Loss of snow persistence during late season drives snow-season shortening in Northeast China under global warming

Northeast China lies on a mid-latitude boundary where small warming can shift snowfall to rain and shorten the snow season, but climate models often simulate snow fields with large regional biases. Here we show that satellite-constrained, 1-km machine-learning emulators project snow-cover duration without directly using modeled snow variables. They relate observed daily snow cover during 2001–2020 to temperature, precipitation, snowfall fraction and terrain, and are driven by bias-corrected climate projections. Snow seasons shorten almost everywhere, with more than 98% of land grid cells agreeing on the sign of change. By 2081–2100, domain-mean snow-cover duration decreases by 42, 64, and 85 days under low-, intermediate- and high-emission pathways, respectively, and areas with fewer than 50 snow days expand from 19% to 99% under high emissions. Snow-season shortening is dominated by loss of late-winter and spring snow persistence, especially at low–middle elevations and along the present rain–snow transition zone. The loss of snow persistence during late winter to spring across Northeast China leads to shorter snow season under global warming, as suggested by bias-corrected CMIP6 climate projections based on Satellite-constrained machine learning framework.