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

Deforestation and forest degradation are large sources of carbon emissions and negatively impact biodiversity, food security, and human well-being. The ability to quickly and accurately detect forest disturbance events is essential for preventing future deforestation and degradation and mitigating their negative effects. Combining optical and radar data has the potential to achieve faster detection of forest disturbance than using an individual system. The main challenge is the methodological approach for fusing the different datasets and scaling it for operational monitoring over very large areas. Here we developed a near-real-time monitoring system for tropical forest disturbance by fusion of Landsat, Sentinel-1, and Sentinel-2 data. For each individual sensor system, the algorithm fits a harmonic model over data for a three-year training period prior to the desired monitoring period. The algorithm then produces a change score for each observation during the monitoring period based on model residuals and RMSE. Change detection is then performed on integrated time series of change scores across all three sensors. A stratified random sample is selected and an assessment framework introduced by Bullock et al (2021) is adopted for accuracy assessment of the results. This study aims to understand the effectiveness of data from each or combinations of the three sensors in the context of near-real-time monitoring of tropical forest disturbance.