Interconnections, trend analysis and forecasting of water-air temperature with water level dynamics in Blue Moon Lake Valley: A statistical and machine learning approach.

Glacier-fed lakes serve as vital indicators of climate change, yet their temperature and water level dynamics are insufficiently studied, particularly in high-altitude basins. Examining these interactions is fundamental for the effective management of water resources in sensitive environments. This study investigates the interactions between air temperature (Ta), water temperature (Tw), and water level (WL) in the Blue Moon Lake Valley (BMLV), using advanced statistical and machine-learning techniques to address gaps in predicting these complex dynamics. We employed the Mann-Kendall test and statistical tests (F-test, t-test) to detect trends in Ta, Tw, and WL. Granger causality analysis explored directional relationships, while wavelet analysis captured variations across multiple timescales. Extreme value analysis assessed the influence of temperature extremes on WL. We compared the performance of machine learning models (GBM, XGB, DT, RF) and proposed a hybrid Quad-Meta (QM) model, which combines the strengths of these approaches, leading to improved prediction accuracy. Results indicated significant warming trends for Ta and Tw, with Ta increasing more rapidly, while WL exhibited stability. This indicates that hydrological factors may play a central role in moderating water levels. In terms of prediction, the QM model demonstrated superior performance and achieved the lowest RMSE (0.117 °C for Tw, 0.326 °C for Ta, and 0.002 m for WL) and the highest R values. We recommend continued global monitoring of Ta, Tw and WL in glacier-fed lakes and developing a hybrid QM model to enhance prediction accuracy in high altitude sensitive environments.