deFOREST: Fusing Optical and Radar satellite data for Enhanced Sensing of Tree-loss

A new deforestation detection pipeline fusing optical and Synthetic Aperture Radar satellite data has been developed and tested on a roughly 92 km × 92 km region of the Amazon forest, according to research published on arXiv [1]. The method, dubbed deFOREST, constructs anomaly maps from optical data using the residual space of a discrete Karhunen-Loéve expansion. Anomalies are quantified through a concentration bound on the distribution of residual components for the forest's nominal state, an approach that does not require prior knowledge of the data distribution [1]. This contrasts with parametric statistical methods that assume a known distribution, an assumption the authors describe as impractical for high-dimensional satellite data [1]. Once the optical anomaly maps are computed, they are combined with Synthetic Aperture Radar data, and a Hidden Markov Model classifies the forest's state [1]. The pipeline was evaluated using Sentinel-1 SAR and Sentinel-2 optical imagery [1]. The results indicate that both the hybrid optical-radar method and the optical-only method achieve accuracy superior to the most recent state-of-the-art hybrid approach [1]. The hybrid method also proved more robust when optical data were sparse, a common condition in heavily cloud-covered regions [1]. The research team included Julio Enrique Castrillon-Candas, Hanfeng Gu, Caleb Meredith, Yulin Li, Xiaojing Tang, Pontus Olofsson, and Mark Kon [2]. The paper was first submitted in October 2025 and revised in June 2026 [1]. The use of SAR data to overcome the limitations of optical sensors in cloudy environments is a recurring theme in remote sensing research. A separate study on forest and non-forest segmentation noted that Sentinel-1 radar acquisitions are unaffected by atmospheric conditions, making them critical for monitoring areas frequently obscured by clouds [4]. That work demonstrated that while optical data enable faster and easier segmentation in cloud-free contexts, radar data definitively reveal patterns such as urban expansion and deforestation when optical images are partially occluded [4]. Another research effort introduced a curated dataset and deep learning approach for deforestation estimation and fire detection in the Amazon, using imagery from Sentinel, Landsat, VIIRS, and MODIS satellites [3]. That team found that missing satellite data and frequent cloud occlusion made deforestation segmentation challenging, and that combining Sentinel-1, Sentinel-2, and Landsat-8 data allowed neural networks to detect deforestation with accurate results [3]. The deFOREST pipeline builds on this principle by formally integrating optical anomaly detection with SAR data through a probabilistic temporal model [1].