Abstract Collecting reliable meteorological data in mountain environments is difficult because harsh conditions, complex terrain, and limited access disrupt long-term observations. In the Snake Range, Nevada, snow cover of low-cost temperature and humidity sensors creates winter gaps in near-surface air temperature records. To address this challenge, we developed three Random Forest Regression models to predict 2-m daily maximum, mean, and minimum air temperatures from snow covered sensors using temperature and humidity inputs. We trained the model with data from two well instrumented sites and deployed it at 27 additional sites, spanning 1639–3976 m elevation. The model achieved mean absolute errors of 0.49 °C to 1.52 °C and reproduced daily temperature patterns well despite predictable biases in maximum and minimum temperatures. It also preserved long-term temperature trends, allowing reconstructed values to fill winter gaps without distorting multi-decadal warming rates. Applying the model across the network showed that previously developed snow-free methods, which removed all snow-covered days from the analysis, overstated warming by excluding many of the coldest days. The model-adjusted dataset produced a park-wide temperature increase of 0.79 °C from 2006 to 2025. This is substantially lower than the nearly 3 °C suggested by the snow-free approach. Snow-covered sensors paired with a validated temperature adjustment model can reliably recover near-surface air temperature records across complex terrain. Using only low-cost temperature and humidity inputs, the method supports denser and more accessible climate monitoring in mountain regions where instrumentation remains difficult to deploy and maintain.
Mazan et al. (Mon,) studied this question.