Abstract
The Sierra Nevada mountain range in Spain, located at 37°N, hosts the southernmost snow-covered area in Europe. The highly changeable weather conditions significantly affect the energy and mass balance of the snowpack, resulting in a high interannual variability of snow phenology, often characterized by multiple accumulation and ablation periods. Up to 30% of the total annual snow ablation is attributed to the effect of evaposublimation, which is amplified by intense solar radiation and strong dry winds.
Since the snow portion underlying evaposublimation does not contribute to the melting process and reduces the amount of available water resource in downstream areas, analyzing evaposublimation is crucial for predictions on water availability in Andalusia. However, intense snowpack monitoring in this area is not feasible given the challenges in accessing and collecting data on a continuous basis in the terrain. Thus, snow dynamics are simulated with a physically based model at plot scale. One of the key input parameters is the roughness length (z0), which influences the coefficients of latent-heat and sensible-heat transfer, and consequently evaposublimation. For simplicity, z0 is often assumed to be temporally and spatially constant, even though this parameter is known to vary in a range of several magnitudes depending on the state of snow metamorphosis. This introduces a large uncertainty in modeled evaposublimation rates.
To address this issue, we conduct a sensitivity analysis to evaluate the impact of a temporally variable z0 on evaposublimation estimates. For this purpose, we introduce different classes of snow surface roughness derived from daily terrestrial photographs at a pilot area in Sierra Nevada based on an unsupervised learning algorithm that utilizes snow cover area and texture features related to the current snow phenology obtained from the photographs. We use the resulting time series of roughness classes to calibrate the snow model with respect to measured snow height and snow cover fraction thus obtaining a temporally variable z0 for our study site.
The results of the sensitivity analysis demonstrate significant changes in evaposublimation rates with z0 at a local scale. Introducing a temporally variable z0 thereby improves snow height estimates and thus evaposublimation which is estimated to contribute with up to 40% to the total annual snow ablation, surpassing previous estimates in this region. Our results suggest that the use of a temporally variable z0 derived from terrestrial photography is an easy implementation towards improving snow modeling and water resource forecasts in this region.