Caribou Habitat Analysis

Woodland caribou · Ontario · 2020 and 2015–2025

Seven Ranges, Checked and Extended

We rebuilt the 2020 caribou habitat analysis of Mackey et al. (2024) from the same data, then carried it forward to 2025 using only the extra records that satellite imagery confirms.

Ranges
7Berens to Kesagami
Area analysed
–
ranges inside managed forest
Pixel
30 mLandsat scale
Annual maps
112015 to 2025

Forest-dwelling caribou, Parc national des Grands-Jardins, Quebec · Photo: Alain Caron, Wikimedia Commons, CC BY-SA 4.0 (cropped)

Ranges reproduced within ~3 points–of the paper's 2020 figures
Ranges above the 35% threshold–Environment Canada, 2025
Core habitat, 2015 → 2025–all seven ranges

Range by range

What a range is, and what goes into each one

Ranges are Ontario's planning units for boreal caribou. Each is drawn to contain one local population from aerial surveys, telemetry, historical occurrence and calving and wintering areas, with edges set along rivers, lake chains, highways and administrative lines. They are not herd boundaries: ranges include unoccupied habitat, and Sydney, Berens and Kesagami share their populations with Manitoba or Quebec. Like the paper, every figure here covers the part of each range inside Ontario's managed forest, outlined in the maps.

Woodland caribou range

Churchill

Churchill · Land cover, 2019
The paper's inputs
Results, 2020

Method

From land cover to core habitat

Five steps turn national land cover into core habitat, as in Mackey et al. (2024). Steps 3 to 5 run once for each year: harvest counts from 1980, roads accumulate by construction year, and fire counts for the previous 35 years and not after 2018, which gives 1985 to 2018 in 2020, the years of the authors' fire layer.

Processing chain each input feeds the step above it
STEP 1AlignOne 30 m grid in EPSG:3978 for every layer
  • Caribou range boundariesLand Information Ontario
  • Forest Management Unitsthe managed forest
STEP 2Potential habitatConiferous forest + treed wetland inside the range
  • Forest land cover (VLCE2)NRCan · 2019 · 30 m
STEP 3DisturbanceHarvest, roads and human footprint buffered 500 m, plus fire
  • Forest Resources InventoryOntario · harvest, planting, unclassified
  • Forest harvest 1985–2020NRCan · 30 m
  • MNRF Road Segments + NRNby construction year
  • Land Cover Compilation v2classes 18, 25, 27, 28
  • National Fire DatabaseNRCan · fire polygons
STEP 4Undisturbed habitatStep 2 minus step 3
STEP 5Core habitatMSPA: strip a 30 m edge, join patches, size them
  • Five patch-size classes100 · 500 · 1,000 · 10,000 ha
Close-up

Land cover, 2019

Follow one 27 × 27 km window in the Churchill range. Every 30 m pixel carries a class from NRCan's national forest land-cover map.

Potential habitat

Keep the classes caribou use

Only coniferous forest and treed wetland stay in. Everything else is set aside before any disturbance is considered.

Potential habitat: – of the window

Disturbance, 2020

Union every disturbance layer

A pixel counts as disturbed if any source touches it: harvest since 1980 from the inventory or the national Landsat record, planted stands, roads, mines, settlements, farms, corridors and the land-cover disturbance class, all buffered 500 m, plus fire from 1985 to 2018. Overlaps count once.

Disturbed: – of the window

Undisturbed habitat

Subtract disturbance from habitat

What remains is potential habitat that no disturbance reaches. In this window, – of the potential habitat is lost.

Undisturbed habitat: – of the window

Core habitat

Strip the edges, size the patches

Morphological Spatial Pattern Analysis removes a one-pixel edge, joins touching pixels into patches and measures each one. Patches under 100 ha are treated as sub-functional; the rest are grouped at 500, 1,000 and 10,000 ha.

Core habitat: – of the window · – of it in patches over 1,000 ha

Land cover, 2019 Churchill range · 27 × 27 km · 30 m pixels
Land cover in the close-up window: mostly coniferous and mixedwood forest with lakes. Potential habitat: coniferous forest and treed wetland in green. Disturbance in 2020 shown in plum over potential habitat. Undisturbed habitat in green; habitat lost to disturbance in pale plum. Core habitat coloured by patch size, with eroded edges in pale green.

The paper's result

All seven ranges reproduce within about three points

Share of each range disturbed, 2020
Mackey et al. (2024), Table 4 This analysis Range inside managed forest, water included, the basis of the paper's Table 4. Bands mark Environment Canada's risk classes; above 35% a range is considered unlikely to support a self-sustaining population.
Show the numbers

Results, 2015–2020

The paper's method, year by year

Share of each range disturbed, 2015–2020, the paper's data only range inside managed forest · hover for a year · ring = paper's 2020 value

Our data lines

Two additions carry the analysis past 2020

The paper's inventory ends with each forest's last survey and its harvest layer ends in 2020, so on the paper's data alone nothing changes after 2020. Two sources continue the record. The next section tests them against satellite imagery before they are used.

What we added

Two extra sources, and only two

The paper's inventory ends with each forest's last survey, and its harvest layer ends in 2020. To continue past 2020 we added the Analysis Ready Inventory 2025 harvest records, which carry Ontario's annual reports to 2023, and the National Burned Area Composite for fire from 2000. The satellite check below shows why these two were kept and the inventory's other records were not.

Everything else is the paper's method, including the land-cover disturbance class and the managed-forest basis used for the paper's comparison. Harvest counts from 1980, as in the paper. From here on, fire counts for 40 years, as Environment Canada does.

What the two added sources contribute range inside managed forest

Where

Where the two additions fall

Harvest reported after 2020, and burned area the National Fire Database does not record, for each range.

ARI harvest, 2021–2025

Satellite check

Harvest records hold up; most extra fire and blowdown records do not

How the check works

Look for the canopy loss each record claims

For 400 random pixels in each group, we took the year the record gives and built cloud-free summer–fall images (15 June to 15 October) for every year from two before to three after. From each image we computed the Normalized Burn Ratio (NBR) and NDVI; harvest, fire and blowdown all lower both.

A record counts as confirmed if NBR drops by more than 0.1 between the year before and the lower of the two years after. Allowing the second year after catches harvest done after the year it was reported.

Two control groups show what the test can and cannot see:

  • Control +: harvest recorded by both the Ontario inventory and the national Landsat layer, known to be disturbed. Its rate is the most a real record can reach.
  • Control −: habitat with no record in any source, known to be undisturbed. Its rate is the false-alarm floor.

Imagery: Landsat 8/9 Collection 2 (NASA / USGS) for records from 2013 to 2020; Sentinel-2 L2A (ESA Copernicus) for 2018 to 2024.

Share of records confirmed

Bars show the share confirmed; lines show the 95% confidence interval (Wilson), from about 400 pixels per group.

Do the records show up in the imagery? median NBR and NDVI per group, aligned on the recorded year
Four panels of NBR and NDVI over time. Harvest and recent fire records drop sharply at or after the recorded year, like the positive control. Extra fire and blowdown records stay flat, like the negative control.

A real event drops between the year before and the year or two after. The flat lines match the negative control.

Size of the drop per group dNBR, before minus after
Box plots of the drop in NBR per group. Harvest and recent fire sit well above the 0.1 threshold; extra fire and blowdown sit on the control line.
Surface reflectance before and after median per band
Spectral signatures before and after the recorded event. Confirmed groups show NIR falling and SWIR rising after the event; the others barely change.
What went into the 2015–2025 maps

Newer years, 2021–2025

The ranges since the paper

Share of each range disturbed, 2015–2025 range inside managed forest · hover for a year
Show the numbers

Disturbance by type

What the disturbed area is made of

Disturbance by type

Where layers overlap, a pixel is counted once, human layers before fire. Everything except fire is human disturbance; fire is the only natural disturbance in the paper's method.

Since the paper

What changed between 2020 and 2025

Core habitat, 2020 and 2025 drag the slider to compare
2020 2025

Human or natural

What is disturbing each range: people or fire

Human disturbance is harvest, planted stands, roads, mines, settlements, farms, corridors and the land-cover disturbance class, all with their 500 m buffers. Natural disturbance is fire, the only natural cause in the paper's method. Where both cover the same ground it is counted as human, as Environment Canada does; that overlap is shown separately.

Human and fire disturbance, 2015–2025 % of each range · hover for a year
Human Fire only 35% threshold

Risk class

Human disturbance alone puts ranges past the threshold

Share of range disturbed under each cause

Core habitat

How much core habitat each cause takes away

Core habitat under four scenarios

Core habitat as % of each range's potential habitat, with 3×3 core extraction run on each scenario. "No disturbance" is the ceiling set by how patchy the land cover itself is.

Core habitat lost to each cause, 2015–2025 k ha · compared with no disturbance
Lost to human disturbance Lost to fire

Where

Human and fire disturbance on the map, 2025

Disturbance by cause, 2025
Human only Fire only Both Water Colour strength shows how much of each display cell is disturbed.

Caveats

Before quoting a number

Keep in mind

  • Percentages are of each range inside Ontario's managed forest, lakes included, as in the paper.
  • The remaining 1 to 3 points come mostly from roads: the road layers are only published as current editions, newer than the paper's 2020 and 2022 copies.
  • The reproduction and the 2015 to 2020 results count fire from 1985 to 2018, as the published layer does. Once our data lines are added, fire counts for 40 years, as Environment Canada does.
  • Fire drops out after 40 years. Sydney's dip in 2024 is its 1983 fires ageing out, not recovery on the ground.
  • Harvest and road inputs end around 2023, so several ranges show identical values for 2023 to 2025.
  • Roads never recover: closed and decommissioned roads stay disturbed.

Data sources

  • Caribou range boundaries and Forest Management Units: Land Information Ontario
  • Forest land cover 2019 (VLCE2) and forest harvest 1985–2020: NRCan, Canadian Forest Service
  • Forest Resources Inventory, Packaged Products v2 (17 management units): Ontario MNRF
  • Ontario Land Cover Compilation v2.0 (classes per the 2014 data specification): Ontario MNRF
  • MNRF Road Segments: Land Information Ontario · National Road Network v18: Statistics Canada
  • National Fire Database and National Burned Area Composite: NRCan, CWFIS
  • Analysis Ready Inventory 2025 (harvest records only): Ontario MNRF
  • Landsat 8/9 Collection 2 (NASA / USGS) and Sentinel-2 L2A (ESA Copernicus), via Google Earth Engine