Streamflow Forecasting Using Wavelet Coupled Soft Computing Techniques and Fuzzy Logic-Based Approach for Stream Water Quality-Quantity Assessment

Abstract

Hydrologic time series is a collection of timely recorded variables such as streamflow, temperature, evaporation, etc. over a period of time. Forecasting of such time series necessarily aid future predictions based on past records as well as filling of missing data or extension of available data. Accurate and timely forecasting of hydrologic time series can be a great aid for various applications in water resources planning and management. During the last few decades, several types of stochastic models have been proposed as well as developed for modeling hydrological time series and generating synthetic stream flows. Some of such stochastic models are autoregressive (AR), Moving Average (MA), Autoregressive Moving Average (ARMA), and Autoregressive Integrated Moving Average (ARIMA). In contrast to the analytical models, soft computing methods learn from past records and require limited input parameters. These techniques are very useful in cases where there are limitations in terms of data availability. The collection of techniques under Soft Computing renders low-cost solutions to imprecisely formulated problems and attempt to mimic the behavior and learning ability of human beings into computers. One such soft computing technique is “Fuzzy Logic”. We have developed three soft computing models to forecast daily streamflow time series for different lead times for Malaprabha sub-basin in Karnataka state of India. The performance of Support Vector Machine (SVM), Adaptive Neuro-fuzzy Inference System (ANFIS), and Fuzzy models to forecast daily streamflow is tested for 1-day, 3-days and 5-days ahead forecasts. The results indicate that the performance of the models significantly decreases with an increase in lead times. The models show high R2values for 1-day ahead streamflow forecasts, whereas it is low for 3 and 5-days lead time. It is necessary to provide a powerful tool to reduce the noise in the data so that accuracy of the model is increased. Wavelet transformer is one such powerful tool used to decompose the data set into different scales. The wavelet method effectively decomposes the original time series in to sub-series at different resolution levels there by facilitating denoising of the data. In this research work, discrete wavelet transform is coupled with the fuzzy logic method to improve the accuracy of the forecast. The iperformance of all the three models significantly increased when the wavelet is coupled, especially for longer lead times such as 5 days. The Wavelet coupled fuzzy (WT-fuzzy) model outperformed Wavelet coupled ANFIS (WT-ANFIS) and Wavelet coupled SVM (WT-SVM) models. However, WT-ANFIS performed better than WT- SVM.Longer lead time forecasts find applications in flood forecasting and evacuation programs. This research aims at improving the efficiency of forecasting models especially for longer lead times such as 3 days and 5 days which are crucial times for undertaking quick flood evacuation measures. The second phase of this research is stream water quality-quantity modeling. Water quality and quantity are the two aspects that are interrelated and hence should be studied together within an integrated framework. In today's world, demand for water essentially takes into account both quality and quantity aspects for various uses of water. Having a sufficient accessible quantity of water becomes meaningful only if this quantity of water is acceptable in terms of its quality. This study aims at studying the role of the quantity of water in determining its quality along with the other quality parameters. The Water Quality Index (WQI) is an efficient tool which can describe the status of water by translating a large amount of data in to a single value. The results in this study indicate that streamflow can be considered as one of the inputs to determine the WQI.