Quantifying snowmelt contribution to streamflow using ensemble machine learning and remote sensing

Abstract Snowmelt plays a vital role in sustaining river flows and regulating water resources in mountain ecosystems, particularly under changing climatic conditions. This study aims to quantify the contribution of snowmelt to streamflow by estimating snow cover and snow water equivalent (SWE) in the Greater Caucasus Basin of Azerbaijan. MODIS and FLDAS remote sensing products were analyzed using the Random Forest machine learning algorithm, with data processed on the Google Earth Engine (GEE) platform to improve prediction efficiency and accuracy. Results show that Random Forest significantly enhanced snow cover estimation compared to satellite data alone. MODIS and FLDAS observations revealed peak snow cover in December 2023 (1,310 km²) and maximum SWE in February 2023 (23 kg/m²). Over the longer period 2013–2023, the highest snow cover was recorded in 2016 (5,052 km²), while peak SWE occurred in 2019 (188.7 kg/m²). These findings indicate a weakening role of snow in regional runoff regulation and water supply, driven by declining snowfall and streamflow. The study demonstrates the value of machine learning for snowpack monitoring and highlights the need for long-term simulations and expanded ground-based networks to improve prediction of peak flow events in mountain basins.