How big and how dense is the control network in each group — its devices and how far apart they sit, independent of season? For the per-area coverage % of the protected-area cores, see Modelled coverage →
What the camera markers and shaded surface show.
Relative activity (RAI) — detection rate for the chosen species group, effort-adjusted by camera-hours. Zoom out for a line-level RAI (the DOC protocol's per-line index); zoom in for the per-camera rate. The shaded surface interpolates it between cameras.
Two roles at once — pick both a predator and a protected species and, zoomed in, each camera becomes a predator / protected pie (sectors ∝ detections). Pooling the two into one rate would be meaningless, so the shaded surface switches off.
Where to act — the predator-vs-protected Predator pressure surface now lives on its own Insights → Predator pressure feature, with a toggleable trapping overlay.
For every
protected
species detection on camera, this measures how long
until the nearest
predator
detection — at the
same camera
, or, if you
set a
Within
radius, anywhere in the camera's
neighbourhood
(the camera plus that radius around it) — the
time gap
between the two species sharing the same ground.
That neighbourhood reuses the Coverage map's proximity engine: cameras are sparse, so “same camera” is strict — a wider radius counts a predator seen a few hundred metres away as nearby.
Three views: a map (each camera coloured by how soon a predator turned up near a protected detection), a distribution of those gaps (how often predator and protected are close in time vs far apart), and a seasonal trend of the median gap. Each view has its own “how to read it” (?), and clicking into any of them drills to the detections behind it, down to the record.
The default predator is your project's highest-concern one — switch to the predator that matters to you.
Camera timestamps must be true clock times for this (trapping has no clock time, so it's camera-only). Treat proximity in time as co-occurrence risk, not a confirmed interaction.
How close in time do predators and protected species share the same ground?
Each camera is coloured by the soonest a predator turned up near a protected detection ( red = soonest ) — the sharpest risk signal for that spot; the dot's size is the number of pairs there.
The Surface layer interpolates between cameras for a smooth gradient; toggle Surface / Pairs / Cameras / Boundary top-right.
Click a camera (or a table row) for the detections behind it. Hover a row to preview that camera on the map.
It's exploratory : proximity in time is co-occurrence risk, not a confirmed interaction.
Short gaps (bars on the left) mean the two species share the same ground close in time; long gaps (to the right) mean they mostly don't overlap. Each bar's height is the number of protected detections whose nearest predator fell in that gap.
Click a bar for the detections behind it — each protected detection paired with its nearest predator, both linked to their own record.
It's exploratory : sharing a camera close in time is suggestive of risk, not proof of a kill or even a meeting.
Click a bar for the detections behind it.
The line is the median gap by season — robust to a few extreme values.
Once the median gap reaches weeks or more , the two barely share ground at all — treat the trend as a weak signal there. It's exploratory: proximity in time is co-occurrence risk, not a confirmed interaction.
Click a point for that season's co-detections.
A grid with one cell
per camera location × season
(most recent first).
The number in a cell is its
animal detections
— the same data that feeds RAI — and the
colour
is the deployment's
health
, so you can tell a genuine quiet patch
from a camera that wasn't really working.
It's a data-quality check: a dead or misfiring camera makes RAI lie (a real hotspot reads as empty, or blanks swamp the rate). Spot those here before you trust the activity numbers. Click a cell for that deployment's summary — its effort, detections, health and annotation state (how much of what came back has been classified) — with a button through to the camera's full home (records, species, map). Click the camera name for its all-time home.
The bottom Season status row is the system's verdict per season: ✓ Complete when it's a full monitoring season (near all its cameras are in and classified), or amber ⋯ Filling when it's still coming in from the field or being annotated — each cell shows the cameras-in / annotated tally, and hovering gives the detail. That's the flag behind “latest full season” for monitoring.
Use Seasons shown to set the window (1–2 years or all) and Highlight to focus on just the grades you're chasing.
Each deployment gets
two independent checks
for the effort it ran, and the
cell takes the
worse
of the two:
A grade is about the camera working, not about the place being empty — a true quiet site reads OK with a low count, a dead camera reads Problem.
One animal that lingers in front of a camera trips it again and again — counted naively,
that one visitor looks like
many
and inflates RAI. The
duplicate window
collapses a quick run of the
same species at the same camera
into one visit.
The window is set per species — a lingerer needs a longer one than a species that passes through — so this screen tunes one species at a time : see what it is set to now, propose a new value, and confirm against the actual images before you change the config.
RAI and the Overview detection counts are reported net of duplicates, so this setting moves published numbers — that is why it is worth the look.
It's a heuristic on time gaps, not animal recognition — the images are how you sanity-check that the window matches what actually walked past.
A run is what the window collapses to one visit: same species, same camera, every internal gap inside the window. Scan the photos along a run — same individual, or a new arrival?
Two ways to view them. A run's heading opens that run on its own — one camera, one occasion. Step through all N photos (or clicking any photo) opens every run as one sequence, run by run. Full screen either way: arrow keys or the on-screen arrows, Esc to close.
Every frame of a trigger is in the viewer. One trigger writes a burst — three frames here — and the record says only that the species was in the trigger , never which frame it walked into. So the strip's thumbnail can be an empty scene for a perfectly real detection; the badge on it says 3 frames , and the viewer steps them all. Judge a detection on the burst, not on its first frame.
The caption is two lines: where and when on top, then where you are. Three words, each shown with its size so it explains itself —
(Click the timestamp under a photo for the full record instead.)
Fetch every detection in a run governs the long ones. Off, only the detections in the strip are fetched and viewable — at most eight a run, so the page stays cheap. On, the viewer steps every detection in the run, including the ones the strip elided ( +16 more ) — worth it to judge a long run, but a run of fifty means pulling fifty triggers off the source the first time you look.
Show what fell either side adds the detection immediately before and after each run — the ones the window did not collapse, so you can see whether a run was cut short. They are dashed and dimmed in the strip, tagged outside , and the viewer says so in words. Their gap is by definition larger than the window: that gap is why the run ended.
Images load once. Photos not yet held locally are fetched from the source in the background and cached, so they appear a few seconds after you ask for them and are instant thereafter. That is why this shows a handful of runs and not a hundred — the fetch is the expensive part, so it is spent only on the runs that decide something.
Rodent
tracking tunnels
— inked cards in tunnels along a line, read for the
footprints left overnight. The
Rodent Tracking Index (RTI)
is the share of a line's
tunnels that recorded a species' prints. Lower is better for pests; watch it fall around toxin work.
Each line on the chart is a species; one point per survey season, in calendar order — the same picture as your recording spreadsheet, kept live.
The value is the
mean across the lines
in the selected group(s), and the
band is ± 1 standard error — the
line is the unit of replication
(DOC method). Wide band
= lines disagree; tight = they move together.
High-use only restricts the count to tunnels marked heavy/high-use (your high rat / mouse / insect use rates). Species and Group are in the plot's options panel (the sliders button, top-right).
Presence per tunnel, not a count — you can't tell how many rats crossed a card, so the index is the proportion of tunnels tracked, exactly as the field sheets record it.
A tracking-tunnel index of relative abundance — the percentage of a line's valid tunnels that recorded each species' prints, meaned across lines (the DOC protocol), from a short survey each season.
Click a point (or a table row) for that season's tracking-index card.
The raw records behind the maps and summaries — one row per detection (camera) or capture (trap) in your current selection. Use it to inspect individual records, find something specific, or export the underlying data.
Scope it with the chips on the title line — Data period, Group and Species — plus the Filters gear (Line, Device, exclude possible duplicates, and which Columns to show). Those decide which records fill the table.
The raw records behind the maps and charts — filter any field, search, and export the current selection.
A line is a run of devices set out and worked together — a trapline, a camera transect, a row of tracking tunnels. It is the unit this programme is actually run and analysed on: camera activity is a per-line index, trap servicing and cadence are judged per line, and a tracking tunnel is only meaningful as part of its line's survey.
Click a line for its home — the same view a map's line marker or a device's “step up to the line” link opens: what it is, its devices, what it has caught or recorded, its trend against its neighbours, and a map.
Every line you run, with its lifetime work and where it stands — click one for its home.
How each trap is coloured and sized on the map — pick the quantity in the chip bar's Map options ⚙ (the Map shows toggle).
Every trap that caught something is a purple square — the app's catch colour — and its size is the number caught , in both views. A trap that caught nothing stays a small grey square.
Catches (the default) — colour and size both climb with the tally: purple for one catch through to magenta for the busiest traps, so the map answers where have we caught the most at a glance. Robust to effort: a trap checked only once or twice can't dominate.
Catch rate — predators of the chosen group caught per 100 trap-nights (effort-adjusted), so heavily- and lightly-checked traps compare fairly — but noisy at low effort, so a trap with a catch or two over a few trap-nights can read a disproportionately high rate. Here the colour carries the rate while size still shows the catch count, so a low-effort, high-rate trap reads as a small magenta square — flagged, not dominant.
Ignore traps under N catches — drops small-count traps from the map to declutter it (and, in Catch-rate mode, to hide the low-effort spikes). The records table below always keeps every catch.
Servicing health — how recently each trap's been checked, against the check interval declared for its ground — has its own map on the Trap review page (the Map tab).
The trap network
at a glance
— servicing beside catch, across its tabs (
Summary
,
Priorities
,
Servicing map
,
Modelled coverage
,
Trend
,
What we caught
). They sit side by side because a
well-checked network that isn't catching, and a catching one going stale, are different stories.
The colour everywhere is a trap's servicing status (key above every tab); the grades are next door.
Status is how recently a trap's been checked , judged against the interval this project declares for that ground — set per locality or line in the settings, and the same number in every season . The season moves the standard instead — what a project intends to do doesn't change with the weather, and this network's own checking rhythm shifts barely a day between summer and winter, so grading against a doubled winter interval only excused the winter it was there to judge. Nor is it the trap's own measured habit: a yardstick taken off the behaviour it judges cannot fail, and a trap checked every 120 days used to read On schedule for keeping to a rhythm no season of the standard would accept.
There is no band between the first two , and that is deliberate. A middle grade — past the interval but inside a short grace window — put a third name on one number and was the least defensible of the three: it ran 1 to 6 days wide depending on the ground, held 106 of this network's 1,160 traps, and swung from 2 traps one season to 105 in another while the checking rhythm it was grading moved barely a day across the year. That is snapshot timing, not servicing. It also collided with the set-aside below, which retires a trap at ~ 3 mo whatever its rhythm: on ground with a declared 84-day interval a trap ran out of calendar before it could be graded late at all, and went straight from On schedule to Dormant — 62 and 15 of them across 18 seasons, with not one graded late in between.
Days past the declared interval replaces it: signed, so −5 d is time in hand, 0 is the day itself and +12 d is the size of the miss, and never cut off — a trap set aside a year ago still reports how far past it is. The word is a reading of that number; the number is what a threshold argued in days can be set against.
Doing what we said ( On cadence ) is a separate question from whether the rhythm we actually run is fast enough ( Meets standard , vs the seasonal (Summer 10 · Spring 14 · Autumn 14 · Winter 28) d standard ). Different numbers, different yardsticks: On cadence reads how recently a trap was checked against the interval declared for its ground, Meets standard reads the trap's measured whole-history cadence against the standard. So a place can sit faithfully On schedule and still be too slow, and the second is the question the first cannot ask. Neither figure is on the Summary : On cadence is the % line on Trend , Meets standard is the Against standard tab, and both are gauges on the Overview's network health.
Graded on
time since the last check
, read
as of the latest data
— no peek at later checks; an in-progress period is read from its freshest check, so a half-done period
doesn't paint every trap overdue.
The trap network at a glance — servicing, catch effort and data quality.
A worklist for
limited time
: every
trap line
that needs attention, grouped by
why
it needs it, and ranked so you can take the
top of each list
first. It reads the network
right now
(status as of the latest check), and the numbers match the
Summary
tab.
Lines flagged Quick win are small , already past the check interval declared for their ground and either being checked too slowly for what they catch or a strong former catcher — a fast visit for a good return; pick them off within any group first.
A line sits in the one group that matters most; any other reasons it also fits show as tags on the row. The Priority meter blends payoff (catch), urgency (how far past its declared interval, counted in whole intervals — so a week late on a fortnightly line outranks a week late on an 84-day one), slip (how much catch fell) and revival value (former yield × how long idle).
Four panels stacked on one
shared time axis
— the
servicing
story,
across
seasons or years
, read top to bottom as
effort → work → gaps → yield
. Every one of them measures the
whole period
, so a column reads straight down:
Hover a season for a tooltip of every value at that point; the season highlights across all four panels. Each panel has its own y-axis — read each for shape and direction over time, not absolute height against the others.
This trend deliberately ignores the selected Period — it spans all of time, including the current period , which is marked (to date) and sits to the right of a dashed rule. That period is still accumulating : its effort and Checks bars count a part period and are drawn faded for that reason — read them against the periods beside them only once the period closes. For a right now reading of which traps are Overdue and by how many days, the Summary cockpit is the surface that answers it: the servicing states are a snapshot, and this figure deliberately no longer draws one.
The per-night catch rate (the trapping outcome/yield) lives on Catch results → Trend , not here — this panel is the servicing signal, catch % .
A
trap-night
is
one trap hunting for one night
. The bar counts
every night the traps on the register could have offered over the period — roughly
traps × the length of the period
— and colours each night by
how long it had
been since that trap was checked
.
A set trap stops pulling animals in some weeks after its last visit — the bait is gone or spoiled, so the ground around it is no longer being held. So the bar splits each night by how long it had been since a check — within 42 days , then older, then older than a year . Because the rule is applied to the night rather than to the trap, the bands can simply be added up over a season.
The catch rate divides by every night under a year old — everything but the oldest.
Both cutoffs are project settings, and they answer to different questions: the first is where a check stops holding ground, the second (a year) is where a night leaves the catch rate.
It is a claim about recency — a check was near that night — and not about the gear. A single-set trap that fired on day 2 sat dead for the rest of its interval, and those nights still sit in the freshest band here. That loss is real and it is estimated separately, on the Saturation view. The cut at a year is kept far away from the one before it because the two answer to different readers: the first has to be strict , or the app claims ground the traps were not holding; the second has to be generous , because a strict one would let a lapsing network divide by fewer nights and report a higher catch rate for being neglected.
Every night that every trap stood during the period is counted once — that is the
denominator
, and it is the same total the
Effort
panel's bar sums to.
Now stamp each of those nights with its
age
: how many days had passed since that
trap was last checked. A cell is the share of the period's nights whose stamp is
N days or less
.
So a cell reading 41% within 7d says: 41% of the effort this network put into the field that season was put in by a trap that had been visited within the previous week. The other 59% of the nights were worked by traps that had been standing longer than that. Both sentences are about nights. Neither is about how many traps got checked.
A trap checked once and then left contributes one check and a great many nights — and every one of those nights lands in the bottom rows, which is where they belong. That is the whole reason the grid is built on nights: anything counted per CHECK gives that trap one vote and a well-walked trap thirty, which is exactly backwards.
A night 10 days after a check and 2 days before the next one is not “within 7d”. The question is always time since the trap was last attended, never time until it next will be — because the first is what the trap was actually experiencing that night.
The thresholds are absolute days, so a three-month season and a twelve-month year ask the same question of different nights — nothing in the numerator or in the ladder stretches with the width of the column. That is why columns compare directly, and why the part period is kept rather than dropped: a share survives a season being cut short, where a count does not.
A share of effort is an unfamiliar quantity. Days between visits is the one everyone
here already thinks in — and they are the
same fact
, so the tooltip gives both.
A round walked at an even interval of T days spreads its nights evenly across that interval, so exactly N ÷ T of them fall within N days of a check. A fortnightly round is 50% on the 7d row. A monthly round is 25% . Inverted, a cell hands the cadence back: T ≈ N ÷ the share — which is the “same as an even round every ~17 days” line on the tooltip.
It says even round and ~ because that assumption is doing the work. A ragged round — and every real one is ragged — comes out longer than its own median check interval, and the gap between the two is itself a reading: it is the neglected tail, the nights after a trap's last visit, which no interval statistic can see at all because there is no closing check to measure to. They are 4.4% of this network's nights.
Hover two rows of the same column and the implied round disagrees with itself — Autumn 2026 gives 30 days on the 7d row and 46 on the 30d row, ending at 395 on the last. Nothing is wrong: N ÷ share is the same number on every rung only for a round that is genuinely even, so every day it gains per rung is tail .
That is why the sentence on the tooltip names its own row. A network with a flat ladder is walked to a rhythm; one that looks monthly from the fresh end and annual from the stale end has a core being kept up and an outer edge that has been let go — and it is the same network, reporting one median interval, either way.
Each row is looser than the one above it, so the shares rise as you go down — every night within a week is also within a fortnight. The row where a column first reaches ~100% is roughly that period's worst gap : past it there is no effort left that was any staler. A column that only saturates on the 180d row had traps that went half a year.
Cumulative rows answer “how much is at least this fresh”, which is the right question but hides the concentration — so each cell's tooltip also gives the share it ADDS over the row above it (“22 points of that fell between 14 and 28 days”). That is the reading a bucketed grid gave, in the one place a bucket's unequal width cannot follow it.
within 42d
— the row labelled “= Effort panel”
— is exactly the share the Effort panel's
first band
draws: the nights on which
the trap was still covering the ground around it. One row of this grid equals the number on the
figure above it. Change the cutoff in the project settings and the ladder grows a rung for it.
It is not the catch rate's cut — that one sits far higher up the ladder, and nothing on
this grid moves when it changes.
The cadence standard is the other reference, and it cannot be a row: it is seasonal ( seasonal — Summer 10 · Spring 14 · Autumn 14 · Winter 28 days ), so it lands on a different row in each column and is drawn as an outlined cell tracing a stepped path across the grid. Read along it for how much of each season's effort met the standard that season actually had. A by year column carries no mark — a seasonal standard has no meaningful average, and a line drawn against a bar nobody had to meet is worse than none.
The gear above the plot chooses which to mark: the cutoff, the standard, both, or neither.
A bucketed version (0–7d, 7–14d, … 1yr+) drew every bucket as a row of the
same height while the buckets themselves ran from 7 days wide to 185 — the widest was
26 times
the narrowest. Two rows' colours were then not comparable quantities. A
threshold has no width, so every cell here answers a question that stands on its own:
76% of this season's trap-nights were within 60 days of a check
.
One ramp, and deeper is better in every row — a bigger share of nights close to a check. That is the same direction it means on Checks per trap beside it, so the two grids can be read without switching conventions. (A bucketed grid could not do this: “more” would be good at the top and bad at the bottom.)
Every night in a cell traces to a dated check on a named trap — a night is “within 30 days” because the check that set it was. So clicking a cell opens that cell's own traps : not the period's traps, but the ones with nights inside that threshold, and from there each trap's checks. A servicing status cannot be audited that way: “this trap was Overdue on the 31st” is a statement about a moment, computed from the absence of a visit, with no record underneath it to show.
Rows fading as you read left to right is a network being checked less often. The current period is marked (to date) — shares are still meaningful in a part period, unlike counts, so it is kept rather than dropped.
The panel beside this one says the network's rhythm moved. This says
on which lines
— one row per trap line, one column per period, grouped under the
place the chip bar is filtering on.
Checks per trap — that line's visits in the period ÷ the traps it had on the register then. Never the raw count: a 32-trap line would outrank a well-walked 6-trap one on effort alone. Retired traps are out of the denominator, so a line is not marked down for retiring traps tidily.
Its traps on the register — the denominator every cell in that row is divided by. It is on the axis because the rest of this panel cannot be read without it: the lines are nothing like the same size, and a row of cells gives no hint of that.
The count is the line's size in the last period it held traps — so a line still being run reports what it is now, and a retired one reports what it was when it went. Which means the line counts in a block will not always add to the block's own: Mokoroa's read 84, 18, 12, 11, 6 and 5 against a block of 107, because three of those lines are gone. That is why the All lines row carries no count of its own — every number on the axis is true, and only adding them up is wrong. Exact traps for any one period are on that cell's hover.
Each block closes with a row under a rule: that place's total checks ÷ its total traps for the period. It is pooled , not the average of the cells above it — averaging the cells would give a 6-trap line the same vote as a 32-trap one, which is the very thing checks-per-trap was chosen over a raw count to prevent.
Which is why it will not look like an average, and should not. Mokoroa's Summer 2025/26 reads 1.9, 1.0, 0.0 and 0.0 across its four lines and pools to 1.5 — because 84 of the block's 117 traps are on the line reading 1.9 , and the two zeros are 9 traps and 13 between them. The eye averages the four numbers it can see and gets 0.7; the traps are not distributed that way. The row follows the traps, and the counts on the axis are what let you see it coming. Each pooled cell's hover also names the biggest line in the block.
It is here because nothing else on this page carries the number: the Checks panel on the figure above is a count of visits, which ranks places by how many traps they have. Clicking it opens the whole block for that period.
The current period is drawn and marked (to date) . It is short by construction — a count over two months of a three-month season is a paler cell for a reason that is not the field — so the cell says so on hover. It is drawn rather than dropped because the figure above and the grid beside this one both draw theirs, and a view whose last column is a different period from its neighbours' is its own kind of misleading.
Rows run best-worked at the top, so faltering lines gather at the foot as a block. Click a cell for that line's traps in that period.
Where the traps are, coloured by servicing status — the in-service traffic light : On schedule green, Overdue red (the loud, go-check-it state). Two colours, because in service there are two answers: inside the interval declared for that ground, or past it. How far past is on each trap's hover and in the table beside the map, in days — a map can only carry a handful of colours, and the amber middle it used to carry stood for a band 1 to 6 days wide that a reader had no way to see the width of. Traps that are set aside drop OUT of the colours into grey: Dormant a solid grey square, then Retired a hollow grey outline (long out of use). Each status is a selectable layer (top-right) — turn off the ones you don't need. The colour is each trap's status right now (as of the latest data), so the map always shows what needs checking today — even when the period picker is on an earlier season (the catch and cadence figures in the table stay scoped to that period).
Two dashed footprint hulls sit under the markers, each its own layer: Footprint (locality) (on by default) traces the ground the traps of each locality stand on — the axis this page scopes by — and Footprint (group) (off by default) the same envelope per declared reporting group. Where the two disagree, a trap's label and its ground part company.
Under-cadenced — a bold black ring marks a trap whose own catch history implies a quicker check interval than it actually gets , and which is overdue right now (past its declared interval, and still in the rotation — a set-aside trap is a different problem). The requirement is measured from that trap's catches, not a fixed catch-% bar, and the ring appears only when the entire plausible range of required intervals is quicker than the achieved cadence — so it under-reports rather than cries wolf. The table below lists them worst-first. But missing blank entries fake this same signature, so a ringed trap on a data-flagged line (the ⚠ flag on the Summary) may be a data-entry issue, not real pressure — check the entries first.
Click any trap for its full check history. Grade meanings are in About the Field operations .
Every trap is kept to
some
rhythm — but is that rhythm
fast enough
? This tab benchmarks each trap's
normal cadence
(its established whole-history check interval)
against the project's
cadence standard
(
seasonal — Summer 10 · Spring 14 · Autumn 14 · Winter 28 days
)
.
It's the Meets standard axis of Network health, and a different question from On cadence : a trap can be faithfully on cadence (checked as often as the interval declared for its ground asks) yet still be too slow — what this project intends for that ground, and the rhythm it actually runs, can both be slower than the standard.
Where the bar can be tighter than the standard. Two things are measured here and they stay independent. The standard is a declared policy — what good practice asks of this ground. Saturation is a measurement: at the rate a line actually catches, how often must a trap be emptied for it still to be catching rather than sitting full or sprung? Where a line's own catches say it needs checking faster than the standard, this tab holds it to the faster number — the bar is whichever of the two is tighter , never an average and never a swap.
It only ever moves one way . A line whose catch rate implies a slower interval than the standard keeps the standard — measured across this network, saturation is looser than the base standard on most lines, so letting it loosen a cadence would change ‘meets standard’ from a minority of traps to a comfortable majority without one extra check being done. The two readings are still shown separately in the shortlist's last column, because they disagree for good reasons and collapsing them into one number would be checking the data against itself. A tightening is that disagreement acted on in the only direction the evidence supports, not a merge.
The
histogram
is the spread of trap rhythms; the
dashed line
is the declared standard. Bars
green
meet the bar that applies to them;
red
run too slow.
The shaded band, and why a red bar can sit left of the dashed line. Where saturation tightened some lines, the band spans from the tightest bar in view up to the declared standard. A trap inside that band is judged against its own line's quicker number, so it can be red while standing to the left of the rule — the rule is the standard, not every trap's bar. The banner names how many lines moved and how far; the accordion marks each one on its row.
The shortlist below lists the below-standard traps, slowest first — the ‘speed these up’ list. Its last column is the interval that trap's own catches imply, with the range around it, marked trap or line for the grain it was fitted at: a per-trap range spans roughly 5.6× here against 1.7× per line, so a trap-grain figure is evidence to weigh and a line-grain one is a schedule to change. Click any row for that trap's home & full check history.
The headline % and counts are the same figures the Meets standard gauge shows on the Overview.
On all data the headline changes question. With a season selected, everything here — headline, histogram and shortlist — is that season's bar. Select all data and there is no one season to judge by, so the headline pools trap-seasons : of every season each trap has worked, the share in which it ran the rhythm that season's standard asked for. The histogram and shortlist below it stay on today's bar, because a shortlist is something you act on now — the line under the headline says which is which.
What-if the standard were different? The plot's options gear has a Test a different standard tuner — turn it on and drag the target to watch the verdict, histogram and shortlist re-assess against a hypothetical cadence. A planning lever only: it never changes the configured standard or the Overview gauge, and the banner flags the figure as What-if .
Every trap check records two independent things: what the
gear
was doing (it held a catch, it had fired with nothing in it, or it was still set) and what
the
bait
was doing (fine, degraded, or gone). Each band's width is its share
of that place's checks, and each ribbon is one combination of the two.
Every place is drawn to the same height whatever its size, so the shapes compare directly; the place's actual number of checks is in the heading above it.
Both axes are already reported on their own — the fired-but-empty arm drives
Calibration
, the bait axis drives
Bait effectiveness
. What
nothing else shows is how they
combine
.
The combination worth hunting for is still set with the bait gone : the trap never fired and its lure has been taken, so it has been sitting there working with nothing on the hook, and neither axis on its own says so.
The
caught
band is safe to compare between places — a catch is
recorded the same way by every source.
The fired but empty band is not. Whether a check gets written down as "sprung, empty" rather than just "empty" depends heavily on the person doing the checking: comparing people who check the same traps, the checker accounts for roughly ten times as much of the variation as the trap does. Read that band within a place over time , and use Calibration — which models the checker away first — for any judgment about an individual trap.
A not recorded band is neither good nor bad news: it is a statement about how completely that place fills the field in, and it is shown rather than dropped so a share is never quietly computed over a smaller denominator than you think.
A check that finds the trap
sprung with nothing in it
is a wasted
capture opportunity — and worse, the trap then sits dead until someone comes back. A trap doing
this repeatedly may need
recalibrating, cleaning or replacing
.
Misfires are rare — a couple of percent of checks — so most traps show none and a handful of events proves nothing. Only traps with enough checks are assessed, and only those clearly above expectation are flagged.
Recording a sprung trap is partly a
habit
. Comparing people who check
the same traps
, most sit within a percentage point of each other — but the extremes
differ by ten, which against a ~2% base rate is a six-fold difference.
Rank traps on raw misfire rate and you rank the people , not the gear: one person's traps would fill the top of the list. So each trap is compared against what the habits of its own checkers predict, and the p-values are corrected for the number of traps tested at once.
Check the
Mostly checked by
column. If nearly every check on a flagged
trap was done by one person, the trap and that person's habit
cannot be told apart
— the excess may belong to either. A flagged trap checked by several people is the stronger case.
Either way this points at which records to read , not at a conclusion. Open the trap and look at the checks themselves before sending anyone up the hill.
Your
best-performing traps
— the ones catching the most per night of
effort, not just the most overall. Ranked by
catch rate
(captures per
100 trap-nights
), so a trap run only a few nights can't beat a steady performer on one lucky catch.
The Hero traps highlights the heroes — the traps whose catch rate is a statistical outlier above the norm (not an arbitrary top-N). They're numbered and sized by rate over a field of every other trap — purple if it caught anything (no number), faint grey if it caught nothing — so the standouts pop while the field still shows where removal happened. A removal-intensity surface is available as an opt-in map layer. Pick a species (top) and the period / group (chips on the title line).
Rate rewards efficiency, not just busy lines — a quietly effective trap on a low-traffic line can out-rank a much-checked one on a hot line.
What the network is removing, and how it's performing — the Output cockpit ranks caught · rate · pressure by locality → line → trap. A high-pressure LINE runs on a check rhythm slower than its own catch rate implies it needs — punctual or not, the schedule is too slow. Saturation estimates what the network is missing because traps sit full between visits. The Hero traps highlights those whose catch rate is a statistical outlier above the norm.
The network's
catch rate
(captures per
100 trap-nights
) over
seasons or years
, across all of time. The
outcome
twin of the servicing trend: are we removing
more or fewer
predators over time? Starts on the
overall total
and breaks down by species on demand.
Pooled Σcaptures ÷ Σtrap-nights over the traps in view (the Group chip narrows the scope). This trend ignores the selected Period — it spans every season. The per-visit catch % (a servicing signal) lives on Trapping → Servicing over time, not here.
The species mix your traps caught — what you're actually removing, not just how much.
Catch
composition
— not just how much you caught, but
what
.
The headline strip totals it up (removed · species groups · the leader · the highest-
concern
predator caught); each cell drills to those captures.
The ranked board lists every species group biggest-first, coloured and badged by its silhouette. Switch Display to the pictograph for an at-a-glance icon read (one icon ≈ a fixed number of animals). Split the board by locality to compare places or by season to watch the mix shift (a rat irruption, a mustelid pulse); Show as share switches those to percentages. Click any bar, icon row or cell for the captures.
Scoped to the selected
Period
and
Group
— so 'by season' shows
the seasons in the selected window (pick
All data
for the full history).
Other
gathers anything caught that isn't in a configured species group. The
pictograph
is a quick visual — read exact numbers off the ranked board (which also downloads).
Counts are individuals recorded at trap checks. A self-resetting trap can log several at one check; a blank check adds none — this is the catch, not the effort (for the effort-normalised catch rate, see Field operations).
A single-set trap catches
at most one animal per visit
. If it fires on
day 2 of a 30-day interval it sits dead for the other 28, and every predator that passes in that
time leaves no trace. Catch % cannot see this — it counts that interval as one successful check.
So two lines with very different catch % can be under identical pressure : a line at 70% caught on 57-day checks and one at 24% on 14-day checks work out the same. The corrected rate puts every line on one scale regardless of how often it is walked.
Missed captures is what the correction implies you did not get, for the whole line over the selected period , counted in animals — not a rate. It is all species together , because a trap fires once whatever walks into it; the app does not try to guess which species you missed. The species column beside it is an observed catch count, there so an under-checked line that takes the animals you care most about can be picked out first.
Recovers is the headline: the share of everything that walked into this line's traps that still got recorded, on the cadence the line actually ran. It is the answer to “what am I missing?”, and unlike the column beside it, it means something on every line.
Should check every is deliberately blank on most lines , and a dash is a finding rather than a gap. A cadence appears only where the check interval is genuinely what limits the line — quicker than the interval it already gets and quicker than the 28-day standard . Everywhere else the honest reading is the Recovers figure.
A dash never means you could check the line less often. Saturation can only ever argue for checking more often. Every other reason to visit a trap — bait, gear condition, misfires, welfare obligations, keeping a volunteer in the habit — is invisible to this model, which is why the cadence standard is the floor and this only ever tightens past it.
The
20%
is a
choice, not a research finding
— it is the amount of saturation loss this project is willing to live with, and it is set in
configuration. Nothing in the trapping literature names it. Change it and every
Should check every
on this page moves.
What the literature does inform is the check interval itself — the seasonal cadence standard on Field operations → Against standard comes from DOC and Predator Free NZ guidance. The two are reconciled here, not averaged : that standard covers every reason to visit a trap this model cannot see, so it is the floor, and saturation is allowed to tighten a line's cadence past it but never to loosen one. That is exactly why a required interval slower than the standard is not shown — it is not the constraint.
The trade-off the number encodes — how much of what walks past a trap is lost, at each rate of finding the gear sprung:
| Checks finding it sprung | Share of opportunities lost | Uplift on what you recorded |
|---|---|---|
| 5% | 3% | +3% |
| 10% | 5% | +5% |
| 20% | 10% | +12% |
| 30% | 16% | +19% |
| 40% | 22% | +28% |
| 50% | 28% | +39% |
Loss climbs slowly to about a fifth, then accelerates — which is the only real argument for putting the target where it is rather than higher.
This measure reads a high catch % as high pressure, and the commonest data fault does
exactly that for a false reason: if blank checks go
unrecorded
, catch % climbs
toward a spurious 100%. Nobody invents a capture, so the error only ever runs
one way
— it inflates saturation, never deflates it.
Each line is therefore checked against its visits — one person, one line, one day. If blanks go unentered, a visit records only the traps that caught, so the share of the line's traps appearing that day collapses. That share is unrelated to catch %, so it can tell the two apart:
The number is always shown. Investigate a flagged line in Records before acting on it.
Every trap carries three measurements of the ground it stands on —
how high
it is, how far it sits from the
nearest mapped water
, and how far from the
nearest formed track or road
. This cuts the catch record by those bands and asks whether the chosen species turns up in one
kind of ground more than the rest of the network would predict.
Departure from expected is the headline: 1.0 means this band caught exactly its share, 2.0 means twice it, 0.5 half. The bar beneath shows how often the traps in each band were actually checked — because a band that is walked half as often is not comparable ground, it is comparable ground under different servicing.
This is not habitat preference. Traps are placed where people can walk and where someone judged an animal would pass, so the ground the network samples is not the ground that exists. Read every line here as a statement about this network , not about the animal.
This ranks baits by how often they catch — captures per 100 trap-nights (a rate that accounts for how long each bait was out), or total captures if you switch Measure to “Total caught”.
A catch is found at a check, but it was caught by whatever was in the trap before that check — so each catch is credited to the prior check's bait, not the fresh bait set on the day it was found.
Full recipe treats the whole bait set as one thing — “Ping Pong + Salted Rabbit” is its own line, separate from either alone. Cleanest to read, but it splits the data into many small recipes.
Ingredient credits every ingredient for the catches in intervals where it appeared, so one catch on a two-ingredient bait counts for both. Good for spotting ingredients that show up in baits that work — but it's a correlation, not proof : ingredients that always travel together blur, so you can't tell which did the work. Read it as “appears in successful baits”, not “causes catches”.
Baits with under 100 trap-nights of use, or fewer catches than the in-plot Min captures floor, are dropped — too little to compare (one lucky catch on a rarely-used bait would otherwise top the chart). The chart shows the top 15.
Click a bar for the individual captures, click a catch to open its full record, or click a trap name to see that trap's whole check history and watch the bait change around each catch.
Captures by bait — attributed to the bait that was in the trap during the catching interval (the prior check's bait). Click a bar to see the captures behind it, then a trap to see its bait history.
A friendly
leaderboard
of the volunteers checking the traps, for one
completed season
at a time. Each trap check is credited to whoever did it; a check
that found a kill is a
catch
.
Badges are the backbone — most active trappers earn at least one.
Score = the sum of catch
weights
— predators that matter most to
kiwi count for more, so quality beats quantity (a stoat outweighs several rats):
Weights are project settings (
trapping$target_weights
in project.R). Bycatch
(birds, unidentified) scores zero.
A catch is credited to whoever checked the trap and found it — so this rewards active checkers, which is rather the point, but it isn't the same as who set the trap well.
Who's checking the traps and catching the predators that matter — ranked for one completed season. Scored by weighted catches (a stoat counts for more than a rat); badges reward effort and variety, not just the top of the table.
The raw records behind the maps and summaries — one row per detection (camera) or capture (trap) in your current selection. Use it to inspect individual records, find something specific, or export the underlying data.
Scope it with the chips on the title line — Data period, Group and Species — plus the Filters gear (Line, Device, exclude possible duplicates, and which Columns to show). Those decide which records fill the table.
The raw records behind the maps and charts — filter any field, search, and export the current selection.
A line is a run of devices set out and worked together — a trapline, a camera transect, a row of tracking tunnels. It is the unit this programme is actually run and analysed on: camera activity is a per-line index, trap servicing and cadence are judged per line, and a tracking tunnel is only meaningful as part of its line's survey.
Click a line for its home — the same view a map's line marker or a device's “step up to the line” link opens: what it is, its devices, what it has caught or recorded, its trend against its neighbours, and a map.
Every line you run, with its lifetime work and where it stands — click one for its home.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Apteryx mantelli
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history .
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Time on camera between this protected and the nearest predator — at the same camera, or within a chosen radius. Shorter gaps mean they share ground close in time.
Click a bar for the co-detections.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Gallirallus australis
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history .
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Where Weka is being caught — group → line → trap, ranked by catches and rate. The Output cockpit from Catch results, scoped to this species.
Which baits catch this species best — by capture rate or total caught. Group by the full recipe (the whole bait set) or by individual ingredient.
Click a bar for the captures behind it.
Time on camera between this protected and the nearest predator — at the same camera, or within a chosen radius. Shorter gaps mean they share ground close in time.
Click a bar for the co-detections.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history .
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Where Mustelids is being caught — group → line → trap, ranked by catches and rate. The Output cockpit from Catch results, scoped to this species.
Which baits catch this species best — by capture rate or total caught. Group by the full recipe (the whole bait set) or by individual ingredient.
Click a bar for the captures behind it.
Time on camera between this predator and the nearest protected — at the same camera, or within a chosen radius. Shorter gaps mean they share ground close in time.
Click a bar for the co-detections.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Rattus · Rattus exulans · Rattus norvegicus · Rattus rattus
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history (which reaches back over the tracking record's decades) .
Tracking index (RTI) is the rodent tracking-tunnel rate — the % of valid tunnels that recorded this species' prints (per line, meaned across groups, ±1 SE). Survey rounds within a season are pooled to the season point; for the round-by-round detail (e.g. a pre/post-toxin knockdown) see the Tracking index page.
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Where Rats is being caught — group → line → trap, ranked by catches and rate. The Output cockpit from Catch results, scoped to this species.
Which baits catch this species best — by capture rate or total caught. Group by the full recipe (the whole bait set) or by individual ingredient.
Click a bar for the captures behind it.
Time on camera between this predator and the nearest protected — at the same camera, or within a chosen radius. Shorter gaps mean they share ground close in time.
Click a bar for the co-detections.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Felis catus
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history .
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Where Cats is being caught — group → line → trap, ranked by catches and rate. The Output cockpit from Catch results, scoped to this species.
Which baits catch this species best — by capture rate or total caught. Group by the full recipe (the whole bait set) or by individual ingredient.
Click a bar for the captures behind it.
Time on camera between this predator and the nearest protected — at the same camera, or within a chosen radius. Shorter gaps mean they share ground close in time.
Click a bar for the co-detections.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Erinaceus europaeus
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history (which reaches back over the tracking record's decades) .
Tracking index (RTI) is the rodent tracking-tunnel rate — the % of valid tunnels that recorded this species' prints (per line, meaned across groups, ±1 SE). Survey rounds within a season are pooled to the season point; for the round-by-round detail (e.g. a pre/post-toxin knockdown) see the Tracking index page.
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Where Hedgehogs is being caught — group → line → trap, ranked by catches and rate. The Output cockpit from Catch results, scoped to this species.
Which baits catch this species best — by capture rate or total caught. Group by the full recipe (the whole bait set) or by individual ingredient.
Click a bar for the captures behind it.
Time on camera between this predator and the nearest protected — at the same camera, or within a chosen radius. Shorter gaps mean they share ground close in time.
Click a bar for the co-detections.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Canis lupus familiaris
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history .
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Time on camera between this predator and the nearest protected — at the same camera, or within a chosen radius. Shorter gaps mean they share ground close in time.
Click a bar for the co-detections.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Trichosurus vulpecula
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history (which reaches back over the tracking record's decades) .
Tracking index (RTI) is the rodent tracking-tunnel rate — the % of valid tunnels that recorded this species' prints (per line, meaned across groups, ±1 SE). Survey rounds within a season are pooled to the season point; for the round-by-round detail (e.g. a pre/post-toxin knockdown) see the Tracking index page.
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Where Possums is being caught — group → line → trap, ranked by catches and rate. The Output cockpit from Catch results, scoped to this species.
Which baits catch this species best — by capture rate or total caught. Group by the full recipe (the whole bait set) or by individual ingredient.
Click a bar for the captures behind it.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Mus musculus
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history (which reaches back over the tracking record's decades) .
Tracking index (RTI) is the rodent tracking-tunnel rate — the % of valid tunnels that recorded this species' prints (per line, meaned across groups, ±1 SE). Survey rounds within a season are pooled to the season point; for the round-by-round detail (e.g. a pre/post-toxin knockdown) see the Tracking index page.
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Where Mice is being caught — group → line → trap, ranked by catches and rate. The Output cockpit from Catch results, scoped to this species.
Which baits catch this species best — by capture rate or total caught. Group by the full recipe (the whole bait set) or by individual ingredient.
Click a bar for the captures behind it.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Oryctolagus cuniculus
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history .
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Where Rabbits is being caught — group → line → trap, ranked by catches and rate. The Output cockpit from Catch results, scoped to this species.
Which baits catch this species best — by capture rate or total caught. Group by the full recipe (the whole bait set) or by individual ingredient.
Click a bar for the captures behind it.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Lepus europaeus
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history .
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Where Hares is being caught — group → line → trap, ranked by catches and rate. The Output cockpit from Catch results, scoped to this species.
Which baits catch this species best — by capture rate or total caught. Group by the full recipe (the whole bait set) or by individual ingredient.
Click a bar for the captures behind it.
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Sus scrofa
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history .
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Everything this project knows about one species (or species group), camera and trap
together.
Summary
— its footprint across the network, the settings that govern it,
and where it turns up most;
Trend
— how its activity and catches move over time;
Map
— where it's seen and caught;
Records
— every detection/capture.
Data period & group come from the chips on the title line; Data period defaults to All data . Summary and Trend always span all time (a note sits in the Period's place there); from the Map tab on you can set a Period to focus the rest.
Insecta
Camera activity is RAI (detections per 2,000 camera-hours); Catch rate is captures per 100 trap-nights — each its own panel (different units, don't compare heights between them). Lines are the network mean across groups and the shaded band is ±1 standard error of that mean (how much the groups disagree); toggle By season / By year. Add a compare species to overlay a second line on the matching panel. Show sets the time window — the last 5 years by default so the recent multi-device story is legible; pick All data for the full history (which reaches back over the tracking record's decades) .
Tracking index (RTI) is the rodent tracking-tunnel rate — the % of valid tunnels that recorded this species' prints (per line, meaned across groups, ±1 SE). Survey rounds within a season are pooled to the season point; for the round-by-round detail (e.g. a pre/post-toxin knockdown) see the Tracking index page.
Break lines into
(the plot-options gear) decides what one line on this chart
is
.
Combined
draws one pooled network line.
Group
draws one line per
in-scope group.
Line
draws one line per monitoring line,
within its group
— line 3 in one reserve and line 3 in another are different places and never share a series.
Lines are shaded by their group's colour, so the legend reads in families.
A per-line series carries no ± band, and that is the point:
the line is the unit of
measurement here — RAI and the tracking index are computed per line and then meaned across lines, so
the band on the other two views is the disagreement
between
lines. One line on its own
has nothing to disagree with. Read a per-line view for
which ground is doing what
; read the pooled view for the network's level and its uncertainty. Catch rate is not split by line
(its interval is a bootstrap over lines, which at one line resamples to itself).
Traps with a catch shows the species' trap extent — the share of the trap network that caught it that season (binomial ±1 SE, √(p(1−p)/traps), so it widens when fewer traps ran). A different axis from activity (how widespread vs how active ), and the only extent charted here. Camera extent is effort-sensitive, so it lives in the Occupancy tab as detection-corrected occupancy ψ with a proper trend test.
Click a point for that season's breakdown.
Where this species is detected and caught — an activity surface, per-site points, the boundary, and the records (hover a row to highlight it on the map; toggle layers top-right).
Every detection (camera) and capture (trap) of this species in the selection. Click a row to open the full record.
When this species is active on camera — its overall Diel class (fixed across all groups & all data — a stable rhythm a thin slice shouldn't flip) and the daily Activity Pattern (which tracks the period & group above, and can split by group).
Detections by hour of the local day — the species' daily rhythm on camera. Hours come straight off each detection's time, with no effort adjustment: cameras watch around the clock, so every hour gets the same exposure. (For the dawn/day/dusk/night split, which does correct for unequal period lengths, see the Diel Activity card.)
This chart honours the period and reporting_group above. Turn on Split by group (the plot-options gear) to break it into one panel per in-scope group — each with its own hourly counts and its own night/twilight shading — to compare rhythms between groups. Hover a bar for its exact count.
Night and twilight are shaded behind the bars from the selected period's sun times (mean civil dawn, sunrise, sunset and civil dusk for the chosen group). Because those shift through the year, the shading has two tones:
One headline
diel class
summarising when this species is active
on camera, with the four diel periods' shares beneath it. Computed across
all groups
and
all data
(not the period / group chips), so the class is a stable rhythm
rather than a noisy per-slice read.
Shares come from effort-normalised rates — detections per available hour in each period, not raw counts. This matters most for Matutinal and Vespertine : civil twilight is only a thin sliver of the day (often under an hour), so a species could be intensely active at dawn or dusk yet log few raw detections simply because the window is so short. Dividing by each period's available hours puts all four on an equal footing.
| Matutinal | Civil dawn to sunrise (first light). |
| Diurnal | Sunrise to sunset (daylight). |
| Vespertine | Sunset to civil dusk (last light). |
| Nocturnal | Civil dusk to civil dawn (darkness). |
| Diurnal | Day-active. |
| Nocturnal | Night-active. |
| Crepuscular | Dawn/dusk-active. |
| Cathemeral | Active intermittently across day and night. |
| Arrhythmic | Enough observations, but no clear diel rhythm. |
Click a bar for the detections in that hour.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
Figures only — the map, coverage and the Overview tiles stay whole for the selection.
Is control working here? One place's trapping and monitoring, and which way each is heading.
A
trap-night
is
one trap hunting for one night
. The bar counts
every night the traps on the register could have offered over the period — roughly
traps × the length of the period
— and colours each night by
how long it had
been since that trap was checked
.
A set trap stops pulling animals in some weeks after its last visit — the bait is gone or spoiled, so the ground around it is no longer being held. So the bar splits each night by how long it had been since a check — within 42 days , then older, then older than a year . Because the rule is applied to the night rather than to the trap, the bands can simply be added up over a season.
The catch rate divides by every night under a year old — everything but the oldest.
Both cutoffs are project settings, and they answer to different questions: the first is where a check stops holding ground, the second (a year) is where a night leaves the catch rate.
It is a claim about recency — a check was near that night — and not about the gear. A single-set trap that fired on day 2 sat dead for the rest of its interval, and those nights still sit in the freshest band here. That loss is real and it is estimated separately, on the Saturation view. The cut at a year is kept far away from the one before it because the two answer to different readers: the first has to be strict , or the app claims ground the traps were not holding; the second has to be generous , because a strict one would let a lapsing network divide by fewer nights and report a higher catch rate for being neglected.
A
protected area
is ground this programme defends, with a surveyed boundary.
This report compares what happens
inside
that boundary with the ground
just outside
it.
The outside can be trapped too — it may be the same lines carrying on outward. Where it is, this is a gradient across a boundary rather than a protected-versus-untreated trial. A difference is worth understanding; on its own it is not proof the area is working.
The tabs. Overview is what this area is and where its devices stand. Trap network asks whether both sides were actually checked alike — a difference there affects every rate, so it is worth reading before the contrast. Inside vs outside is the comparison itself, with everything that qualifies it. Species trends is each animal over time, and Approaches the trapped ground this area does not claim. Devices lists every device and its distance to the boundary; Method defines the terms and states the limits.
Every device is sorted by where it stands:
"On boundary" is about measurement, not ecology. It is a few metres either side of the surveyed line, and it means we cannot say which side the device is on . It is not an edge-effect zone — a true ecological edge is a hundred metres or more, and nothing here measures one.
A trap standing near the line takes whichever class the band puts it in, so the band's width decides which side of the comparison it counts on. It is a project setting, chosen per area against how well that boundary is mapped, and the Overview prints the metres this area uses.
Is the protected area actually protected? What we catch and see inside the boundary, against the ground around it.
A build-your-own trend chart. Configure a
plot
— its
species
,
metric(s)
and
places
— then
Save
it;
Add plot
to stack another, or click a plot to
Edit
it. Pick
any species
, not just predators and protected natives (mice, possums…).
All plots share ONE Data period and time axis (Group by), so they read together top to bottom. Use it to look for relationships over time — it's exploratory : lines moving together is suggestive, not proof.
Pick
more than one metric
to compare a species' signals: same-scale metrics
(RTI, occupancy — all %) share one axis, and up to two scales sit on a plot (left + right); more spill
to further plots. Species is otherwise the line
colour
.
A figure has two ways to tell series apart, and the editor names one each. Colour the lines by puts every value on ONE pair of axes, so they lie over each other — the easiest comparison to read. Separate plots by gives each its own plot instead: better once there are too many to overlay. They combine , which is how you get a plot per reserve with a line per boundary band inside it. Naming the same thing in both falls back to a plot per metric — one channel is enough to tell them apart. Metrics can only ever be separate plots (RAI, catch rate and the tracking index share no y-axis), so colouring by anything gives each metric its own.
Max lines per plot (under the plot list) caps how many lines one plot has to carry — a plot with more is dealt onto as many plots as it takes, and those share a y-axis so a line doesn't look to have jumped between them. Lines that belong together are kept together: a species keeps all its metrics, and an effort line rides on every plot it was there to explain.
Places are sets what a place means for that plot, and the rest of the editor follows it. A Reporting group is a declared grouping — what a figure is published under. A Locality is the physical block people work. A Protected Area is ground inside a boundary you can measure against. Neither of the first two is more correct, but the choice is not cosmetic: it is the key the metric groups on, so a device with no value on it is left out of the figure entirely and two axes can put different networks behind the same-looking plot. The note under the plot list says what the live axis omits.
On the Protected Area axis you can narrow further to a boundary band — Interior (well inside the boundary), On boundary (astride the line) or Final approach (the first collars outside it) — and colour by those bands, or give each its own plot, to compare inside against the ground around it . That is the one contrast where both arms are your own devices, on the same ground, under the same effort. It pools the two outer bands into Periphery by default, because that is the arm the protected-area report's headline ratio uses; untick to see all three. Traps carry this contrast — cameras and tunnels here sit almost entirely inside the boundaries, so a band split of RAI or the tracking index is nearly all Interior.
The effort metrics answer a question a catch rate cannot answer about itself. Catch rate already divides by trap-nights, but a trap that stops being checked keeps counting trap-nights it cannot catch in, so the rate falls whether or not anything changed in the field. Cadence says whether the round slowed, Traps overdue what share of the network is past the interval declared for its ground, and Catches per check is the one that settles it: if catch rate fell while catches-per-check held, the traps still catch when you get to them. Click any of their points for the servicing cockpit as at that season. They stop one period short of the other lines on purpose — the season in progress has not been re-checked yet and would read as a servicing collapse.
Traps overdue counts a trap from the first day it is late — no grace period. Until August 2026 the first stretch past the declared interval had a name of its own ( Due ) and sat outside this line, so the series counted only traps more than a week over. That band was between one and six days wide depending on which ground a trap stood on, and on the slowest ground the state beyond it could not be reached at all, so it was measuring the yardstick rather than the fieldwork. Retiring it moved those traps into this line: the whole series steps up at that release , every season at once, with nothing having changed in the field. Earlier seasons are not restated to the old definition — the line shows one definition throughout, and this is the note that says which.
Hover a point for its values; click an RAI/catch-rate point for that species' per-group breakdown in that season, an occupancy point to walk its per-group then per-site basis (which deployed sites counted as occupied), or an RTI point for that group's tracking-index card for the survey. RAI is detections ÷ camera-hours; catch rate is captures ÷ trap-nights; occupancy here is naïve — the raw % of monitored sites with any record (the modelled estimate is on the Occupancy page); RTI is the % of tracking tunnels marked.
Build a plot from any species and metric(s), Save it, and stack more — all on one shared time axis.
One map carrying
both devices
so you can see where the protected
species are and whether predator control is reaching them.
Colour says what a marker
is; size says how much
— one legend, no clashing colour scales.
A green hotspot ringed by purple and grey is
covered
; a green hotspot
with little around it is a
gap
— somewhere a protected species is active but
trapping isn't reaching.
Camera markers are sized by the number of detections — a count of 1 is the same small dot everywhere, and the green and red dots share one scale so predator and protected compare like-for-like (a single hotspot can't shrink the rest — the scale clamps at the busiest 10%). The effort-adjusted rate (per 500 camera-hours) is in each marker's hover, where the comparison across differently-run cameras matters. Catch markers are a count of predators removed, on their own scale. Period and group come from the chip bar; click any marker for its detail.
Whether the network is even dense enough to work — footprint, devices per km², spacing — now lives on the main Overview → Network tab; the Coverage gaps table below ranks the gaps that remain numerically.
All for the selected period & group; everything redraws when you change them.
Camera lines ranked worst-first by how well nearby trapping reaches the protected species they detect.
Per surveyed
area
(each protected-area polygon), how much of it was
within range of a working trap —
averaged over the period you have chosen
, not
measured on its last day. Split into
designed
(nights a trap stood) and
maintained
(nights it was still credited).
Each trap is modelled as controlling a radius you set (starting from the project's configured radius). For every night we take the share of the polygon inside that merged footprint, then average across the nights. It sees gaps that traps/km² can't — in space and in time.
A night is credited when the bait was plausibly still working: within the days a check is taken to keep that trap's ground covered. An interval that caught something keeps all of its nights, however long it ran — a catch is evidence the trap was working, and it is the one signal that survives a check nobody wrote down.
The gap between
designed
and
maintained
(the
Coverage lost
column) is control forfeited to nights that stopped counting. It is
only
that: both columns average over the same period, so an area whose traps went in
halfway through the season shows a lower designed figure rather than a servicing failure.
Each Protected area is its own row — a group that holds several protected areas shows each separately, because they are distinct parcels with distinct boundaries. The dashed outline on the map is the designed extent (all traps).
Credited/night is the average number of traps credited on any one night of the period. It is a mean, not a headcount: a trap credited for half the season contributes a half. That is why it is not a whole number, and why it can be lower than the traps you can see on the map.
An average hides its own shape. 41% of a season is the same number for ground covered evenly all winter and for ground covered fully for five weeks and not at all for eight — which is why the export also carries the worst single night, the number of nights under half cover, and the longest unbroken run of them.
It is a
model
: the absolute % depends on the control radius, but the
designed-vs-maintained gap
and the ranking between areas are robust to it. Coverage is
% of the
surveyed protected-area polygon
, and only areas WITH a polygon appear — a % needs a
real boundary to be a % of. Every trap within range of an area counts toward it,
whichever group it's assigned to
(a neighbour's trap over the line still covers ground
inside). The radius is
one number for the whole project
, not one per species: a stoat
is drawn from further than a rat, and this page cannot say so. It would take more than a per-species
radius to fix — the trap pool is species-blind too, so a narrower disc for mice would still credit
every trap in the network with controlling mice. Read the figures as network geometry, not as cover
against a particular predator.
Coverage is not a cadence judgement. A night counts when the trap was inside its coverage window — a fact about the gear and the calendar. It does not ask whether the trapper is behind schedule: a line declared at 10 days but walked just inside that window is badly off cadence and still entirely credited here. That question has its own surfaces — the servicing traffic light, and Field operations → Against standard — and deliberately does not move these figures.
Credited does not mean armed. A credited night is one where the trap had been visited recently enough that its bait should still have been attracting — not one where it was known to be baited, set and empty. Ground within range counts as covered for every night of that window whatever the gear was actually doing. Read it as recency of servicing, expressed spatially , never as predator suppression. Whether a single-set trap had already fired and stopped hunting is a different question with its own model — trap saturation , fitted per line from how often traps are found sprung — and because that happens most where pressure is highest, the areas with the most predators get the most flattering coverage.
The coverage window is deliberately generous, which is the wrong direction here. It is a stated assumption about how long a lure keeps attracting after a visit, set generously so that the nights it removes are ones nobody would defend as coverage. Catch yield per 100 trap-nights on this data runs 0.61 within 30 days, 0.38 at 30–60, 0.29 at 60–90. Coverage credits every night inside the window at full strength, so a trap catching at half rate still paints a full-strength disc. Defensible as a claim about the bait, optimistic as a numerator of area controlled.
The window depends on the gear, so areas trapped with different gear are not compared like for like. Self-resetting traps declare a longer window than hand-baited ones, and rightly: a gas canister and a long-life lure do outlast a re-baited box, and leaving one unvisited for a season is what the machine is for. But it means an AT220 holds ground green for months unvisited while a DOC200 beside it fades at the project's own window — gear, not servicing, separating the two figures. It is negligible at present (a handful of such traps, a few dozen trap-nights) and would stop being negligible if the network shifted toward self-resetting gear.
The denominator is the whole polygon, including water. An area fronting the sea is capped below 100% by ground no trap could ever cover, so read maintained against designed rather than against 100%. Perimeter geometry does the same thing more quietly: a boundary line spends half of each disc outside the polygon.
Two artefacts can read as neglect. Where a trap's history hands over between record sources the gap belongs to no interval at all, so it shows as uncovered ground that is data lineage rather than a lapse. And a retired trap's coverage stops at its last check — that one is evidence, not an assumption: we were told the trap is gone.
A catch can change a past figure. Because an interval that caught something keeps all its nights, a capture recorded later can raise coverage for nights already gone. That is the intended reading — the catch is evidence about those nights — but it means a season's coverage is not final until its checks are in.
The figures are computed on a 25 m grid , which is why they can differ by a tenth of a point from an exact geometric answer. The grid, the radius and the coverage window are all published in the analysis pack so a number here can be reproduced from it.
The metres one trap is modelled to control. Starts at the project's configured radius; nudge it for a what-if.
How much of each surveyed area does your control network physically cover — and where are the holes? A model: the green area is ground within the control radius of a working trap.
A single
where to act
picture. The shaded field is
predator pressure
from the cameras — high (
red
) where the chosen
predator
is active
and
the chosen
protected
species is scarce. Onto it you can lay where trapping is actually reaching.
It synthesises both devices: the cameras say where the pressure is, the traps say where control is happening. A red patch with no catches or traps over it is an uncovered frontier — somewhere predators are winning and nothing's pushing back.
Hint: select a single Group and zoom in — the surface and the per-camera scores are most readable at group scale.
Toggle these top-right.
Each layer carries its own encoding
— the score layers use the ramp for the value itself, the trap field uses colour for servicing;
the bullets below say which is which.
Cameras for the selected period & group. The two trap layers follow the Traps shown chip instead — by default every trap within 2 km of that group's footprint, so the neighbours that explain a frontier are kept. Everything redraws when you change them.
Ground shaded where predators are busy and protected species scarce — camera by camera, and the ground between them.
An
expert-mode
map: pick any species (or a group like Mustelids) and see
where the
cameras
detect it against where the
traps
catch it —
the two devices on one canvas, for the chip bar's period and group.
On the cameras, colour says the role — predators (red), protected (green), other (grey) — so a mustelid never looks like a kiwi. Shape says the device : a disc is a camera detection, a square is a trap catch (purple — a removal). Size says how much . Where more than one role is at a camera the disc is split (e.g. half red / half green) so neither hides the other — click it for the full per-species breakdown .
Markers (chip): By role draws one marker per location — the camera disc split by role, the trap catch in purple, and (where tunnels run) the tracking index as a pressure-ramp disc; By species gives each picked species its own toggleable layer (cameras role-coloured, catches purple, and a role-coloured tunnel arch for the tracking index; markers fan out where a species' devices share a spot).
Layout switches between one map and two — side by side or swipe . In two-map mode a Second map section appears inside the Data period , Species and Boundary class chips: set one to give the right map a different period ( prior period , same period last year , or any other), a different species, or different ground — interior on the right against the boundary on the left , say. Leave a Second map section alone and the right map mirrors the left for that setting. Link views pans/zooms them together. The group, locality and line are always shared.
Boundary class (chip) filters the map by where a device stands relative to the protected areas : Interior (inside one), On boundary (on the line, within the survey band either side), or Beyond boundary (outside every one). The same measurement the Devices and Lines tables carry as a column. It cuts across the group and locality chips rather than nesting under them: one locality can hold traps inside a reserve, on its fence line and well outside, and this is how you ask for one of those.
Why there is no "approach" option, and no approach collars on the map: an approach belongs to a named area , and the collars overlap — traps between two reserves are in both. Drawn for every area at once the collars can only shade the nearest, naming no area at all, and "every trap in a final approach" has no single answer at the device level either. Both live on the Protected area report instead, where one area has been picked and each ring carries its own catch and effort.
Hint: select a single Group and zoom in — the markers are most readable at group scale.
Toggle these top-right.
All for the chip bar's period & group; everything redraws when you change them.
See all the data for any species on one map — monitoring (cameras) and control (traps) together, where it's detected against where it's caught.
A camera tells you where a species
was seen
, not everywhere it
was
— a cryptic animal can use a site and trip the camera rarely, or never.
Occupancy
models correct for that, estimating the proportion of camera sites a species
actually occupies. This page runs the
dynamic
version across your seasons.
The headline signal is the Trend — has occupancy risen or fallen year on year (this year's seasons vs the same seasons last year). That's the “is a predator spreading despite trapping?” answer, and it's what to read.
The model also reports two turnover rates, from how detections at each site come and go between consecutive seasons:
These say how much the species shuffles between sites each season (and together they set the level occupancy tends toward). They are not the trend — a species that just moves in and out with the seasons can have high γ AND ε while its occupancy is flat year on year. Both are corrected for detection p (so “went quiet” isn't mistaken for “gone”).
The line is modelled occupancy
ψ
per season with a
95% CI band
(bootstrapped); the
×
marks the raw naïve rate for reference. Read the
shape
— a rising band despite trapping means a predator is gaining ground.
The Trend compares the most recent year of occupancy (its 4 seasons averaged) with the year before — whole years, so seasonal ups and downs cancel rather than skewing it. It's called increasing / decreasing only when that change's confidence interval — from bootstrapping the model — excludes zero . That bar adjusts to each species' precision, so a well-detected species needs a smaller move to count than a rarely-seen one; anything else reads no clear trend , with the actual move shown in points. So a big-but-uncertain swing won't get a confident label.
Built on camera locations as sites, seasons as periods, and weekly detection occasions within each 21-night pulse. Only consistent-coverage seasons are modelled — early ramp-up seasons (before the full camera network was running) are dropped so a trend compares like with like. The Data period is a rolling 2-year window you slide with the “as at” season — a fixed length that always fits (the full span is too sparse for some species) and lets you see how the trend looked at earlier dates. Honest limits: at ~36 sites the bands are wide, so lean on clear separations, not small wobbles; sparse species can't be fit and are flagged; the first modelled season is the least certain; and baiting lifts detection (absorbed into p if consistent). Occupancy is about where (occurrence), not how many — and unlike counts it's immune to a weka or rat camping on the bait.
How the share of camera sites each species occupies has shifted across seasons.
The whole project as
data
, not charts — every group, season and
species_group on tidy tables, with the
raw counts and effort
behind every rate.
The Data bundle (.md) is one self-contained file (the brief, then the guide to reading the columns, then every table) — upload it to an AI assistant and ask for the assessment. The context estimate tells you how big a model you need.
Or take it as two files : the brief is the ask alone (~43 KB) and the data file carries the tables plus the export metadata and column definitions (~815 KB). The data file stands on its own — upload it and simply ask it questions, no brief needed. It is also the same file for every brief and register, so you upload it once and re-run reworded briefs against it. Both carry an export fingerprint naming the scope they were cut from.
The Data bundle (.pdf) is the SAME content, as a presentation-ready document (prompt on portrait pages, the data tables landscape with compact type) — pair it with the AI's report output to show exactly what was asked and the data behind it. The All data (.zip) is every table as CSV plus the brief and the column guide, for anyone who wants the raw tables.
Outcomes are reported
per species
(`species`, e.g. Stoat / Ferret /
Weasel), with each row carrying its
species_group
(Mustelids) — so the combined
trapping-target figure is a sum over the group's rows and the per-species view is always there.
It also carries
every
predator group (rats, cats, possums…), closing the
mustelid-only Group-report blind spot.
Today the assessment is generated
externally
(upload the bundle to a
Claude/AI session). The data rebuilds whenever
ik_data
changes, so a future in-app
generator can cache one report per data-import.
Two independent settings shape the brief. Neither changes a single number in the tables:
They compose, so you can ask for the full operations analysis written for someone who has never met a tracking tunnel — previously only the Volunteers brief was plain-language, and it drops the analytical chain.
Register never changes content. It cannot alter a figure, a confidence rating, a verdict, which findings appear, or whether a failure is reported — two registers over the same data must be reconcilable line by line. That floor is written into the export itself.
Each register option ships a built-in instruction. To override one for this deployment,
add
register-<axis>-<mode>.md
(e.g.
register-candour-encouraging.md
) beside the brief files in the data dir's
analysis-pack/
folder.
Two optional project files feed the brief, with
different
roles:
Rule of thumb: a fact to assume → context.md; a claim to test → hypotheses.md.
Each group, season and species_group as tidy tables — raw counts + effort behind every rate. Set the Data period and Group scope in the chips above and the audience and register in the Report brief tab, then download the bundle and have an AI assistant generate the assessment.
The first view of a new Data period or Group scope builds the analysis dataset — up to a minute, then it is cached until the next data import. The file sizes and the brief preview fill in when it lands, and the downloads wait for it. Settings in the Report brief tab need no rebuild: they change the brief text only.
The tables below hold every row of each grain (paginated) — the same data as the downloads above.
The instructions that travel with the data — what to produce, and how to say it. None of these four settings changes a number in any table on the other tabs; the brief below rebuilds as you change them.
The preview rebuilds as you change these. On a scope that has not been built yet it waits for the analysis dataset first — the spinner below means it is building, not that nothing happened.
reporting_group × season × species — the core outcome table. Split groups resolve to their members (Stoat / Ferret / Weasel) and every row carries its species_group, so a group figure is the sum over that group's rows.
One row per trap with coordinates — the only per-device frame on the trapping side. No season and no species axis; servicing_state is AS AT THE EXPORT DATE, with the declared interval, the signed days past it and the two check dates beside it so recency and any tolerance can be recomputed. Traps with no reporting_group or no coordinates are kept: NA is a fact about the records.
terrain axis × species_group × band — catches per band of GROUND (elevation, distance to water, distance to a formed track), TRAPPING only. Devices stand where people can walk, so terrain partly explains where the gear is: read `baseline` (peers beats effort), and treat `enough = FALSE` or `concentrated = TRUE` as overriding any ratio.
RARE-PREDATOR EVENTS — a predator absent across its whole monitored history in a group that is now present, at any level. Rarity is measured per group against its OWN prior monitored seasons (p_appear, Jeffreys-corrected), not against a fixed threshold, so the set maintains itself as abundances change. Seasons with no camera hours and no trap checks are never counted as zeros. Worst first.
reporting_group × season — trapping health: servicing %, cadence, catch rate + catch-per-check %, coverage is on `ep_operational` as an area-time mean over the season, effort.
reporting_group × trap_line × season — per-line traps/checks/captures, catch rate + catch-per-check %: the grain servicing effort and data-entry behaviour actually vary at. TRAP lines — a different set of places from the camera lines.
reporting_group × trap_line, NO season axis — what the catch numbers could not record. A trap that fires early sits dead for the rest of its interval, so catch_rate_recorded understates a slow-checked line: arrivals_recovered_pct is the share that still got recorded, catch_rate_offered what the traps were actually offered. recommended_interval_days is populated ONLY where saturation is the binding constraint — a blank means it is not what limits this line, never that the line could be checked less often. peak_season is where the pooled figure is weakest. All-data per line (a single season rarely gives one line enough checks), so never trend it. Read `confidence` beside every row.
reporting_group × trap type (cameraModel) — n_traps + pooled checks/captures + catch-per-check %; flags any self-resetting gear.
reporting_group × season — the CAMERA monitoring window (start/end dates, days). Trapping is continuous; cameras run a window each season.
reporting_group × season × species_group — rodent tracking-tunnel RTI % with the survey window (start/end); rti_pct is the DOC per-line rate, tracked/valid_tunnels the raw counts.
network × season × species_group — naive + modelled occupancy ψ. NETWORK-WIDE (all camera sites pooled), NOT per group.
network × species_group — colonisation γ / extinction ε + start→end verdict ('are predators spreading despite trapping?').
per CAMERA line (reporting_group + camera_line) — protected hotspots with weak nearby trapping. The *_nearby columns and nearby_trapping_status describe traps within the gap radius of the line, NOT the line itself — a camera line is never the thing that is overdue. Worst first. Here predator/protected POOL every predator group (Mustelids, Rats, Cats, Hedgehogs, Dogs) and protected group (Kiwi, Weka); the per-predator 'Gaps — …' tabs break predators out vs Kiwi alone.
protected area × season — each protected area's structure (size, boundary band, device counts per class) and its coverage. The coverage % pools every trap in the project by distance, so it can exceed what the class counts suggest.
Protected area × season × class × species_group × signal — inside vs outside the boundary, with raw counts and effort behind every rate. Read pooled, not metric; the contrast is internal against edge + external pooled. A blank pooled with n_devices 0 means no devices of that kind on that side, not zero animals.
protected area × locality — how every named block of ground stands relative to each area: distance to the polygon (0 = inside) and TRAP counts per distance band, plus the approach trapping by ring. Keyed on locality, not group (a third of this project's traps carry no group); the distance bands count every trap on the block INCLUDING those inside the polygon, so they are not the rings. How ground in no area's report becomes legible.
ep_taxa_absent — taxa this area has NEVER recorded by that signal, so they carry no ep_activity rows. The honest form of a zero series; the effort is on the rows that remain.
per camera line — Kiwi hotspots with weak/uncaught Mustelids control nearby (this predator group ALONE vs Kiwi). Worst first.
per camera line — Kiwi hotspots with weak/uncaught Rats control nearby (this predator group ALONE vs Kiwi). Worst first.
per camera line — Kiwi hotspots with weak/uncaught Cats control nearby (this predator group ALONE vs Kiwi). Worst first.
reporting_group × camera_line × season × species_group — the CAMERA per-line grain (detections + effort + RAI). Camera-line labels repeat across groups, so a line is the group AND the label.
per camera SITE, pooled over ALL seasons, with coordinates — the spatial signal (where activity concentrates).
the project's species_groups classification — role, priority, split, members.
per group — spatial + effort summary (cameras, traps, area, density, centroid).