Book

Description

Tropical lake-basin wetlands rarely have a clean edge. Papyrus hides water from optical satellites, seasonal floods redraw the shoreline, and radar and optical sensors often disagree about what lies beneath. Conventional maps force these shifting landscapes into fixed categories and say nothing about how reliable their boundaries are. This book presents an uncertainty-aware approach to wetland mapping in the Kenyan sector of the Lake Victoria Basin. It combines OLMoEarth geospatial foundation model embeddings with Sentinel-1 radar, Sentinel-2 multispectral imagery and a physically informed two-stage Random Forest classifier to separate wetlands, open water and land across wet and dry seasons. The framework achieves about 90.8 percent overall accuracy in both seasons and introduces the Wetland Uncertainty Field, a map that shows where classifications can be trusted and where they cannot. A seasonal dynamics product separates 33,274 hectares of permanent wetland from 2,904 hectares of seasonal wetland. Written for researchers, remote sensing and GIS practitioners, and planners working on wetland management, flood risk and climate adaptation in data-limited regions.