NH-SDN: A Deep Learning-Derived Daily 0.25° Snow Density Dataset for the Northern Hemisphere (1982–2025)

The Northern Hemisphere Snow Density dataset (NH-SDN) provides daily spatially continuous snow density estimates at a spatial resolution of 0.25° across the Northern Hemisphere from 1982 to 2025. The dataset is generated using SDN-Net, a spatiotemporal deep learning framework that combines Vision Transformer (ViT) and Convolutional Neural Network (CNN) architectures. SDN-Net is trained using quality-controlled in-situ snow density observations from North America and Eurasia and integrates multisource environmental variables associated with snow conditions, meteorological forcing, vegetation dynamics, topography, geographic location, and temporal evolution. SDN-Net captures nonlinear relations between snow density observations and environmental controls. The ViT-CNN architecture enables the model to represent large-scale spatial dependencies and local heterogeneity. NH-SDN provides spatially continuous and temporally consistent snow density estimates across the Northern Hemisphere, addressing the spatiotemporal discontinuity of in-situ observations and the limited accuracy of existing reanalysis products. NH-SDN is distributed as daily GeoTIFF files in a geographic coordinate system (WGS84, EPSG:4326). Each GeoTIFF file represents snow density for one day, and the file naming convention follows YYYYMMDD.tif, where YYYY, MM, and DD indicate the year, month, and day, respectively. Snow density is provided in units of g cm⁻³ and stored as Float32 data without scale factor or offset. Snow-free grid cells are assigned a value of 0. NH-SDN provides a long-term and spatially continuous snow density dataset for investigating snowpack variability. It supports applications including snow water equivalent estimation, hydrological and cryospheric modeling, snow thermal process analysis, and climate change research.