Process-informed deep learning for interpretable forested wetland mapping under dense canopy

Canopy-obscured hydrologic signals and gradual wetland–upland transitions hinder the delineation of forested wetlands, particularly when heterogeneous Earth-observation evidence is fused as an unstructured feature stack. We developed a process-informed multi-stream convolutional neural network (CNN) for mapping forested wetlands on a 10 m grid in the Changbai Mountains, China. Eighty-five covariates from Sentinel-1/2, ALOS-2 PALSAR-2, and Copernicus DEM GLO-30 were organized into vegetation, hydrology, phenology, topography, and temporal-context streams before spatial fusion. Relative to a capacity-matched single-stream baseline, the grouped architecture increased forested-wetland F1 by 3.42 percentage points across five validation folds, whereas the fixed grouped held-out test set showed a smaller margin of 1.03 percentage points. For map production, five-fold probability averaging achieved an overall accuracy of 0.873, macro-F1 of 0.882, and forested-wetland F1 of 0.836. To interpret correlated predictors, group-aggregated Gradient SHapley Additive exPlanations (SHAP) and k-nearest-neighbor conditional replacement distinguished global model reliance from conditionally non-substitutable evidence. Among correctly classified forested-wetland samples, context and phenology were the leading sources of global reliance, whereas hydrology, followed by topography, provided the strongest conditionally unique evidence. Direct pixel counting yielded a map-derived forested-wetland area of 686.6 km², concentrated mainly in peripheral plains and hills. This framework combines process-informed feature organization with dependence-aware interpretation to support transparent regional mapping of canopy-obscured forested wetlands.