Nature (and the likeness thereof), mapped across Vancouver Island

We have built a nature-likeness score for every 10 m pixel across Vancouver Island, the Gulf Islands and the Sunshine Coast. About 48,000 km2 of land and inland water, for the year 2024, on a 0 to 1 scale. High means the pixel looks like intact forest, wetland or open water. Low means it looks like the built environment. The sea is masked out.

The map below is the finished layer, so pan around it before reading on. You can also view it here, full screen.

map

It is built on Google Satellite Embedding V1 and public land cover, with each pixel scored by its similarity to reference land-cover signatures. There is no training data and no hand-drawn polygons anywhere in it.

The map

Pan and zoom anywhere in the region. Click a pixel and you get its score, and where that score sits against the rest of the region. Drag the two ended slider under the legend to show only part of the range, and the area in km2 updates as you drag, so you can ask "how much land scores below 0.4" and read the answer directly. Park and protected area boundaries toggle on top.

The median score of natural likeness across the region is 0.88. This is high, as the region is a forested coast. The interesting parts are the tails of the index: the settled strip down the east side, the working forest around Campbell River where a quarter of pixels fall below 0.62 (cut blocks!), and the Cape Scott wind farm at 0.693 against roughly 0.9 in the trees around it.

Natural-likeness and protected areas

We crossed the layer with BC's protected area boundaries: national and provincial parks, ecological reserves, conservation lands, and regional and municipal parks. An quick afternoon's work that uses the index and open data.

A quarter of the region scores 0.90 or higher and has no protection at all. That is 11,862 km2, roughly two and a half times all the provincial parkland here.

What you can do with this information

• Screening for acquisition or covenant. Find high scoring land with no designation, filter to a region or watershed, and rank candidates before anyone drives anywhere.

• Offsetting and no net loss. A consistent before and after measure across a whole study area, rather than a patchwork of inventories with different vintages.

• Cumulative effects and referral review. A common yardstick for "how modified is this already" when assessing what one more project does.

• Local plan indicators. An Official Community Plan or Regional Growth Strategy naturalness indicator that can be recomputed each year from public data.

If you have a boundary you would like run against this layer, or a question it looks like it could answer, get in touch!

Appendix for nerds

• Inputs. Google Satellite Embedding V1, a per pixel representation distilled from multi sensor satellite imagery, plus public land cover as the reference. Each pixel is scored by its similarity to reference land cover signatures. There is no training data and no hand drawn polygons anywhere in the layer.

• Resolution and extent. 10 m native, 2024, about 48,000 km2 of land and inland water. The sea is masked out. Lakes and rivers are kept, and score high, because open water is part of what the layer treats as unmodified.

• Above treeline. Rock and ice do not look like forest, so alpine land scores lower without being disturbed. Carmanah Walbran, old growth on a valley bottom, averages 0.91. Strathcona, roughly half of it alpine, averages 0.86. Neither is degraded.

• What it is not. Not a measure of ecological integrity, habitat value, or biodiversity. Not a substitute for field assessment. Values are comparable within this region, not against a layer built separately for somewhere else. Special protection areas are a forestry designation rather than a conservation one, so we report them separately from parks rather than folding them in.