The changing climate in recent decades has caused rapid, widespread frozen ground degradation of the cryosphere, and changes in frozen ground may largely affect the ecosystem function. This study presents the long-term shrinkage of frozen ground, driven forces and resultant ecological impacts at three high-latitude sites (namely Nenjiang, Jiagedaqi, and Tahe, from low to high latitude) in Northeast China, an area that has received less attention in previous permafrost studies. As the key indicators of frozen ground, the ground surface and subsurface soil temperature (down to 40 cm) were found to increase in the period from the 1970s to the 2000s. The annual mean air temperatures at observation sites have increased significantly at rates of 0.24~0.40 °C/decade from 1971 to 2019, and were found to be negatively correlated to the change in soil freeze depth at the Jiagedaqi and Tahe sites. The annual snow depth at Jiagedaqi was found to decrease at a rate of 1.8 cm/decade, and positively correlate to the soil freezing index. The vegetation status of the three watersheds improved, as indicated by the increased normalized difference vegetation index (NDVI) from 2000 to 2017. Our results can make a valuable contribution to knowledge of climate–frozen ground interactions, and can highlight the importance of examining such interaction for permafrost protection in similar high-latitude permafrost-dominated basins.

Despite advances in hydrological Deep Learning (DL) models using Single Task Learning (STL), the intricate relationships among multiple hydrological components and model inputs might not be comprehensively encapsulated. This study employed a Long Short-Term Memory (LSTM) neural network and the CAMELS dataset to develop a Multi-Task Learning (MTL) model, predicting streamflow and evapotranspiration across multiple basins. An optimal multi-task loss weight ratio was determined manually during the validation phase for all 591 selected basins with streamflow data-gaps under 5%. During test period, MTL showed median Nash-Sutcliffe Efficiency predictions for streamflow and evapotranspiration at 0.69 and 0.92, consistent with two STL models. The MTL's strength appeared when predicting the non-target variable, surface soil moisture, using probes derived from LSTM cell states—representative of the internal DL model workings. This prediction showed a median correlation coefficient of 0.90, surpassing the 0.88 and 0.89 achieved by the streamflow and evapotranspiration STL models, respectively. This outcome suggests that MTL models could reveal additional rules aligned with hydrological processes through the inherent correlations among multiple hydrological variables, thereby enhancing their reliability. We termed this as "variable synergy," where MTL can simultaneously predict varied targets with comparable STL performance, augmented by its robust internal representation. Harnessing this, MTL promises enhanced predictions for high-cost observational variables and a comprehensive hydrological model.