To reveal the recharge effects of snowmelt on soil moisture at different depths in plain regions of Heilongjiang Province, this study utilized long-term ERA5-Land data from 1960 to 2024 on the Google Earth Engine platform, comparing and analyzing the spatiotemporal evolution characteristics of snow water equivalent (SWE) and soil moisture in the Sanjiang Plain and Songnen Plain. By integrating Mann-Kendall tests, Pearson correlation analysis, multiple linear regression, and random forest models, we examined the driving mechanisms of various meteorological factors on soil moisture. Results show that both SWE and soil moisture in the Sanjiang and Songnen Plains exhibited significant declining trends from 1960 to 2024. Pearson correlation analysis indicated a statistically significant positive correlation between SWE and deep soil moisture in the Sanjiang Plain ( r =0.55, p <0.01), whereas the correlation was relatively weaker in the Songnen Plain ( r =0.42, p <0.01). Both multiple linear regression and random forest models consistently demonstrated that the contribution of snowmelt to soil moisture increases significantly with increasing soil depth. In the variation of deep soil moisture, snow water equivalent is the most critical driving factor, with its importance far exceeding that of precipitation and temperature. The random forest model demonstrated higher simulation accuracy across all soil layers compared to linear regression. The study confirms the key role of winter snowmelt in replenishing springtime deep soil moisture, providing scientific support for the protection of Northeast China’s black soil and drought monitoring.