Good-quality Long-term Forecast of Spring-summer Flood Runoff for Mountain Rivers

Abstract

The universal simulation model was developed with the use of system-analytical modeling to ensure a long-term forecast of mountain river runoff during spring-summer floods. Prediction quality of this SAM-model is characterized by Nash-Sutcliffe efficiency of 0.68–0.88 and is very high for long-term flood forecasts, including ones for inundations and mountain reservoirs filling in spring. The model was tested on the example of 34 medium and small rivers (1630 values of runoff observations for 1951–2016) located in the Altai-Sayan mountain country (2,000,000 km2). Its input factors include monthly precipitation, monthly mean air temperature, GIS data on landscape structure and orography of river basins. Meteorological factors are calculated as percentage of their “in situ” long-term mean values averaged for the whole study area. This helps to explain and quantify the influence of autumn-winter-spring soaking, freezing and thawing of mountain landscape soils on spring-summer flood. We apply a simple novel method to evaluate model sensitivity to variations in environmental factors expressed in terms of their contribution to variance of the observed flood runoff. It turns out that sensitivity of the latter decreases in the following sequence of factors: autumn precipitation, landscape structure of river basins, winter precipitation, winter air temperature, landscape altitude. The developed SAM-model provides a three-month lead-time estimate of runoff in a high water period with the threefold less variance as compared to forecasts based on the observed long-term mean values.