Method for reconstructing the monthly mean water transparencies for the northwestern part of the Black Sea as an example

Abstract

A model based on neural networks with a teacher is suggested to reconstruct the data of observations of hydrophysical parameters. The indices of global climate oscillations in the ocean–atmosphere system were used as input signals for the model. The processes of model training and adaptation, permitting the most accurate solution of the modeling problem, is described. A comparison of model-predicted monthly mean of the Danube runoff volumes with observational data showed their good agreement. Gaps in data of observations of the water transparency (visibility depth of a white disk) in different years are filled by reconstructed data. A close correspondence in absolute value between reconstructed and measured visibility depths of the white disk is reported. Some features of interannual variations in reconstructed data on water transparency, caused by both climatic factors during 1950–1962 and changes in the chlorophyll а concentration during 1998–2010, were revealed.