To study the water cycle and water resources evolution laws in the inland river basins of Northwest China under climate change, the sub-basins of Changmabao Station on the Shule River and Kaqun Station on the Yarkand River were selected as the research objects. The Sen’s slope method was used to analyze the variation characteristics of hydro-meteorological elements from 1956 to 2016. The Mann-Kendall trend test was used to identify the years of runoff abrupt changes. The random forest model and SHAP method were applied to quantify the effects of precipitation, snow meltwater, and glacier meltwater on runoff. The future runoff change trend was predicted based on CMIP6 multi-model data. The findings indicate that from 1956 to 2016, both the Changmabao and Kaqun station sub-basins experienced a significant warming-wetting trend, with abrupt change years in annual runoff occurring in 1993 and 1997, respectively. Under the background of warming-wetting climate, precipitation and glacier meltwater in the two sub-basins showed increasing trends, while snow meltwater exhibited divergent trends with one rising and the other declining. The random forest model can effectively simulate the monthly runoff variations in both sub-basins, achieving the Nash-Sutcliffe efficiency coefficient of 0.975-0.976 and 0.984-0.985 in the calibration period, respectively. In the Changmabao Station sub-basin, glacier meltwater had the greatest impact on runoff, followed by snow meltwater and precipitation; in the Kaqun Station sub-basin, glacier meltwater had the greatest influence before the abrupt change and during the overall period, while influence of precipitation exceeded snow meltwater after the abrupt change. Projections based on CMIP6 multi-model dataset indicate that the annual runoff of the two sub-basins may continue to increase in the future, but the increasing trend will weaken.