57 · № 1 УДК 551.578 doi:10.15356/2076-6734-2017-1-34-44 Применение методов машинного обучения для моделирования толщины снежного покрова © 2017 г. Г.В. Айзель Институт водных проблем РАН, Москва, Россия hydrogo@yandex.ru Use of machine learning techniques for modeling of snow depth G.V. <...> Ayzel Institute of Water Problems, Russian Academy of Sciences, Moscow, Russia hydrogo@yandex.ru Received Мarch 12, 2016 Keywords: boosting, machine learning, modeling, open data, snow depth. <...> Summary Snow exerts significant regulating effect on the land hydrological cycle since it controls intensity of heat and water exchange between the soil-vegetative cover and the atmosphere. <...> Estimating of a spring flood runoff or a rain-flood on mountainous rivers requires understanding of the snow cover dynamics on a watershed. <...> In this research we used the daily observational data on the snow cover and surface meteorological parameters, obtained at three stations situated in different geographical regions: Col de Porte (France), Sodankyla (Finland), and Snoquamie Pass (USA). <...> Statistical modeling of the snow cover depth is based on a complex of freely distributed the present-day machine learning models: Decision Trees, Adaptive Boosting, Gradient Boosting. <...> It is demonstrated that use of combination of modern machine learning methods with available meteorological data provides the good accuracy of the snow cover modeling. <...> The best results of snow cover depth modeling for every investigated site were obtained by the ensemble method of gradient boosting above decision trees – this model reproduces well both, the periods of snow cover accumulation and its melting. <...> The purposeful character of learning process for models of the gradient boosting type <...>