Demonstrating How FPCA Can Leverage SAR Time-Series Information to Distinguish Wetlands and Uplands Based on Seasonal Backscatter Trends

Wetlands are important but vulnerable ecosystems that must be accurately mapped and monitored to effectively guide restoration and conservation planning. In this study, we used functional principal component analysis (FPCA) to leverage synthetic aperture radar (SAR) time-series information and explore whether common wetlands and uplands can be separated based on seasonal trends in backscatter intensity. To contextualize the results, we first identify the drivers of change and analyze variations in seasonal backscatter intensity trends using four years of $C$-band Sentinel-1 VV and VH data. We then trained an FPCA-based feature extraction engine to, first, evaluate the potential of FPCA to improve SAR time series handling and interpretation, second, evaluate the spatio-temporal consistency and separability of derived scores, and third, investigate whether adaptive training can improve the predictive power of FPCA scores. The results showed that microwave-surface interactions vary seasonally between classes. This was primarily due to changes in phenology and hydrology, whose effects on backscatter varied depending on target characteristics such as plant functional type. On the other hand, scores were relatively consistent within each specified class, though shifted according to some significant changes in target characteristics. Classification of scores using random forest demonstrated that the method was effective in generating discriminant features. Independent overall accuracies ranged from 83% to 89% even when the model was applied to unseen data and in spite of inherent difficulties distinguishing between swamp and forest. Retraining and reapplying FPCA to better capture the variation specific to these classes also demonstrated that the predictive power of scores remained constrained by the inherent limitations of $C$-band VV and VH polarized data for detecting surface water in forested areas. Overall, these findings highlight the potential of FPCA to improve the handling and interpretation of SAR time series, and that seasonal backscatter intensity trends, captured by FPCA scores, can effectively separate multiple common wetlands and uplands.