About
This pilot investigates the functioning and resilience of European forest ecosystems under climate change and disturbance events. It focuses on monitoring forest productivity, health, and structural changes using a combination of ecological modelling and satellite observations. The pilot covers three representative forest regions: Rumperöd in southern Sweden, Berchtesgaden National Park in Germany, and the Carpathian mountain forests in Central and Eastern Europe.
Current progress
Current work has focused on the Rumperöd case study and on developing a workflow for downscaling coarse LPJ-GUESS outputs using remote sensing data as fine-resolution spatial proxies. LPJ-GUESS outputs have been organized for multiple coarse grid cells and key ecosystem variables, including LAI, GPP, NPP, biomass and FPC, with additional forest-structure variables available for later stages.
Remote-sensing predictors, including Sentinel-2-derived FCOVER/LAI proxies, Sentinel-1 SAR data and DEM-derived terrain layers, have been explored as spatial information sources for the downscaling task. Initial machine-learning experiments tested several model families, including convolutional, transformer-based, graph-based, diffusion-based and hybrid architectures. These experiments provided a useful prototype and helped identify the main methodological requirements for a reliable downscaling framework.
The work also revealed important challenges that need to be addressed before operational use. This includes the need to avoid over-emphasizing coarse mean matching, reduce spatial artifacts, prevent irrealistic smoothness, and ensure that downscaled maps preserve both LPJ-GUESS consistency and realistic fine-scale spatial patterns. Based on these findings, the workflow has been improved with a cleaner decoder design, valid-data masks, interior-focused evaluation, spatial fidelity metrics, edge-aware smoothness and spatial cross-validation.
Scope & background
Forests play a crucial role in climate regulation, carbon storage, and biodiversity conservation. However, forest ecosystems are increasingly affected by climate change, extreme weather events, pest outbreaks, and other disturbances. Monitoring these impacts requires models capable of simulating ecosystem processes and predicting how forests respond to environmental stressors. The pilot therefore combines ecological models with Earth Observation data to monitor forest productivity and ecosystem functioning across large spatial scales.
OBSGESSION's contribution
OBSGESSION’s contribution to the development of the pilot includes adapting the LPJ-GUESS dynamic vegetation model to simulate forest productivity while incorporating disturbance modules related to storms and bark beetle outbreaks. These simulations are combined with satellite data from missions such as Sentinel-1, Sentinel-2, and Landsat to improve spatial resolution and monitoring capabilities. Machine learning techniques are used to downscale coarse model outputs into high-resolution predictions of forest productivity and ecosystem structure. This integration allows the generation of spatially detailed indicators such as biomass, leaf area index, and gross primary productivity.
Expected results
Expected outputs include high-resolution maps of forest productivity and ecosystem structure across the pilot regions. The pilot will also produce improved modelling approaches for predicting forest responses to climate change and disturbance events. By combining process-based models with remote sensing data, the pilot will generate spatially detailed indicators of forest ecosystem functioning and provide uncertainty estimates for these predictions.
Stakeholders
Key stakeholders include forest management authorities, environmental agencies responsible for biodiversity monitoring, and policymakers involved in implementing the EU Forest Strategy and biodiversity policies. Researchers studying forest ecology, climate impacts, and ecosystem services are also important users of the pilot’s outputs.
