Study region: The northern slope of the Kunlun Mountains is a high-cold arid region on the northwestern edge of the Tibetan Plateau characterized by extreme altitudes (>4500 m), low precipitation, and extensive glacier coverage (3.5%). Study focus: We reconstructed past (1990-2020) and predicted future (2021-2030) responses of lake water storage to climate change for typical lakes along the northern slope of the Kunlun Mountains using multi-source satellite datasets (Landsat, altimetry, and GRACE) and back propagation (BP) neural network modeling. New hydrological insights: Significant lake area expansion (+77.19%) and increases in lake number (+162.79%) were observed post-2003. Lake Ayagkumu showed the most significant increase in water storage change rate, increasing at 0.0418 km(3)/a post-2003 compared to -0.0071 km(3)/a pre-2003. Monthly-scale analysis revealed abrupt changes in lake water storage, with significant increases observed for Lake Aqqikkol in April 2003. Seasonal patterns in lake water storage were also observed, with the warm season (May-September) exhibiting peak growth. BP neural network predictions showed continued expansion for most lakes with good accuracy (+/- 10% deviation vs. observations), confirming the good predictive capability of the model. Validation using GRACE data further confirmed model reliability (0.0895 km(3)/a vs. 0.0903 km(3)/a observed). These findings clarify hydrological processes in high-cold arid regions and serve as an important scientific basis for assessing water resource vulnerability and formulating adaptation strategies.