Project aim

Forecast how forest above-ground biomass, and therefore carbon, evolves over the next decade under different climate scenarios, to support REDD+ planning.

  • 10 yrForecast horizon
  • 5Data sources fused
  • RCPClimate scenarios
  • GNNModel family
Illustrative map for SMART-FOREST Illustrative graphic

Overview

SMART-FOREST fuses Sentinel-1 and Sentinel-2 imagery, GEDI spaceborne LiDAR, SMAP soil moisture and ERA5 climate data in an object-based and temporal graph neural network framework. It forecasts above-ground biomass over 10-year horizons under RCP climate scenarios, evaluated with spatial-block cross-validation.

The challenge

GEDI measures canopy structure only along sparse tracks, while imagery covers everything but does not measure biomass directly. Forecasting a decade ahead also means accounting for climate pathways, not just extrapolating past trends.

Data & inputs

  • Sentinel-1/2
  • GEDI LiDAR
  • SMAP soil moisture
  • ERA5
  • RCP scenarios

Approach

  1. 01

    Multi-source fusion

    Sentinel-1/2, GEDI LiDAR, SMAP soil moisture and ERA5 are brought onto a common analysis grid.

  2. 02

    Object-based graph

    Forest objects become graph nodes, so neighbouring stands inform each other.

  3. 03

    Temporal GNN

    A temporal graph neural network models biomass change over time.

  4. 04

    Scenario forecasting

    Forecasts run over 10-year horizons under RCP scenarios, evaluated with spatial-block cross-validation.

Results

The fusion framework and forecasting pipeline are under active development.

Validation

Spatial-block cross-validation.

Status & caveats

This is an ongoing project. Results will be published here as validation is completed.