Chapter 22 - Application of machine learning models in hydrology: Case study of river temperature forecasting in the Drava River using coupled wavelet analysis and adaptive neuro-fuzzy inference systems model

Abstract

In this study, a hybrid model (WA-ANFIS) which couples wavelet analysis (WA) and adaptive neuro-fuzzy inference systems (ANFIS) is proposed to forecast daily river water temperature as a case study. Four mother wavelets, including Daubechies, Symlet, discrete Meyer and Haar, are considered to develop the WA-ANFIS model. The hybrid model is applied to predict daily water temperature of the lower Drava River in Croatia, Central Europe. Long-term observed daily water temperatures in two river gauges as well as daily air temperatures of two meteorological stations are used. The performance of the WA-ANFIS model is evaluated by comparing the modeling results with those obtained from linear and non-linear regression models as well as the traditional ANFIS model. The results show that the WA-ANFIS models perform well in forecasting river temperature time series, and significantly outperform the linear, non-linear and the traditional ANFIS models. Among the four mother wavelets applied, the Daubechies at level 10 performs the best, slightly better than the discrete Meyer and Symlet, while the Haar mother wavelet has the lowest accuracy. The outcomes of this study have important implications for water temperature forecasting and ecosystem management of river systems.