← Back
Canopy

How it works

What Canopy is, the three scoring models, and the three pipelines behind them.

Canopy is a pan-African conservation intelligence and finance platform covering all 54 African countries and 200 key protected areas. It is designed to support anyone deploying conservation capital across Africa, giving them an independent, apples-to-apples comparison of country financial need, country risk, country conservation performance, and protected area performance. Beyond Africa, it is adding the world’s most important protected areas from the Canopy Global 1000, starting with 331 across 32 countries in Asia, Oceania and the Americas.

Using 3 scoring models and 3 data pipelines (including 11 satellite feeds), Canopy brings real-time ecological, financial, and geopolitical intelligence into natural capital investment decision-making. The platform also provides an independent monitoring and evaluation tool, tracking conservation performance and land-use of protected areas and buffer zones over time.

Canopy: three scoring models Canopy is a pan-African conservation intelligence and finance platform covering all 54 African countries and 200 key protected areas, supporting anyone deploying conservation capital across Africa with an apples-to-apples comparison of country risk, country conservation performance and protected area performance. Three model boxes, each with the question that model answers. PACE for Protected Area Conservation Effectiveness with 200 PAs and 162 KPAs answers: is this PA working? C-RISK for country risk to conservation capital, covering 54 countries, answers: how risky is this country? It is published both as a readiness call and as a country score from 20 to 100, on which higher is safer. C-FIN for conservation finance covering 54 countries answers: is this country funded? It uses three scoring models across parks, countries and finance: PACE Protected Area Conservation Effectiveness 200 PAs, 162 KPAs Is this PA working? C-RISK Country risk for conservation capital 54 countries How risky is this country? C-FIN Conservation finance by country 54 countries Is this country funded?

It uses three scoring models across parks, countries and finance:

  • PACE (Protected Area Conservation Effectiveness, 200 PAs, 162 KPAs)
  • C-RISK (country risk for conservation capital, 54 countries, published as a readiness call and as a country score from 20 to 100, higher is safer)
  • C-FIN (conservation finance, 54 countries)
Canopy: how it works Top box: How does it work? Canopy pairs large language models with structured datasets and direct satellite observation to drive its quantitative indices. Every entry runs through a rigorous pipeline: drafted or computed, challenged by a second critical pass, then versioned and snapshotted so the full history is auditable. Three arrows point down to three pipelines. The left box (LLM Pipeline) describes the reading work for every PA, country, and carbon project: operator reports, academic papers, journalism, NGO investigations, and disclosures, scored on a five-rung rubric. The right box (DATA Pipeline) describes the hard data: credit ratings, bond markets, conflict data, protected area mapping, and the Verra carbon registry. The wide bottom box (SATELLITE Pipeline) shows eleven Earth-observation layers per protected area: fire (NASA FIRMS active-fire thermal), forest (DIST-ALERT on Sentinel-2 and Landsat), vegetation (MODIS NDVI greenness anomalies), surface water (Sentinel-2 NDWI open-water extent), night-time lights (VIIRS DNB monthly), burned area (MODIS MCD64A1), land use inside the boundary and in the buffer (Dynamic World with Sentinel-2 spectral unmixing), biomass (ESA CCI above-ground carbon), rainfall (CHIRPS week-of-year anomaly), soil moisture (SMAP L4 root-zone, fused with rainfall as a drought cross-check), and forest structure (GEDI canopy height and cover). They range from near-daily anomaly sweeps through monthly monitors to multi-year land-use and carbon analysis. How does it work? Canopy pairs large language models with structured datasets and direct satellite observation. Every entry runs through a rigorous pipeline: drafted or computed, challenged by a second critical pass, then versioned and snapshotted so the full history is auditable. LLM Pipeline For every PA, country, and carbon project, the model draws on operator reports, academic papers, independent journalism, NGO investigations, and official disclosures. It scores each dimension on a five-rung rubric and produces a composite. DATA Pipeline Canopy pulls in hard data. Credit ratings, bond market data, conflict data, global protected area mapping, plus the Verra carbon registry. SATELLITE Pipeline FIRE NASA FIRMS active-fire thermal FOREST Sentinel-2 + Landsat DIST-ALERT forest loss VEGETATION MODIS NDVI greenness anomalies WATER Sentinel-2 NDWI open-water extent NIGHT-TIME LIGHTS VIIRS DNB monthly settlement + activity BURNED AREA MODIS MCD64A1 monthly burned footprint LAND USE Dynamic World + Sentinel-2 interior + buffer, unmixed BIOMASS ESA CCI above-ground carbon RAINFALL CHIRPS week-of-year anomaly SOIL MOISTURE SMAP L4 root-zone drought cross-check FOREST STRUCTURE GEDI canopy height + cover, per park 1 2 3 4 5 6 7 8 9 10 11 Eleven Earth-observation layers per park, from near-daily anomaly sweeps through monthly monitors to multi-year land-use and carbon analysis.

How it works

Canopy pairs large language models with structured datasets and direct satellite observation. Every entry runs through a rigorous pipeline: drafted or computed, challenged by a second critical pass, then versioned and snapshotted so the full history is auditable.

For every PA, country, and carbon project, the model reads operator reports, academic papers, independent journalism, NGO investigations, and official disclosures. It scores each dimension on a five-rung rubric and produces a composite.

Alongside this reading work, Canopy pulls in hard data. Credit ratings, bond market data, conflict data, global protected area mapping, plus the Verra carbon registry.

And Canopy watches from orbit. Eleven Earth-observation layers per protected area: fire (NASA FIRMS active-fire thermal), forest loss (DIST-ALERT on Sentinel-2 and Landsat), vegetation (MODIS NDVI greenness anomalies), surface water (Sentinel-2 NDWI open-water extent), night-time lights (VIIRS DNB monthly), burned area (MODIS MCD64A1), land use inside the boundary and in the buffer (Dynamic World with Sentinel-2 spectral unmixing), biomass (ESA CCI above-ground carbon), rainfall (CHIRPS week-of-year anomaly), soil moisture (SMAP L4 root-zone, fused with rainfall as a drought cross-check), and forest structure (GEDI canopy height and cover). They range from near-daily anomaly sweeps through monthly monitors to multi-year land-use and carbon analysis.

The satellite stack is being extended. An above-ground biomass trend now runs alongside the live feeds, and two analytical layers, a fire-risk forecast and few-shot ground-feature detection, are in development, each entering as context first and held to a precision bar set in advance. The method, its acceptance bars, and its known limits, including a measured audit of the fire feed's geometry, are documented in full.