Draw one or more areas on the map, or load a KML or KMZ on the Inputs tab. Draw more than one to compare them.
No areas drawn yet
Points of interest
3 · What do you want to see?
4 · Trend settings
20%
Lower is stricter. 20% or below is recommended for reliable composites — raise it only if too few clear images come back.
Default is the full year. Narrowing this to the same months each year (e.g. your peak growing season) stops a wetter or cloudier season in one year from being compared against a different season in another — Sentinel-2 revisits every ~5 days, but how many of those are usably cloud-free varies a lot by season.
Pick more than one index to compare them as separate lines — but only when exactly one area is drawn. With more than one area drawn, only the first selected index is used, comparing that index across areas instead.
Shows whichever index button above is selected first (if more than one is active) — that button gets a black border while the preview is on, and "1 of N selected" in the legend below means the same. Standard scale uses the same colour range on every area and year, so tiles are comparable to each other. "This area's range" stretches the colours to the 2nd–98th percentile of the drawn areas (all of them together, so multiple areas stay comparable with each other), which makes subtle variation stand out more — at the cost of no longer meaning the same value from year to year or run to run.
Runs ~12x more sub-queries (one per month instead of one per year, per area), so it's slower — but it's the only way to see the normal seasonal wave and get a real, this-location-specific baseline for "how much does this area's index normally fluctuate."
Each date becomes a cloud-masked median of every clear Sentinel-2 image within the window around it (using the cloud % above), and the two are subtracted pixel by pixel. Pick dates at the same time of year so ordinary seasonal change doesn't dominate. NBR is the default here because it responds most strongly to bare ground and clearance.
3× typical
With "allow for area-wide change" on, each pixel's change is measured against what the area as a whole did (its median change), so a season or drought that shifts everything together cancels out and only pixels that changed differently from the rest are flagged. The slider sets how different: a pixel is flagged when it's more than that many times the area's typical spread of change (a robust spread that the outliers themselves can't inflate). Lower = more sensitive, more false alarms; higher = only the starkest changes. Changes under 0.05 are never flagged. Draw the area wider than the suspected clearance, taking in surrounding untouched vegetation — if most of the area was cleared, the clearance becomes the "typical" change. Turn the checkbox off to measure raw change from zero instead.
Loss
Gain
slightmoderatestrong
No clear view (cloud or shadow on every pass)
Low confidence (only one clear image)
Click a flagged patch on the map to see when it changed.
Shades show how big a flagged change was in index units — slight 0.05–0.15, moderate 0.15–0.3, strong over 0.3 (measured from the area-wide change when that's on). These are sizes of change, not burn severity: on the standard burn-severity scale for NBR (Key & Benson 2006) a 0.27–0.44 drop is only "moderate-low". The slider still decides which pixels are flagged at all. These are rough defaults, not calibrated thresholds: a strong NBR drop in forest usually means the canopy is gone, but the same number means something different in grassland or scrub. Grey pixels had cloud or shadow on every pass in one of the windows, so they couldn't be assessed — not "no change". Light grey pixels rest on a single clear image, which a missed cloud edge or shadow can easily throw; widen the window to firm them up.
The control is analysed with the same methods, and the report sets the two side by side: when and how much each was disturbed, how much is recovering, and their whole-area trends. Pick a control of the same kind of vegetation, close by and not much smaller (a reference-site comparison, as in Before-After-Control-Impact designs).
Sensitive confirms a break on 4 clear images in a row at a 95% chi-square probability, so smaller or patchier drops in the canopy signal (such as scattered dead canopy trees) are picked up, with more false alarms. Standard uses CCDC's published 6 images at 99%.
Published methods at their published settings, apart from break detection when Sensitive is picked (details in About and in the report's Appendix A). Disturbance and recovery: CCDC (Zhu & Woodcock 2014) on every clear Landsat 8/9 image since 2013 (NASA's Harmonized Landsat Sentinel-2 Landsat product), on its 30 m grid (0.09 ha pixels). A disturbance is a break in the period where the chosen index and its fitted line fall (by at least 0.05); it is recovering when the fitted line after the break rises. Falls CCDC misses (it is blind for a year or more after each break) are caught on Sentinel-2: a pixel whose yearly value, against the area's median, fell by 0.1 or more from one year to the next and stayed down the year after. Steady decline or improvement: Mann-Kendall and Theil-Sen on each 10 m Sentinel-2 pixel's yearly values, allowing for the season, with a false discovery rate correction. Nothing is tuned to particular sites. The smallest spot only decides what is reported. Single trees and patches under about 0.1 ha are below what 30 m pixels can map, and every result is unvalidated until a validation sample has been labelled.
1 · Areas
Draw one or more areas on the map. Draw more than one to compare them.
No areas drawn yet
2 · LiDAR survey
Draw an area to see its surveys
3 · Layer
auto
1 · Area
Draw or pick one or more areas: they all load together, to compare. Point clouds are big: even a hectare usually means reading 20 to 60 MB, so start small.
No areas drawn yet
2 · Survey
Draw an area to see its surveys
3 · Load
4 · Check the trees
1 · Earth Engine
2 · Areas
Draw one or more areas, or add them from a layer. Each gets its own climate.
No areas drawn yet
3 · Climate
to
Normals are always 1991–2020, and the long-term change always runs from 1950, whatever years are picked.
Inputs
Every file goes in here. Drop files on a box, or anywhere on this tab: KML and KMZ become areas, CSV files are read as occurd. exports, and GeoJSON becomes a private layer. To add a KML as a layer rather than as areas, drop it on the Layers box.
1 · Areas
Each polygon placemark becomes an area, named as in the file. You can also draw areas on the map on any tab.
2 · Species records from occurd.
The records CSV (Occurrences CSV, which the Biodiversity tab maps) or the checklist CSV, exported from occurd. with NZTCS status unlocked. For a citable report, fetch in occurd.'s Big Data Mode: its exports carry the GBIF download DOI, which becomes the source of the species facts. It's used exactly as exported: statuses are occurd.'s, including any conflicts you settled there, and records are matched to the areas by occurd.'s polygon name. The files are kept with the open report.
3 · Layers
Each file becomes a private layer on the map, listed with the reference layers. Private layers stay in this browser: never uploaded, and never in Export for Louped or other packages.
No areas yet: draw on the map, or load a KML on the Inputs tab
Load KML and KMZ files on the Inputs tab. While this report is open its areas are the ones on the map, on every tab.
3 · Species from occurd.
4 · Fact sheet
Runs the checks on the other tabs for these areas (or uses their results if they're already there for the same areas), and reads the occurd. files. The results also stay on those tabs and on the map.
5 · Export
A zip for writing the report elsewhere: the ticked facts with their IDs and sources, the areas as GeoJSON and KML, a map of the areas, and a summary of what was run. The writing, and every call on what the facts mean, happen outside inferrd.
For a client map: tick layers in the Louped column of the map's Layers panel, then make or update the site in the Louped tab.
1 · Earth Engine
2 · Areas
Draw one or more areas, or add them from a layer, then press Check areas against layers at the foot of the Layers panel. Only what turns out to matter for an area is reported; the rest is listed as checked.
No areas drawn yet
3 · Map
Tree cover loss, canopy height and surface water are under Trees & water in the Layers panel on the map, for your areas: switch each on or off, fade it, and drag it up or down. ✂ on an area cuts it by any of them, or by any other layer. For what covers the land now, switch on LCDB v6 among the reference layers on the Layers tab.
Layers
Switch on what you want on the map. The map's Layers panel then lists each tab's own layers, and everything that's on, wherever it came from; tick Louped there or here to put a layer in the client site.
Layer check
Tests every area against every reference layer, on or off, the same check as occurd.'s Projects > Layers. The Report tab's Gather the facts runs it too, and Export for report includes its CSV.
Published GIS layers, not the legal plan maps: confirm anything decisive against the operative plan. The CSV has a row for every area and layer, including those with nothing.
1 · Records
Export Occurrences CSV from occurd. for your areas and add it on the Inputs tab. The files are kept with the open report, so the Report tab's fact sheet uses them too.
2 · Selections
Each selection is a layer on the map, under Biodiversity in the Layers panel. Tick it in the panel's Louped column to put its records in the client site.
3 · Choose
Years to
Nothing ticked in a group means all of it. Untick single taxa in the list below the map.
1 · Client site
2 · Layers
Untick to leave a layer out of the client's site (it isn't sent at all, and its source leaves the About page). Saved as you tick.
More
3 · Wording
4 · Online
Where sites are kept
1 · Areas
Draw one or more areas, or add them from a layer, then describe them: geology, terrain, erosion-prone land and soils.
No areas drawn yet
2 · Describe
3 · Map
What was found goes under Physical in the Layers panel on the map: switch each layer on or off, fade it, and drag it up or down. Click the map inside an area for what is mapped there.
Use the polygon tool (top-left) to draw your area of interest
◌
Draw an area and get its climate.
Climate for
Average year, 1991–2020
Month by month against normal
Since 1950
Each year's rainfall and mean temperature, with the Theil–Sen trend line where the Mann–Kendall test finds one (p < 0.05).
Year by year
Reading this
❦
Add an occurd. records export, then choose the records to map.
◱
Choose a site, or make one from this map, to see it here as the client will.
✎
Start a report: add its areas and occurd. files, gather the facts, then export them.
What was gathered
Fact sheet
Each line is a measured value and where it came from, with no interpretation. Untick anything that doesn't belong in this report: unticked facts are left out of the export. What the facts mean is yours to say when the report is written.
◌
Draw an area and check it against the layers.
Checking
Checked, nothing to report
Download
Everything checked for every area, including what was left out of the page.
Reading this
◌
Draw an area and describe it.
Describing
Terrain
Geology
Erosion susceptibility and land use capability
Soil
Download
Every table above for every area, one row per item.
Reading this
◌
Draw an area and load its point cloud.
Details, trees and 3D for
Density by height
Where the pulses stopped
Returns
Classes
Trees
Reading this
◌
Draw an area and load the LiDAR.
Change between surveys
Ground change between surveys
Areas compared
Vegetation height classes
Spread of heights
Reading this
◌
Draw an area, unlock and run an analysis.
Vegetation index over time ()
Season used:
Monthly detail — seasonal pattern
Interpretation
Per-year data
Vegetation index over time () — comparing areas
Monthly detail — seasonal pattern by area
Same natural-variability baseline as the single-area view, computed independently per area — see the summary table below.
About this view
Each area gets its own trend analysis below — Mann-Kendall/Sen's-slope trend, the biggest single-year change, and (with monthly detail on) the natural-variability baseline are all computed independently per area, not pooled together.
Per-area summary
Are the areas different from each other?
Index comparison —
Monthly detail — seasonal pattern by index
Same natural-variability baseline approach as the single-index view, computed independently per index — see the summary table below.
About this view
Each index gets its own trend analysis below — Mann-Kendall/Sen's-slope trend, the biggest single-year change, and (with monthly detail on) the natural-variability baseline are all computed independently per index on this one area, not pooled together. The indices aren't on the same physical scale (EVI/NDVI vs. NBR/NDMI use different bands), so compare direction and trend significance across indices, not raw magnitude.
Per-index summary
When did this change? —
Patch minus surroundings
Interpretation
Per-image data (clear images only)
Whole area —
Spots over time
Each spot minus the median of the whole area (or of the control, when there is one), image by image (faint dots), with a 6-month moving median. Above zero: greener or wetter than the area; below: less. The bars above the chart show what each spot's flag compared: hollow = the years before, solid (and the light shading) = the years after, dashed = the year a dropped-and-recovered spot had grown back by, or the whole period for slow decline and growth. Click spots on the map, or tick them in the table, to add or remove them; the chart switches to the index the spot was tested on (NDMI for canopy decline).
Spots inside the area
Click a row to zoom to the spot; tick Chart (or click the spot on the map) to follow it over time. Every spot is a reason to go and look, not proof of what happened.
Validation sample
How often are these spots wrong, and how many are missed? Draw a stratified random sample of 30 m pixels: some inside each spot type and more outside every spot. Label each point from aerial photos or a site visit (reference_label: sudden_drop for a disturbance, dropped_recovered for one that is recovering, slow_decline, gain, or no_spot; leave it blank if you can't tell), then load the file back with "Load labelled validation sample" under Run condition check. Accuracy and areas are worked out as in Olofsson et al. (2014). Points are listed in random order and shown on the map without their type, so label them without looking at the stratum column if you can.
Accuracy —
Error matrix
Rows: what the map says (the stratum). Columns: what the point really was. Each cell: labelled points, then the estimated share of the whole area (each stratum's points weighted by its mapped area).
inferrd.
A free, browser-based tool that pulls Sentinel-2 vegetation-index history for any area you draw, anywhere in the world, and runs a trend analysis to say whether observed change over time is statistically robust or just noise.
How it works
Draw your area of interest on the map
Pick what you want to see: a trend over years (graphs of the index year by year) or change between two dates (a map of where vegetation increased or decreased, described below)
For a trend, choose a year range, an optional season window, a cloud-cover threshold, and an index (EVI, NDVI, NBR or NDMI)
inferrd. builds a cloud-masked composite from Sentinel-2 Level-2A surface reflectance (atmospherically corrected, not raw top-of-atmosphere imagery — avoids false "change" from atmospheric differences between dates) for each year. Cloud and cloud shadow are masked with Google's Cloud Score+ (Pasquarella et al. 2023), which catches thin cloud and shadow that Sentinel-2's own Scene Classification Layer often misses; the Scene Classification Layer still drops no-data, saturated and snow pixels. The index is computed at Sentinel-2's native 10m resolution
The per-year mean is run through a linear trend, a Mann-Kendall test and Sen's slope estimator — the same non-parametric approach standard for short, noisy environmental time series
The four indices
EVI and NDVI both measure how much near-infrared light bounces back relative to red light, which is a proxy for green leaf material — it works because healthy leaves reflect NIR strongly. NBR swaps red for shortwave-infrared band 12 instead, which makes it far more sensitive to bare ground and char left by fire or clearance than to leaf greenness itself — the standard index for burn/disturbance detection. NDMI uses shortwave-infrared band 11, which responds to how much water is actually in the canopy — it can drop under drought or water stress before the plant looks any less green, so it's worth checking alongside EVI/NDVI rather than instead of them. All four move in the same direction (higher = healthier/wetter, lower = stressed/disturbed), so the trend analysis and interpretation text below read the same way regardless of which one you pick.
This is built for vegetation — not water
All four indices rely on near-infrared reflectance, and water absorbs NIR almost completely, driving every one of them sharply negative regardless of what's happening biologically. If your drawn area is mostly or entirely water, the numbers aren't measuring vegetation at all — more likely turbidity, sediment, or algae/chlorophyll in the water column, none of which this tool's interpretation (dieback, drought, pests, clearance) was built to explain. The cloud mask here also deliberately keeps water pixels rather than excluding them, so a marine polygon gets no automatic protection — draw a land area for the results to mean what this tool says they mean.
Why "no trend" doesn't always mean "nothing happened"
Mann-Kendall and Sen's slope specifically test for a monotonic trend — a gradual, sustained direction across the whole period. A sharp one-off disturbance (storm damage, fire, clearance) followed by partial recovery is not monotonic, so these tests can correctly report "no significant trend" even when one year clearly dropped off a cliff. To catch that case, inferrd. separately reports the single biggest year-over-year change in the series and calls it out explicitly whenever it dominates the overall spread — check that figure, not just the trend line, if you're looking for a specific event rather than a gradual change.
"No significant trend" is also not the same as "stable". With the eight or so yearly values Sentinel-2 allows, the test only picks up a fairly large, steady change; with four or fewer values it can't reach significance at all. Read a non-significant result as "not shown either way". The test's variance is widened for any year-to-year carry-over in the series (Hamed & Rao 1998), never narrowed.
Change between two dates
Choose "Change between two dates" in section 3 to map which pixels changed between a "From" and a "To" date — for checking a specific event, like reported clearance. Each date is a cloud-masked median of the clear images in a window around it, and the two are subtracted pixel by pixel. With "allow for area-wide change" on, each pixel is compared with what the whole area did (its median change), so a season or drought that shifts everything together cancels out; a pixel is flagged when it differs from that by more than the slider's multiple of the area's typical spread. Flagged pixels are shaded by how big the change was; grey pixels had cloud or shadow on every pass and couldn't be assessed, and light grey ones rest on a single clear image. Draw the area wider than the suspected change, taking in untouched surroundings, and check the clear-image counts — a date backed by one or two images is easily thrown by a missed cloud edge or shadow. As with everything here, a flagged pixel is a reason to go and look, not proof of what happened.
To find when a flagged patch changed, click it on the map. inferrd. compares the patch with its surroundings (a ring 30–150 m out) on every image of roughly the last two years in which the whole patch is clear, and reports the two clear images the change falls between — where the difference crosses the flag threshold and stays there for two or more images. A one-image dip that recovers straight away is flagged as likely cloud shadow or haze, not change.
Condition check
Choose "Condition check" in section 3 for one area, such as a covenant or Protection Management Area added with "Analyse this area". It reports the area's trend over the years and traces spots inside it, with published methods at their published settings. Disturbance and recovery come from CCDC (Zhu & Woodcock 2014), which fits every clear Landsat 8/9 image since 2013 with a seasonal model and records a break once six clear images in a row depart from it; a disturbance is a break where the canopy signal fell, and it is recovering when the fitted line after the break rises. Steady decline and improvement come from a Mann-Kendall test and Theil-Sen slope on each 10 m Sentinel-2 pixel's seasonally adjusted yearly values, with a false discovery rate correction (Benjamini–Hochberg), since a test at p < 0.05 per pixel would flag about one pixel in twenty in an unchanging forest. Spots are 30 m pixels joined into patches of at least the smallest size; nothing is tuned to particular sites. Optionally, a second area is analysed the same way as a control. The results have not yet been checked against field or aerial-photo ground truth here, so their error rates are unknown until a validation sample has been labelled. "Export for report" saves the results as a zip for writing a condition report. A spot is a place to go and look, not proof of clearance or of a breach.
Sentinel-2 revisit rate & seasonality
Sentinel-2's two satellites together revisit most places roughly every 5 days, but how many of those passes are actually usable depends on cloud cover, which varies by season — the "Clear looks" column in your results table (how many clear views the typical point in your area got) is the real signal for how much data went into each year. "Scenes" counts every scene that passed the whole-tile cloud filter, including ones that were cloudy over your area. By default each year's composite uses the whole calendar year, which mixes seasons together; narrowing the season window to the same months every year (e.g. your peak growing season) keeps years genuinely comparable instead of accidentally comparing a summer-heavy year against a winter-heavy one.
Read this before trusting a result
Satellite vegetation index alone can't tell you why a change happened — dieback, drought stress, pest or weed incursion, and actual clearance can all look similar from space. Treat a statistically significant decline as a reason to send someone to look, not as proof of what's happening on the ground.
The Level-2A surface reflectance data this tool uses only goes back to 28 March 2017 (Sentinel-2's raw, uncorrected Level-1C data goes back further, to mid-2015, but that's deliberately not what this tool queries — see above). ESA only produced it worldwide from December 2018, so outside Europe 2017 and most of 2018 are missing or thin; 2019 is the first full year everywhere, and the default start.
Climate tab
The Climate tab gives each drawn area's climate from ERA5-Land, ECMWF's land reanalysis (a weather model run over the observations, about 9 km cells, hourly from 1950 to a few months ago), averaged over the area and 6 km around it. It runs on Earth Engine, so it needs the access key from the Infrared tab. It shows the 1991–2020 normals (rainfall, mean temperature, frost days, hot days, wet and heavy-rain days, growing degree-days), each month of the years you pick against its normal with the rainfall of the 12 months to then, every year since 1950 with a Mann–Kendall trend, and a table and CSV. Use it to check whether a change in the vegetation signal lines up with a drought, a wet spell or a cold year. A 9 km cell smooths the terrain, so totals in hill country can be well off a local rain gauge: it is for timing and for comparing with normal. Station records for New Zealand are with Earth Sciences New Zealand's (formerly NIWA's) DataHub, which needs its own registration and licence, so inferrd. doesn't use them.
Report tab
The Report tab gathers the data a report is built from; the writing, and the judgement calls, happen outside inferrd. Each report holds its areas (loaded from KML or KMZ files, one area per polygon placemark, or drawn) and species records exported from occurd. Reports are kept in this browser only.
Gather the facts runs the Physical, Trees & water and Climate checks on the report's areas (or uses their results if they're already there), takes LiDAR from the LiDAR tab where it's loaded, runs the layer check against every reference layer (occurd.'s check, so the Layers password must be entered; if it isn't, the facts say the layer check was not run), and reads the occurd. files. The fact sheet is a numbered list of measured values, each with its source and nothing interpreted: no "significant", no "mostly", no reasons. Species statuses are occurd.'s, as exported, and the species facts cite the GBIF download DOI when occurd.'s export carries one (fetched in its Big Data Mode). Untick any fact that doesn't belong. Export for report saves a zip: facts.csv (the ticked facts, keeping their IDs so a report can cite them), summary.json (version, areas, what was run and when, sources), the areas as GeoJSON and KML, layer_check_<date>.csv (the same file as occurd.'s Projects > Layers check), and map.png, the areas on aerial imagery. Export for Louped saves a package for a client map in Louped: the report's areas, the occurd. records (with their coordinates and GBIF DOI), and every vector layer that's on in the Layers panel, in its order, colours and opacity. Set the panel up as the client should see it before exporting. Image layers (index preview, LiDAR, Earth Engine land cover) aren't in packages yet and are listed as left out. occurd. files added before v0.62.0 kept no coordinates: remove them and add them again.
Layers
The Layers panel on the right of the map shows what matters on the open tab: that tab's own layers (the index preview, change map, condition trend and condition spots on Infrared once you show them, LiDAR once loaded, the Physical layers once that tab has described the areas, the three Trees & water layers whenever there are areas, the Biodiversity tab's selections), then Also on the map: everything else that is switched on, from the other tabs or the reference layers, so LiDAR heights can sit over geology, or tree loss over an index preview. The eye switches a layer on or off, the slider sets its opacity, dragging a row by its grip moves it up or down among its own kind, and the Louped tick puts it in a client site. The chevron folds the panel to the side. The Layers tab lists every layer: each tab's, and the reference layers, occurd.'s polygon layers (below), where you unlock and choose them. Reference outlines always draw above the data layers. The panel remembers its order, opacities and what is on in this browser; reference layers start off each time.
Check areas against layers, at the foot of the panel, checks every area against the Trees & water tab's layers (tree loss, canopy height and surface water) and against every reference layer. The results are on the Trees & water tab, each with its CSV.
Cut an area by a layer: ✂ on an area keeps only the part of it inside a layer, or only the part outside, as a new area named after both; the original stays. It can be any tab's layer: canopy over a height, tree cover loss, surface water, a slope class or elevation band, or a reference layer's shapes. Earth Engine layers are cut at their own pixels (10 or 30 m). "Cut areas to this" in a reference-layer popup cuts each area that touches that one shape.
Trees & water tab
The Trees & water tab checks each drawn area against global Earth Engine layers, so it needs the access key from the Infrared tab: tree cover loss by year since 2001 (Hansen / University of Maryland, 30 m), canopy height in 2020 (ETH Zürich, 10 m) and surface water since 1984 (EC JRC Global Surface Water, 30 m). Each is shown only when it finds something for that area: tree loss of more than about two pixels, vegetation 3 m or taller over at least 5% of the area, or any surface water. What was checked and found nothing is listed in one line each, and everything goes in the CSV. For what covers the land now, LCDB v6 (Manaaki Whenua) is among the reference layers on the Layers tab; the global land cover layers (ESA WorldCover, Dynamic World) and the fire layers (MODIS burned area, FIRMS) were taken out in v0.85.0, as LCDB is better for New Zealand and fires big enough for them to see are rare here.
Physical tab
The Physical tab describes what each drawn area is made of and how it lies. Geology: in New Zealand, GNS Science's 1:250 000 geological map (QMAP, 2023 edition) gives the map units in the area with their rock types, ages and descriptions, the mapped landslides, and the faults within 1 km and whether they are active; elsewhere, Macrostrat's compilation of geological maps is sampled at points across the area. Erosion and land capability (New Zealand): MPI's Erosion Susceptibility Classification, with the Land Use Capability class, subclass and unit it is built on. Terrain: elevation, slope (in the LUC slope classes) and aspect from the Copernicus 30 m elevation model. Soil: SoilGrids 2.0 (250 m) texture, stones, pH, organic carbon, nitrogen, cation exchange capacity and bulk density down to a metre. Geology and erosion are read straight from the publishers' map services and need no key; terrain and soil run on Earth Engine and need the access key from the Infrared tab. All of these are regional-scale maps: use them for the setting of a site, and check anything that matters on the ground.
LiDAR tab (New Zealand)
The LiDAR tab at the top uses LINZ's free 1 m elevation models instead of satellite imagery, and needs no access key. Draw or pick an area and inferrd. lists the LiDAR surveys that cover it, newest first, but those with data over all of it before those that only reach an edge (from each survey's LINZ capture area, since a survey's edge tiles can be mostly empty). If a loaded survey turns out to have no data, or almost none, over an area, the tab says so and names one that does. Load one to see vegetation height (the surface model less the ground model, DSM − DEM, from the same flight), a hillshade of the bare ground, a hillshade of the surface, or coloured elevation; with two or more surveys, the change in vegetation height or in the bare ground between the earliest and the latest, with contours of the bare ground over any of them (from the ground model, so they follow the ground under the trees), at an interval set by the height range, or at a fixed 0.5, 1, 5, 20 or 100 m ticked under Contours in the Layers panel. Contours of the surface (DSM) are a second layer, off at first: they follow the tops of trees and roofs, and keep embankments over culverts and bridges, which the ground model may cut through. View in 3D, under Load LiDAR, shows the loaded ground or surface model as a 3D model to tilt and turn, coloured as on the map (every layer that's on in the Layers panel, in its order and at its opacity, over the basemap: set the map up in 2D, then look at it in 3D), by an aerial photo, or by height on a choice of colour ramps, and with contours of the model shown (switched on or off in the 3D view); the heights can be exaggerated to bring out low relief. Aerial photos are LINZ's (Toitū Te Whenua, CC BY 4.0), read straight from the open nz-imagery bucket: the survey flown nearest the LiDAR is picked first, and any other over the area can be chosen. 0.5 m contours are traced at 1 m, fine enough for drains, terraces and scarps; they are best on open ground, since under bush fewer pulses reach the ground. The results pane gives the area in each height class, the spread of heights and a CSV. Click the map for the ground, surface and height at a point.
Heights are for the date the survey was flown. They include anything standing on the ground, buildings and power lines too. The 1 m grid also smooths narrow crowns and gaps. Areas up to about 3 km across are read at the full 1 m; bigger ones are read at 2, 4, 8 or 16 m. Data: Toitū Te Whenua Land Information New Zealand, CC BY 4.0, read straight from LINZ's open nz-elevation bucket.
Trees above the canopy (in the Layers panel, once LiDAR is loaded) puts a pink dot on each tree that rises at least the chosen height (2, 3, 5, 8 or 12 m) above the canopy round it: emergents over a forest and lone trees in open ground. The canopy round a tree is the height in a ring 8 to 20 m from its top that only 5% of the ring is above, so a hedge or a neighbouring tree crossing the ring doesn't count as canopy. Trees in an even canopy get no dot. Click near a dot for the tree's height and how far it rises. Best at 1 m; at 2 m and coarser, crowns blur together.
Several areas, several surveys. Draw more than one area and they load together, compared in one table and on one chart. Tick "Compare surveys over time" and pick 2 or 3 surveys to see how the same areas changed: a change table, and a Height change layer, red where vegetation dropped and blue where it grew. Surveys differ in sensor and processing, so changes under about a metre may not be real.
Point cloud tab (New Zealand)
The Point cloud tab reads the LiDAR itself, every return of every pulse, rather than LINZ's 1 m grids: OpenTopography's copy of LINZ's surveys, as LAZ files. Draw or pick one or more areas (up to six) and inferrd. lists the surveys with points over them, newest first, then sizes up the download; all the areas load together. Each tile's points are stored by flight line, not by place, so even a small area needs a good share of a tile: a hectare is usually 20 to 60 MB. Over 100 MB it warns before loading; over 1.5 GB it won't load, and asks for a smaller area. With more than 8 million points over the area it keeps an even share of them, so the 3D view keeps up.
Once loaded, the right shows where the pulses stopped: the returns in each height band above the ground, and each band's layer density, the returns in it as a share of those that got down to it (the band and everything below), so a layer under a thick canopy isn't read as thin just because few pulses got that far. Understorey density is that share for 0.5 to 2 m. Density by height draws it metre by metre, up from the ground (or the share of all returns in each metre); with two or more areas loaded, their lines are laid over each other to compare; a later load (another survey, say) adds its lines too, up to six. With several areas, Details, trees and 3D for picks which area the figures, tables and 3D view show; the map shows every area's trees. Also the returns (only, first, in between, last) and the survey's classes. Noise points are left out.
View in 3D colours the points by height above ground, true colour (where the survey has it), class, return or intensity. Turn it round, or double-click anywhere (or press Stand) to stand there at 1.5 m: drag to look about, W A S D or the arrow keys to walk, Shift to go faster, Esc to go back. Ground shows the ground as its points, as a solid surface (the ground model the heights are measured from, shaded to show its shape), or hidden, to see the understorey. Show takes the canopy away: Under the trees found leaves out each tree found from its crown down to its lower limit (30% of its height, at least 2 m), leaving what grows beneath it and in the gaps; Up to a height leaves out everything above it. With two or more areas loaded, Compare 2 areas side by side in 3D opens two views, each with its own area: with three or more loaded, pick which two at the top left of each view. To compare one area between two dates, load it from one survey, pick another survey under 2 · Survey and load it again: both stay loaded, the chart shows both dates, and side by side in 3D starts with the area at its two dates. Surveys differ in point density (older ones often have a quarter of the points), and a sparser survey sees less of the lower layers, so inferrd. warns when two dates differ more than twofold; with Move together on, turning, zooming, standing and walking in one view does the same in the other, from the same place in each area (untick it and each view moves on its own, steered when the cursor is over it), and the height colours use one scale for both. Each tree colours every tree found its own colour, with no two touching crowns the same; the same trees are dots on the map (Layers panel), with a table of their heights and crowns on the right and a CSV of every tree. Trees are found from the points alone (after Dalponte & Coomes 2016): best on trees with their own crowns, like pines, emergents and trees in the open; in an even native canopy, touching crowns can be read as one tree or one broad crown as two. Where the survey has photo colour, touching crowns that differ in colour are told apart. Data: Toitū Te Whenua LINZ and each survey's licensor, distributed by OpenTopography, CC BY 4.0; read with laz-perf.
Reference layers
Draw to select: with reference layers switched on, draw a shape round a QEII covenant, a PMA or any other of their areas, and a popup on your shape offers to use what's inside instead of your shape, add it as well, or keep only your shape. An area counts as inside when 90% of it is.
Reference layers (password protected, in the Layers panel) are occurd.'s layers: current land cover (LCDB v6), potential vegetation, the Vegetative Cover Map, threatened environments, protected areas, QEII covenants, NIWA river catchments and Gisborne District Council plan layers. The layers module and its layer list are loaded live from occurd., so a layer added there appears here too. "Analyse this area" in a layer popup turns that boundary into an area for a trend or change map. These are published GIS layers, not legal plan maps.
Your data, your account
inferrd. runs in your browser and talks to Google Earth Engine through a shared Rutherford Ecology service account. An access key is swapped, via a small Cloudflare Worker, for a short-lived read-only token — there is no Google sign-in. The boundaries you draw and your results stay in your browser and are never stored on a server.
If this tool saves you a whole lot of time, feel free to buy me a coffee.
Licence & source
Free to use for non-commercial purposes under PolyForm Strict 1.0.0; redistributing or modifying the code is not permitted. For commercial use, contact Rutherford Ecology. View source on GitHub.