Dry-wet seasonality effects on the satellite-based land cover types identification in the Nile River Basin

作者：Yulong Lv, Dailiang Peng, Zihang Lou, Hongyan Wang, L S Yu, Yaqiong Zhang, Qiaoyun Xie, Jinkang Hu, Shijun Zheng, Enhui Cheng, Hongchi Zhang, Yizhou Zhang, Hao Peng · 发表于：Big Earth Data · 年份：2025 · DOI：10.1080/20964471.2025.2603812 · 被引用次数：1 · 研究领域：Remote Sensing in Agriculture、Flood Risk Assessment and Management、Remote-Sensing Image Classification

Accurate land cover type (LCT) mapping is essential for monitoring ecological dynamics and supporting sustainable resource management. However, in the Nile River Basin (NRB), characterized by pronounced seasonal dry-wet variations, previous mapping efforts showed limited LCT identification accuracy due to insufficient consideration of these dynamics. To assess the impacts of seasonal variations on LCT mapping, we computed monthly Standardized Precipitation Evapotranspiration Index (SPEI) across the NRB to capture spatiotemporal dry-wet transitions. Three Random Forest (RF) models were trained using dry-season, wet-season and combined spectral characteristics, with accuracy metrics and pixel-scale consistency analysis. The results indicate that: (1) The NRB shows clear spatiotemporal alternation between dry and wet seasons. (2) Identification using dry-season spectral characteristics achieved higher classification accuracy (OA = 83.68%, Kappa = 0.8134) than wet-season data (OA = 80.39%, Kappa = 0.7758), with 35.6% of areas showing seasonal discrepancies. The combination of dry and wet season spectral datasets further improved accuracy (OA = 88.14%, Kappa = 0.8644). (3) Discrepancies in identification primarily stem from variations in spectral responses between dry and wet seasons. This study demonstrates that integrating dry and wet season spectral characteristics can substantially improve land cover classification accuracy, particularly in regions with pronounced wet-dry seas...