Estimating the riverine environmental water demand under climate change with data mining models

This paper presents a statistical approach based on data mining to estimate the riverine environmental water demand (EWD). A river’s environmental water demand defines the quantity, timing, and quality of streamflow that are required to sustain riverine ecosystems and human activities. Genetic programming (GP), artificial neural network (ANN), and support vector regression (SVR) are herein applied to model the environmental demand. Input and output data for the use of GP, ANN, and SVR are the average monthly temperature and precipitation in 1995–2005 plus climate projections by the Canadian Land System Model (CanESM2) under the recommended concentration pathways RCPs 2.6, 4.5 and 8.5 in 2025–2035. A case study illustrates this paper’s methodology using temperature and precipitation data and monthly discharge of the Karaj River, Iran. The applied data mining models were evaluated with R2, RMSE, and the NSE criteria. This work’s results show that the largest values of R2 and the NSE equal respectively 0.94 and 0.95, and the smallest value of the RMSE equals 0.07, which correspond to the SVR predictions. These results establish that SVR is a suitable model for the purpose of estimating the environmental water demand in comparison to GP and ANN in the study area. The SVR projections indicate that by 2035 and under the RCPs 2.6, 4.5, and 8.5 projected changes of the environmental water demand with respect to baseline conditions would be respectively 63, 118, and 126 m3/s. It is demonstrated in this work that under climate change conditions the correlation between the EWD index and temperature was 83%, while the said value for rainfall was estimated to be 76%.