Comparison of Decision Tree Based Rainfall Prediction Model with Data Driven Model Considering Climatic Variables

Abstract

In hydrological cycle, precipitation initiates the flow and governs the system. The preciseness in the prediction of rainfall will reduce the uncertainty involved in estimating the associated hydrological variables such as runoff, infiltration, and stream flow. Many research works has been channelled towards improving the accuracy of these predictions. ANN is the most widely used neural networks in Integrated Water Resource Management. Most of these models, utilize the strength of data-driven modelling approach. The reliability of these predictions depends on the preciseness in selecting the correlated variables. If the available historical database fails to record the most correlated variable, then reliability on these data-driven approach predictions is questionable. In this paper, an attempt has been made to develop a methodological framework that utilizes the strength of a predictive data-mining analysis (decision tree). The developed decision tree based rainfall prediction model maps the climatic variables, namely; a) temperature, b) humidity, and c) wind speed over the observed rainfall database. The performance of the developed model is evaluated based on three performance indicators (Nash Sutcliffe efficiency, RMSE and MSE). The performance of the developed model is also compared with the well- known data-driven (Artificial Neural Network) based rainfall prediction model.