Estimating forest degradation by capturing temporal change in biomass density: insights from a tropical dry forest

Forest degradation is typified by reductions in canopy cover and biomass density over time, resulting in decreased capacity and quality of forest ecosystem function and services. It usually occurs in small patches across the forest area, so it is tough to detect using only field data, which is typically collected in widely spaced sampling plots. In this paper, we used multitemporal field measurements of aboveground biomass (AGB) and remote sensing to estimate forest degradation in a tropical dry forest that has been subject to shifting cultivation for many years. First, we obtained field measurements in 2019 and 2022, and then we mapped AGB in these years using random forest regression and Sentinel-1 and Sentinel-2 images. Then, we calculated the difference in AGB between these two dates (ΔAGB = AGB 2022 – AGB 2019 ) and reclassified the map of ΔAGB into three categories, namely AGB gain, no change, and AGB loss. We performed a rigorous accuracy assessment by randomly distributing 1257 sampling points across those three categories to obtain an unbiased area estimation. Although the average AGB density of the data collected in the field was not statistically different between these two years, there was a slight reduction in canopy cover and canopy height, along with an increase in stem density, which is consistent with a shifting cultivation regime in which coppicing occurs. The RMSE of the 2019 AGB model was 14.5 ± 4.8 Mg ha -1 and the 2022 model was 14.0 ± 4.52 Mg ha -1 , suggesting that the mapped AGB are comparable between these two years. The accuracy of our estimate for AGB loss exceeded that of AGB gain, and using ground data at two time periods allowed us to achieve a much more accurate estimate than would have been possible with remote sensing alone. AGB loss accounted for 4.3% of the change in biomass over the total forest area, while AGB gain through natural recuperation represented 1.3%, and most of the forest area (94.4%) remained stable. Thus, over the entire study area, a small negative trend in AGB was registered. The study demonstrates the value of integrating time series ground-level data with remote sensing and modeling to obtain much more accurate estimates of biomass loss and gain in forests where degradation and regrowth are the dominant processes, rather than deforestation. • GLCM textures of both Sentinel-1 and Sentinel-2 bands were among the most important predictors for AGB in the tested biomass models. • The RMSE of the biomass models ranged between 14 and 14.5 Mg/ha. • Approximately 4% of the tropical dry forest in the study area lost AGB. • The area with AGB loss was approximately three times of area with AGB gain.