Probabilistic Melt Model Based On Snow Cover and Lst Using Landsat 8 Images In The Sierra Nevada de Santa Marta

This article presents a diagnosis of snowmelt in the Sierra Nevada de Santa Marta based on Landsat 8 data. The objective is to estimate, between consecutive dates, the probability and expected fraction or area of snow loss by integrating snow cover and surface temperature maps. The methodology combines three steps: supervised classification of snow into five classes using k-nearest neighbors; estimation and harmonization of surface temperature using the thermal product and a complementary calculation from the B10 band with MTL metadata to fill gaps; and fitting a probabilistic model with logistic regression using temperature and thermal change rate, with temporal calibration to obtain reliable probabilities. As a result, the classification achieved an overall accuracy of 99.90% and a Kappa of 0.9872; additionally, the model produced consistent probabilities per pixel with an AUC close to 0.80 for estimating the expected fraction and area of thawing per date pair.