Forest Carbon Diligence Data Release Version 1.3

What's New

This version marks a major improvement and sets a new benchmark in quality for the market, and releases an automated, annual change detection asset. We will publish these results in a peer-reviews and advance scientific engagement with the community.

Improvements

Carbon model

15% increase in pixel-level variance explained. 30m r2 scores increased from 0.51 to 0.65. 1km r2 scores increased to 0.82.
35% average reduction in pixel-level uncertainty.
Systematic reduction in bias (addressing overestimation issues).
3x increase in training data, adding coverage in boreal forests.

Time series model

Higher year-to-year consistency, lower noise.
Higher confidence in reported changes (compared to benchmark).

New asset

New change detection asset for automated forest change analysis, available as "cc-change" and "ch-change" in the Subscriptions API.
These are 30m uint8 categorical rasters, with values v0, 1, 2] corresponding to nno change, fast change, slow change].

Product documentation

Web documentation for tech specs and intercomparison/validation report.
Pre-print manuscript for Diligence methods, validation, and intercomparison with independent data.
Pre-print manuscript for change detection intercomparison.