[A wetland identification method based on hydrological response unit connectivity.]

Abstract

Wetland identification is a fundamental prerequisite for the conservation, ecological restoration, and basin-scale water resources management of wetlands. Conventional remote-sensing-based wetland identification mainly relies on surface spectral information and is susceptible to intra-class spectral variability, inter-class spectral similarity, ambiguous boundaries, and seasonal hydrological fluctuations. It is therefore difficult to reveal the hydrological processes underlying wetland formation and maintenance. To address these limitations, we proposed a refined wetland identification method at the hydrological response unit (HRU) scale oriented toward hydrological connectivity. With Weihe River Basin as a validation area, we examined the applicability of this method in a complex watershed environment. Based on daily HRU-scale hydrological fluxes simulated by the SWAT+ model, we calculated nine dimensionless hydrological connectivity indicators and integrates particle swarm optimization with support vector machine to adaptively determine the classification boundaries of river-fed wetlands, temporary water-supply wetlands, groundwater-fed wetlands, and non-wetlands in a high-dimensional feature space. The identification performance was evaluated through reference validation, typical-area image comparison, and hydrological physical consistency assessment. SHAP analysis showed that the evapotranspiration ratio, detention storage coefficient, channel runoff proportion, and lateral flow ratio were the major contributing features in the identification process. Validation results in the Weihe River Basin showed that the receiver operating characteristic curve values (ROC-AUC) of the three wetland types were all close to 0.90, while the overall accuracy of binary wetland-non-wetland identification was approximately 77.8%. Spatially, wetlands were mainly distributed in a network pattern along the main stem of the Weihe River and its tributaries in Shaanxi Province. During 2018-2022, the average wetland proportion was 3.2% in summer and 0.9% in winter. River-fed wetlands accounted for a multi-year average proportion of 72.0%, making them the dominant wetland type, whereas temporary water-supply wetlands and groundwater-fed wetlands occupied relatively smaller proportions. The results indicated that the proposed method could effectively characterize the spatial extent, type structure, and dynamic changes of wetlands at the basin scale. It would facilitate a shift in wetland identification from "spectral discrimination" to "process diagnosis" and thus provide a new technical pathway for the conservation, ecological restoration, and water resources management of wetlands.