A zero-inflated spatiotemporal analysis for snowpack variations and influence of environmental factors in the Northern Hemisphere

Abstract

Snowpack will vary with climate change and is of vital importance for water supply, the energy balance, and heat exchange. Environmental factors such as thermal, atmospheric, terrestrial, and hydrological factors influence variations in snow mass. This study aimed to investigate associations with environmental factors and variations in snow mass changes based on satellite-retrieved products in the Northern Hemisphere (NH). A zero-inflated space–time autoregressive (ZI-STAR) model is proposed to identify the environmental factors associated with the SWE changes during the snow accumulation and ablation periods in North America and Eurasia because of the spatial patterns, temporal lag effect, and excessive zero observations captured with the satellite-based snow water equivalent (SWE) product. The proposed ZI-STAR model, which considers spatial, time lag, and zero-inflation effects together, offered the best overall performance for SWE simulations with high precision (average R2 of 0.90 while the others range from 0.69 to 0.84) and low spatially random distributed residuals (RMSE 23%–33% less than other model), compared to time-lag, and spatial autoregressive models. Based on the results, air temperature and snow precipitation dominate the variations in SWE across the period of interest. The influence of water vapor varies and the wind speed and topography have little effect due to the missing SWE observations in mountainous areas. The variations in the environmental factors for SWE furnish a foundation for further research about snow changes in the NH. The proposed ZI-STAR model also gives a feasible approach for spatio-temporal continuous variables modeling with excessive zero observations in other fields.