Comparison of artificial neural network and decision tree models in estimating spatial distribution of snow depth in a semi-arid region of Iran

•Snow depth was digitally mapped at a fine resolution in a semi-arid region of Iran.•Two common algorithms were compared.•Different source of ancillary data was applied.

There is no doubt that snow cover plays an important role in the hydrological cycle of mountainous basins. Therefore, it is essential to measure snow parameters such as snow depth and snow water equivalent in these areas. The aim of this study is to estimate the snow depth from terrain parameters in the Sakhvid Basin, Iran using artificial neural networks (ANNs) and M5 algorithm of decision tree. For this purpose, snow depths were measured in 206 sites based on systematic network. Furthermore, 30 terrain parameters were extracted from a digital elevation model (DEM) of the basin. The results indicated that the decision tree model is the most suitable method to estimate snow depth in the study area with a Nash-Sutcliffe Efficiency (Ens) of 0.80, followed by ANNs with an Ens of 0.73. Moreover, the most significant parameters in the M5 decision tree algorithm are: channel network base level, stream power, wetness index and height.