Augmentation of limited input data using an artificial neural network method to improve the accuracy of water quality modeling in a large lake

被引：0

作者：

Kim, Jaeyoung ^{[1
]}

Seo, Dongil ^{[1
]}

Jang, Miyoung ^{[2
]}

Kim, Jiyong ^{[2
]}

机构：

[1] Department of Environmental Engineering, Chungnam National University, 99, Daehak-ro, Yuseong-gu, Daejeon,34134, Korea, Republic of

[2] Smartdata Research Section, AI Research Lab, Electronics and Telecommunications Research Institute, Daejeon,34129, Korea, Republic of

来源：

Journal of Hydrology | 2021年 / 602卷

关键词：

Lakes;

D O I：

暂无

中图分类号：

学科分类号：

摘要：

The performance of water quality models depends on both data from the external inputs and the internal processes of a water body. Limited field data can often be the major cause of errors in water quality prediction when modeling, especially in large environments. The aim of this study was to improve the prediction accuracy of water quality in a large lake using the combined application of an artificial neural network (ANN) method and a numerical model. Multilayer perceptron (MLP) method was used as an ANN method to generate temporal input data by learning complex relationships of water quality variables from two types of water quality monitoring systems at major boundaries. A regular monitoring system analyzes 13 water quality variables in 3 layers monthly or weekly, while the automatic monitoring system analyzes 8 surface water quality variables daily. The Environmental Fluid Dynamics Code (EFDC), 3-D hydrodynamics and water quality model, was calibrated with 55,588 grids to simulate the water quality of the 46.5 km section of Daecheong Lake. The accuracy of the EFDC models was assessed at four locations in the lake for the application of daily data generated by MLP models against that of interpolated data from a regular monitoring system as input of EFDC boundary conditions. According to the averaged index of agreement (IA), the performance of MLP-EFDC showed more accurate results than the EFDC using interpolated data for most variables. In particular, the maximum increase in the average IA was 14.4% for total phosphorus. However, the performances of MLP-EFDC were not significantly improved in the downstream section of the study area, where the input effects were mixed with the internal processes of the lake. This study shows that (1) unmonitored temporal input data can be developed using ANN techniques if data for learning processes are available, and (2) the linkage between the ANN technique and the numerical model can improve the prediction accuracy of the water quality in a large lake. © 2021 Elsevier B.V.

引用