Оценка качества моделирования и прогнозирования в гидрологии – рассуждение о методе

The currently existing hydrological models are provided with extremely limited methodological tools for their calibration, verification, and testing. It seems extremely relevant to develop, on a unified methodological basis, a system of quality assessment for a number of typical hydrological tasks – parameterization and verification of hydrological models, simulation and scenario modeling of the hydrological regime, forecasting, engineering calculations, etc. Without solving this problem, further development and effective practical use of hydrological models are impossible.

The paper demonstrates the logic and, to some extent, the history of the development of some of the most important ideas and approaches used in assessing the quality of modeling and forecasting, as well as to outline some prospects for their development. The necessary systematization of the terms of the subject area under consideration has been carried out, starting with the basic concepts of “prediction” and “forecast”, with consideration of additional definitions: “continuous – aggregated” forecasts; “short-term – long-term” forecasts; “methodological”, “standard”, “ideal”, “inertial”, “regime” forecasts; and etc.

The stages of development of the “skill scores” approach in the development of measures for assessing the quality of hydrological modeling and forecasting are considered, mainly using the example of the emergence and modification of a basic measure for hydrology - the Nash-Sutcliffe efficiency coefficient. Special attention is paid to two decompositions of the NSE measure, demonstrating its reducibility to the basic statistical parameters of the samples of measured and predicted values of the variable – average values, standard deviations and correlation coefficient. A rationale for modifying the NSE estimate by applying it to pre-ranked samples of variable values is proposed. An approach to assessing the quality of modeling and forecasting is formulated based on a consistently statistical approach, using statistical criteria for the similarity of marginal distributions of actual and forecast values of a variable. A number of tasks for future research are outlined in the field of developing methods for assessing the quality of hydrological models and forecasts.