Estimation of maximum water levels during spring flood on Lena river parts using artificial neural networks

• Published in: Journal of Physics: Conference Series
• Main Authors: Struchkova, G. P., Kapitonova, T. A., Timofeeva, V. V., Nogovitsyn, D.D.
• Format: Article in Journal/Newspaper
• Language: unknown
• Published: IOP Publishing 2019
• Subjects: Sakha; lena river; Republic of Sakha; Yakutia
• Online Access: http://dx.doi.org/10.1088/1742-6596/1392/1/012025; https://iopscience.iop.org/article/10.1088/1742-6596/1392/1/012025/pdf; https://iopscience.iop.org/article/10.1088/1742-6596/1392/1/012025

Abstract The Republic of Sakha (Yakutia) possessing a vast territory located in various climatic zones and a developed network of water bodies is exposed to a wide range of natural emergency situations. Among them, spring-summer floods are the most frequent and bringing enormous damage, causing inundations of vast areas and national economy objects, which determines relevance of development and improvement of flood forecast methods for implementation of timely measures to prevent and reduce an inundation risk. Artificial neural networks have proven their effectiveness in solving various forecast problems, especially when using statistical data. In particular, usage of a neural network approach based on the forecast of a time series from previous values gives good results. The artificial neural networks, unlike statistical methods of analysis, are based on parallel data processing, have an ability of self-learning and recognition of nonlinear relationships between input and output data sets. A choice of parts of the Lena river to predict maximum water levels during the floods was determined based on locations of potentially hazardous objects, the inundation of which can cause the considerable material damage. On the basis of the hydrological data obtained during 70 years, the neural network models are obtained, which make it possible to predict flood hazards from the spring floods, two variants of the transformed initial data are considered and different network structures are compared. Relative errors of the forecasts obtained during the work vary considerably (7–20%), which indicates necessity for the processing of the initial data and careful selection of the structure.