Deep Evidential Regression for Sparse Forest Height Estimation from Multimodal Satellite Imagery

2026-08-03

AI Summary

This study addresses the challenge of unreliable uncertainty quantification in forest height estimation from remote sensing, which is often hindered by sparse reference labels and geographic distribution shifts. To this end, we propose a deep evidential regression U-Net model leveraging multimodal Sentinel-1 and Sentinel-2 imagery, equipped with a novel masked evidential loss function. Our approach simultaneously produces high-accuracy canopy height predictions and well-calibrated uncertainty estimates in a single forward pass. Evaluated on the TreeUQ benchmark, the method achieves prediction accuracy comparable to deterministic U-Net baselines while generating spatially continuous and reliable uncertainty maps. This capability significantly enhances trustworthy remote sensing mapping in scenarios where ground truth labels are extremely scarce.

Abstract

Accurate estimation of forest height from satellite imagery is essential for applications such as carbon accounting, biodiversity monitoring, and ecosystem management. While recent deep learning approaches provide accurate predictions, they typically do not quantify predictive uncertainty. This limitation is particularly relevant in geospatial settings characterized by sparse supervision and geographic distribution shift. In this work, we investigate Deep Evidential Regression (DER) for forest height estimation on the TreeUQ benchmark, a large-scale dataset designed for the joint estimation of tree count and average tree height at 10 m resolution, based on Sentinel-1/-2 data as well as tree inventory data over the federal state of Bavaria. To account for the extreme label sparsity of the tree inventory data, we introduce a masked evidential loss for dense geospatial prediction. Using a U-Net architecture with multimodal Sentinel-1 and Sentinel-2 inputs, the proposed approach jointly predicts tree height and associated uncertainty estimates in a single forward pass. Experimental results show that DER achieves predictive performance comparable to a deterministic U-Net while additionally providing well-calibrated uncertainty estimates. These findings demonstrate the potential of evidential learning as an efficient framework for uncertainty-aware forest structure estimation from Earth observation data.

Problem

• forest height estimation
• predictive uncertainty
• sparse supervision
• geospatial prediction
• multimodal satellite imagery

Innovation

• Deep Evidential Regression
• Uncertainty Quantification
• Multimodal Satellite Imagery
• Masked Evidential Loss
• Forest Height Estimation