A double sigmoidal model for snow-rain phase separation

Abstract

This paper presents the use of a double sigmoidal model for precipitation phase partitioning in mountains. For the model development and evaluation, daily rainfall, minimum and maximum air temperatures, and snow depth data from twelve synoptic stations in the mountainous regions of Iran are used. The model performance is compared against six air temperature-based methods using the Taylor diagram, the coefficient of determination (R2), and the Root Mean Square Error (RMSE) performance indicators. Comparing the frequency of snowfall events estimated by the proposed model with those of the existing models and observation at the selected stations showed better performance of the proposed model in precipitation phase discrimination at the studied stations. The average R2 statistic that relates the frequency of snowfall occurrences estimated by the proposed model with the frequency of snowfall events observed at the studied stations is 0.989, which is the highest compared to other available models. Its average error across the stations is 4%, the lowest among the examined models. A split sample approach was used to ensure that the parameters of the proposed model are appropriately estimated. The result showed that the R2 and RMSE values of both calibration and validation subsets are not vastly different from each other, suggesting that the parameters of the two experiments are stable and the corresponding results are robust. The proposed model also performs better than other existing models in accurately estimating air temperature thresholds required for precipitation phase discrimination as well as the snow depth data in all studied stations.