DETER-R: An Operational Near-Real Time Tropical Forest Disturbance Warning System Based on Sentinel-1 Time Series Analysis

Background

Near real-time forest disturbance monitoring is essential for reducing tropical deforestation and supporting enforcement actions in highly threatened regions such as the Brazilian Amazon. Most operational monitoring systems rely on optical satellite imagery, which is strongly constrained by persistent cloud cover in humid tropical environments. Synthetic Aperture Radar (SAR) imagery offers an alternative because it can penetrate clouds and provide continuous observations. In response to this limitation, Brazil’s National Institute for Space Research (INPE) developed DETER-R, a radar-based operational warning system designed to complement optical deforestation monitoring systems and improve detection capacity during cloudy periods.

Goals and Methods

This study presents the development and first operational year of the Deforestation Detection System (DETER-R), an automated near-real-time forest deforestation detection system based on Sentinel-1 SAR time-series analysis. This system processes Sentinel-1 imagery daily in Google Earth Engine and applies adaptive linear thresholding to detect disturbance events. DETER-R integrates information such as historical deforestation proximity, forest masks, and temporal filtering to reduce false positives. Detected anomalies are made into warning polygons and classified into disturbance intensity categories. Validation combines automated comparison with optical DETER alerts and visual interpretation using Planet, Sentinel-2, and Landsat imagery.

Conclusions and Takeaways

The first year of operation demonstrated that DETER-R can effectively support forest law enforcement under persistent cloud conditions. The system produced more than 88,000 disturbance warnings with an extremely low false-positive rate, while also detecting over 105,000 ha of disturbances not captured by the optical DETER system. DETER-R proved particularly valuable during the rainy season, when optical monitoring performance declines substantially. The study highlights the operational value of SAR-based systems for tropical forest governance and suggests that future improvements could incorporate machine learning and deep learning approaches to reduce omission rates and improve detection speed. The framework also demonstrates strong potential for adaptation to other tropical forest regions worldwide.