Near real-time monitoring of tropical forest disturbance by fusion of Landsat, Sentinel-2, and Sentinel-1 data

Background

Tropical forest disturbance contributes significantly to global carbon emissions and biodiversity loss, particularly in regions such as the Amazon Basin, where deforestation and degradation continue to rise. Effective monitoring is critical for climate mitigation initiatives such as REDD+ and for enabling timely enforcement of environmental regulations. While satellite remote sensing provides the only feasible means of monitoring large areas, individual sensor systems present limitations. Optical sensors such as Landsat and Sentinel-2 suffer from cloud cover constraints, whereas radar systems like Sentinel-1 capture structural changes but respond more slowly to disturbance signals. Increasing the temporal density and reliability of observations remains a key challenge for near-real-time monitoring.

Goals and Methods

This study develops the Fusion Near Real-Time (FNRT) algorithm to improve disturbance detection by integrating Landsat, Sentinel-2, and Sentinel-1 data. This model fits harmonic time-series functions using three years of training data (2017–2019) and applies them to detect anomalies during the monitoring year (2020). Change scores are derived from residuals between predicted and observed values and classified into disturbance signals using threshold-based rules. A moving monitoring window evaluates consecutive signals to confirm disturbances. The algorithm is implemented on Google Earth Engine and tested across eight Amazon sites using stratified sampling and lag-time-based accuracy assessment.

Conclusions and Takeaways

This study demonstrates that multi-sensor fusion significantly improves both detection speed and accuracy. The FNRT model detects approximately 69.8 percent of disturbances within 30 days and reaches a peak producer’s accuracy of 91.6 percent. Combining sensors consistently outperforms single-sensor approaches across all lag times. Optical sensors detect early-stage disturbances more rapidly, while radar enhances detection under cloud cover. Practitioners can use FNRT to support near real-time monitoring, enforcement, and climate reporting systems. Future improvements should focus on increasing data density, refining forest masks, and integrating additional high-resolution datasets.