Seasonal floodplain wetlands, as dynamic transitional ecosystems between terrestrial and aquatic environments, play a pivotal role in maintaining regional ecological balance and regulating hydrological processes. However, their spatiotemporally variable inundation regimes and heterogeneous land cover pose significant challenges for remote sensing-based water body extraction. Notably, the intricate coupling of hydrological dynamics and spectral responses constrains the applicability of single spectral indices, leading to inconsistent accuracy and limited monitoring efficacy. Focusing on the Poyang Lake floodplain as a representative case, this study aims to investigate the adaptability of multiple spectral indices to hydrological fluctuations and land cover complexity in seasonal floodplain wetlands, with the primary objective of developing optimized, context-specific extraction strategies tailored to distinct hydrological scenarios and surface conditions.The research was conducted in the Poyang Lake National Nature Reserve, a typical seasonal floodplain wetland in China. Sentinel-2 MSI imagery was selected as the core data source, leveraging its high spatial and spectral resolution. Eight representative spectral indices were constructed, including normalized indices (NDWI, MNDWI), multi-band indices (MBWI, SWI), and auxiliary indices (NDVI, TCB, TCG, TCW). Water body extraction was performed under both wet (July 2021) and dry (March 2021) hydrological conditions using K-means clustering on the Google Earth Engine (GEE) platform. Ground truth data were derived from high-resolution GF-1/2 PMS imagery, with random forest classification employed for validation. Extraction accuracy was evaluated using multiple metrics: Overall Accuracy (OA), Producer’s Accuracy (PA), User’s Accuracy (UA), and Kappa coefficient. Additionally, stratified assessments were conducted for distinct water body types, encompassing permanent waters, seasonal dish-shaped lakes, and paddy fields.Results indicate that hydrological variation exerts a significant regulatory effect on the performance of spectral indices. During the wet season, NDWI achieved the highest extraction accuracy (OA = 95.4%), whereas TCB performed poorest (OA = 73.5%) due to interference from high-moisture vegetation. In the dry season, SWI (OA = 97.4%) and NDWI (OA = 97.2%) demonstrated strong discriminative capacity in mixed spectral environments, while TCW exhibited high spectral confusion with mudflats (OA=75.7%). For different water body types, SWI yielded the best results for seasonal dish-shaped lakes (OA=98.5%); NDWI and TCB performed optimally for permanent water bodies (OA=99%); and only MBWI maintained high accuracy for paddy fields (OA=94%). Boxplot analyses and spectral curve comparisons confirmed that the adaptability of spectral indices is closely linked to their ability to enhance intra-class spectral homogeneity and inter-class separability.This study confirms that no single spectral index can effectively capture water bodies across all seasonal and land cover conditions in floodplain wetlands. The efficacy of each index is strongly modulated by hydrological stages and the spectral characteristics of specific surface types. Thus, adopting a differentiated extraction strategy based on hydrological context and target land cover types is imperative. This study identifies the optimal matching patterns of spectral indices under varying hydrological conditions and water body types. It provides a scientific and technical framework for dynamic monitoring of water bodies in seasonal floodplain wetlands and holds significant practical implications for enhancing wetland ecosystem conservation and refined water resource management.