Coupling high-resolution field monitoring and MODIS for reconstructing wetland historical hydroperiod at a high temporal frequency

Historical wetland hydrology data are instrumental to support the design of wetland management and restoration strategies but are rarely available. In this study, we tested the capabilities and limitations of a simple methodological framework based on publicly available MODIS Land Reflectance Products to estimate wetland soil surface saturation and inundation spatiotemporal dynamics. Using supervised learning and high-resolution groundwater and surface water elevation data, the framework searches for spectral algorithms, referred to as the wet/dry wetland status classifier (WSC) and the continuous wetland dynamics identifier (WDI), that best predict upper soil layer wetness status in the study wetland. We used Google Earth Engine (GEE) for fast access and processing of the full range of MODIS data. The capabilities of GEE also enabled us to readily conduct a comparative assessment of the MODIS 8-day composite and daily collections and test various pixel-level quality filters to select reliable data at the highest possible temporal resolution. We tested the framework on the internationally-recognized Ramsar site Palo Verde National Park wetland in Costa Rica, and we obtained good results (overall prediction accuracy of 86.6% and kappa coefficient of 0.7 for the WSC; r2 of 0.71 for the WDI). High-resolution water level data allowed us to assess the challenges, promises and limitations of using MODIS products for wetland hydrology applications. We then applied the WSC and WDI to map the 2000-2016 subweekly wetland hydroperiod at 500m resolution, achieving a temporal resolution rarely matched in remote sensing for wetland studies. The analysis of the end-products, combined with the field water elevation data, provided new insights into the drivers controlling the spatiotemporal dynamics of hydroperiod within the Palo Verde wetland and did not reveal any significant temporal trends. The WSC and WDI framework developed here can be useful for reconstructing long-term hydroperiod variability and uncovering its drivers for other wetland systems globally