Agricultural Parcel Delineation for the Senegal River Valley: A UNet Segmentation Based High-Resolution Approach

Accurate delineation of agricultural parcels is critical for enabling sustainable intensification and supporting effective land use and food security management, while also facilitating access to credit, insurance, and public services. However, data scarcity remains a major constraint across the African Sahel, including the Senegal River Valley (SRV), where consistent, high-resolution field parcel information is largely unavailable. To address this gap, we propose a boundary-aware deep learning framework based on HED-UNet that integrates structured post-processing and object-level evaluation to generate spatially coherent and analytically robust field objects. The framework is applied to freely available Norway’s International Climate and Forest Initiative PlanetScope mosaics (2017–2025), enabling the derivation of a consistent representation of maximum observed field extent for regional-scale agricultural monitoring. Across all evaluation tiles, the model achieves a global IoU of 0.73 and an F1-score of 0.84, with a precision of 0.75 and a recall of 0.97. Lower boundary scores (0.39–0.59) are primarily driven by spatial resolution constraints, the spectral complexity of irrigation–wetland interfaces, and the fragmented smallholder landscape characterized by narrow and heterogeneous field structures, which challenge precise boundary localization. Overall, this study provides the first wall-to-wall agricultural parcel dataset at 4.77 m resolution for the SRV derived entirely from freely available satellite data, offering a scalable foundation for field-level agricultural monitoring in smallholder systems.