Author(s): Azazkhan I. Pathan; Saif Said

Linked Author(s):

Keywords: Gene expression programming; Multiple linear regression; Rainfall forecasting; Mean absolute error; Root-mean-squared error

Abstract: Predicting rainfall, which is a major component of the hydrological cycle, is critical in water resources engineering, planning, and management, as well as in irrigation scheduling. Being an agricultural area, Aligarh district in Uttar Pradesh, India, has a greater interest in developing an efficient model to predict the monthly rainfall. Therefore, this paper presents a comparative analysis of two different techniques, viz., gene expression programming (GEP) and multiple linear regression (MLR), to predict the average monthly rainfall in Aligarh district. The non-spatial average monthly weather data, including rainfall, temperature, relative humidity, wind speed, and cloud cover from 2009 to 2019, were selected for calibration and validation of the models. Results revealed that the optimized GEP model produced average monthly rainfall estimates with considerably high accuracies, yielding R2 ~ 0.91 and 0.86, RMSE of 18.63 and 19.27, and MAE of 10.6 and 11.57 for calibration and validation data, respectively. The MLR model produced rainfall estimates with moderate R2 ~ 0.77 and 0.73, RMSE of 22.25 and 32.27, and MAE of 15.65 and 21.76 for calibration and validation data, respectively. Analysis of the results revealed that the GEP model performs substantially well with high forecast accuracies, thereby indicating significant rationality in the optimized GEP model. Further, sensitivity analysis performed on the weather parameters showed that cloud cover exhibited the highest significance of 96.7% on rainfall estimation. The developed GEP-based model can be used by town planners and agricultural professionals.

DOI: https://doi.org/10.1007/978-981-97-6009-1_25

Year: 2022