Estimating daily high-resolution snow cover in the Central Pyrenees using logistic regression with Sentinel-2 and MODIS data

The presence or absence of snow in the landscape strongly modulates land surface energy exchanges and governs key ecosystem processes in high-mountain catchments. In mid-latitude mountain regions, such as the Pyrenees, which are dominated by intermittent and ephemeral seasonal snowpacks, pronounced intra-annual spatial variability in snow cover complicates the accurate characterisation of snow temporal dynamics.

Although a wide range of snow products and remote sensing platforms are currently available, many of them have significant limitations when applied to complex, mountainous environments such as the Pyrenees. These limitations include data gaps caused by cloud cover and confusion between snow and clouds; reduced accuracy in areas affected by topographic shadows; insufficient illumination due to low solar elevation at the time of satellite overpass; and the trade-off between spatial and temporal resolution. Furthermore, as most existing products are designed for large-scale applications, they can introduce significant errors when high spatial detail is required. This is particularly pertinent in catchment- and sub-catchment-scale hydrological, biogeochemical, and ecological studies.

In this context, we propose a methodological approach that combines the daily temporal resolution of snow gap-filled MODIS products with Sentinel-2-derived snow cover as the ground truth, using k-nearest neighbour (k-NN) classification. We generated daily binary snow presence/absence maps at a spatial resolution of 20 m over the study area using a logistic regression model incorporating general explanatory variables such as elevation, slope, aspect, monthly solar radiation and the spatial and temporal information of snow cover, such as distance-to-snow maps derived from MODIS.

Preliminary results show that the logistic regression framework generates daily snow cover maps that are spatially and temporally consistent, substantially reducing data gaps and improving the representation of intermittent and ephemeral snow zones, which are expected to become increasingly prevalent under future climate change. Model outputs were evaluated against independent ground-based observations, including snow pole measurements, telenivometer data, showing good agreement across elevation gradients and seasons. Together, these results demonstrate the potential of the proposed approach to capture fine-scale spatio-temporal variability in snow cover, providing a robust basis for catchment-scale analyses of snow–hydrology and snow–biogeochemistry interactions in high-mountain regions.