TY - JOUR

T1 - Experience in Applying Probabilistic Approaches in Predicting the Level Regime of the Marmarik River

AU - Сумачев, Александр Эдуардович

AU - Гайдукова, Екатерина Владимировна

AU - V.G. Margaryan, null

AU - A. M. Sedrakyan,

N1 - Sumachev, A.E., Gaidukova, E.V., Margaryan, V.G. et al. Experience in Applying Probabilistic Approaches in Predicting the Level Regime of the Marmarik River. Dokl. Earth Sc. (2024). https://doi.org/10.1134/S1028334X2460124X

PY - 2024/6/1

Y1 - 2024/6/1

N2 - Abstract: The possibility of short-term and long-term forecasting of water levels, including those associated with dangerous hydrological phenomena on the Marmarik River, using various probabilistic approaches, including regression dependencies, an integrated moving average autoregression model, and multilayer perceptron models, is considered. To evaluate the effectiveness of prognostic methods, the statistical parameters of a random process are calculated, while recommendations are given using the classical criteria for the effectiveness of issued forecasts. For long-term forecasting, the expediency of using the integrated moving average autoregression model was assessed, while it is noted that these models in the classical representation are not applicable due to time gaps, and therefore it is recommended to focus on the mathematical expectation of a random process. For short-term forecasting one or two steps ahead, the method of training artificial neural networks was used. The analysis carried out in the work revealed that in the case of short-term forecasting of water levels for one period in advance (12 h), it is most expedient to focus on the value of the water level attributable to the date of issue of the forecast, the standard error of such a forecast is 5 cm. For a 24-h water level forecast forward, it is expedient to develop neural network forecasting models, taking into account the development of the situation on Gomraget-Meghradzor. A further increase in the quality of the outputs is possible when using data for a longer observation period and a whole year. At the same time, as an alternative to neural network forecasting models, physical and mathematical (hydraulic) models of the formation of water levels can be used.

AB - Abstract: The possibility of short-term and long-term forecasting of water levels, including those associated with dangerous hydrological phenomena on the Marmarik River, using various probabilistic approaches, including regression dependencies, an integrated moving average autoregression model, and multilayer perceptron models, is considered. To evaluate the effectiveness of prognostic methods, the statistical parameters of a random process are calculated, while recommendations are given using the classical criteria for the effectiveness of issued forecasts. For long-term forecasting, the expediency of using the integrated moving average autoregression model was assessed, while it is noted that these models in the classical representation are not applicable due to time gaps, and therefore it is recommended to focus on the mathematical expectation of a random process. For short-term forecasting one or two steps ahead, the method of training artificial neural networks was used. The analysis carried out in the work revealed that in the case of short-term forecasting of water levels for one period in advance (12 h), it is most expedient to focus on the value of the water level attributable to the date of issue of the forecast, the standard error of such a forecast is 5 cm. For a 24-h water level forecast forward, it is expedient to develop neural network forecasting models, taking into account the development of the situation on Gomraget-Meghradzor. A further increase in the quality of the outputs is possible when using data for a longer observation period and a whole year. At the same time, as an alternative to neural network forecasting models, physical and mathematical (hydraulic) models of the formation of water levels can be used.

KW - artificial neural networks

KW - forecasting

KW - long-term forecasts

KW - short-term forecasts

KW - water level

UR - https://www.mendeley.com/catalogue/e1becd77-0742-379f-9ba5-468e8a175a9c/

U2 - 10.1134/s1028334x2460124x

DO - 10.1134/s1028334x2460124x

M3 - Article

VL - 516

SP - 1067

EP - 1074

JO - Doklady Earth Sciences

JF - Doklady Earth Sciences

SN - 1028-334X

IS - 2

ER -