DEEP PROCESS-DATA MINING FOR BUILDING OF ANALYTICAL MODELS: 1. MEDIUM-TERM FORECAST OF SPRING FLOOD EXTREMES FOR MOUNTAIN RIVERS

Abstract

A standard methodology of deep process-data mining for building high-performance process-driven (analytical) models of complex natural systems was proposed. The method- ology (called as system-analytical modeling) is based on a system-hierarchical approach and deep mining of large datasets providing both extraction of the information hidden in such datasets and quantitative characterization of real processes occurring in natural systems. With its help, deep process-data mining of data (1951–2020) on spring flood discharge peaks and troughs (with ice motion) on 34 mountain rivers of the Altai-Sayan mountain country was performed. An analytical hydrological model of high performance (Nash-Sutcliffe criterion NSE = 0.78) was developed for the annual medium-term forecasting of discharge peaks and troughs in April using the data on meteorological conditions of the recent autumn and current winter periods. Flood peaks depend on autumn-winter precipitation (which determines 29% of the peak variance), landscape structure of river basins (14%), and winter air temperatures (0.8%). Spring floods on mountain rivers often threaten the life of local population that makes the developed model topical.