Rainfall Forecasting by Using Machine Learning Models: A Case Study of TRNC

Rainfall forecasting is crucial to making decisions, managing irrigation resources, and agriculture, and even predicting floods. Mediterranean rain regime is effective on Turkish Republic of Northern Cyprus (TRNC). The use of machine learning methods is widespread in many fields, including engineering, agriculture, transportation, and for the prediction. Several machine learning procedures were used in this study to build daily rainfall prediction models, including Decision Trees, Random Forests, Bagging Regressions, and Stacking Regressions. Five climatic parameters, average temperature, specific humidity, relative humidity, wind speed, and wind direction datasets were compiled on daily bases from 1995 to 2022 and used as input parameters after training and test phases. A comparison between the actual rainfall data gathered from NASA and the predicted outcome rainfall data from the machine learning models were used to determine the appropriate model which was having maximum accuracy and minimum error. In order to evaluate them, the statistical measures, R2, MSE, and MAE were used. It is determined that, the two most accurate models for predicting daily rainfall as a whole of TRNC, were Stacking Regression, and Random Forest with R2, 95.66, and 95.43, MSE 0.0428 and 0.045, and MAE 0.0821 and 0.0891, respectively. By applying the similar approach, based on the selected meteorological station as a representative for each region of TRNC, two appropriate machine learning models were found to be the best two fitted models that are Stacking Regression and Bagging Regression.