Impacts of combining meteorological and hydrometric data on the accuracy of streamflow modeling

Abstract

Proper modeling of rainfall-runoff is essential for water quantity and quality management. However, comprehensive evaluation of soft computing techniques for rainfall-runoff modeling in developing countries is still lacking. Towards this end, in the present study two new soft computing techniques of genetic programming (GP) and M5 model tree were formulated for daily streamflow prediction. Firstly, the daily meteorological and hygrometric data including rainfall, temperature, evapotranspiration, relative humidity and discharge were collected for the years 1970 - 2012 throughout Amameh Watershed in Tehran, Iran. Secondly, the input variables were determined using cross-correlation and then 62 different scenarios were developed. Thirdly, the data were standardized in the range of zero to one. Finally, performance of the scenarios was assessed using the mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE). Totally, 80 and 20 percent of instances were used for training and testing, respectively. The results showed that the scenario number 54 was the best using both GP and M5 model tree techniques. However, GP showed much better performance than M5 model tree with MSE, RMSE, and MAE values of 0.001, 0.031 and 0.009 for training and 0.001, 0.032 and 0.009 for testing, respectively. The scenario 54 had eight inputs including rainfall, discharge, and delay for two days, temperature, evapotranspiration and relative humidity.

Keywords

• genetic programming
• model development
• m5 model tree
• scenario analysis
• streamflow prediction