Climate and hydrological analysis of glacial drainage basin based on BP neural network restoration of monitoring data - a case study of Weigeledangxiong Glacier

The drainage basin of the melting water of the Weigeledangxiong Glacier. Under global climate change, increasing meltwater from Weigeledangxiong Glacier has altered watershed composition in its drainage basin. Continuous in-situ monitoring is crucial for identifying sensitive hydrological processes, yet harsh high-altitude conditions and technological limitations restrict data quality and hinder studies of melt-driven short-term water cycle mechanisms. To address this, we developed a BP neural network-based reconstruction method for generating accurate, continuous, high-temporal-resolution datasets. Using complete monitoring records, we applied Mann-Kendall (M-K) trend analysis, M-K mutation testing, and Spearman correlation to assess meteorological and hydrological variations throughout the hydrological year in the glacial drainage basin. (1) The BP neural network exhibits superior applicability and accuracy (average accuracy: 94.82 %) compared to traditional interpolation methods for restoring short-term continuous monitoring data. (2) June 28th marks a critical threshold when the study area's temperature transitioned from persistent decline to significant increase, leading to substantially enhanced DXG melting intensity under temperature control beginning in July. (3) The total annual meltwater runoff from DXG reaches 20,422,030 m³ , with July accounting for 38.18 % of the annual total, reflecting pronounced seasonal concentration in local hydrology. Additionally, river freezing and melting processes significantly influence groundwater levels and temperatures within the watershed, with observed hydrological responses exhibiting an approximately 8-day lag period. • Hourly meteorological, hydrological and groundwater data obtained from glacier basin. • BP neural network effectively reconstruct continuous hourly multi-parameter datasets. • Neural network training reveals common patterns across diverse parameter types. • Continuous datasets allows for robust intra-annual trend characterization.