Extent and dynamics of tropical flooded forests with multi-year SAR composites and altimetry

Flood dynamics regulate greenhouse gas emissions from tropical flooded forests, but monitoring flooding beneath dense canopies is challenging. We introduce a L-band SAR approach that uses more than a decade of ALOS-2 PALSAR-2 observations to detect sub-canopy flooding at 100 m resolution across the Amazon and Congo rainforests. We generate multi-year composites of HH and HV backscatter that represent low-, intermediate-, and high-water stages. These hydrological stages are defined locally using over 7000 satellite altimetry time series, accounting for the large variability in the timings of high-water conditions across both rainforests. Flooded forests display a characteristic dual-polarization signature: HV backscatter is comparable to that of non-flooded forests, while HH backscatter is higher due to double-bounce scattering. Quantifying deviations in the HH-HV space, we determine the extent of flooded forests at high-water stage and estimate their seasonal dynamics using a discrete stage-count metric, informing on the number of hydrological stages at which a pixel is classified as flooded forest. Multi-year compositing reduces speckle noise relative to single scenes, enabling the detection of narrow floodplains, while the local definition of high-water conditions improves the representation of high-water conditions across the two rainforests. We estimate approximately 827,000 km 2 of flooded forests under typical high-water conditions, ~687,000 km 2 in the Amazon and ~ 140,000 km 2 in the Congo rainforest. In the Amazon basin, agreement with estimates from Hess et al. (2015), Rosenqvist et al. (2020) and Fleischmann et al. (2022) varies spatially, with high agreement in wide floodplains (e.g., Solimões and Purus rivers), lower agreement along narrower floodplains, and substantial discrepancies in sparsely vegetated areas, partly explained by differences in forest definitions. In the Cuvette Centrale, our estimates show spatial consistency with the estimated extent of peat swamp forests by Crezee et al. (2022), although our flooded forest extent is more spatially restricted. We also found a good correlation between our stage-count map and the microtopography of the Solimões floodplain. By resolving both the extent and seasonal dynamics of flooded forests at 100 m resolution, these maps can support ecological, hydrological, and carbon studies in tropical forests.