Investigation of Effective Parameters in Modeling Monthly Precipitation using Intelligent Integrated Models Based on Time Series Decomposition

Rainfall forecasting is important in many different aspects of watershed management, such as flood and drought warning systems. Spatiotemporal variations of rainfall cause its prediction to be difficult. In this study, the monthly rainfall of Urmia and Mako stations were assessed using the intelligent kernel-based methods using Ensemble Empirical Mode Decomposition (EEMD) and Wavelet Transform (DWT). For this aim, different models were developed based on teleconnection patterns and climatic elements including rainfall, humidity, and temperature of previous months, and the impact of these parameters on accuracy of the modeling process was investigated. The obtained results showed the high efficiency of the integrated methods used in modeling process. It was observed that in the monthly precipitation modeling, the simultaneous use of teleconnection patterns and climatic elements related to previous months improves the accuracy of the models by up to 35%. Comparison of the wavelet transform and ensemble empirical mode decomposition showed that time series decomposition based on wavelet transform led to more accurate outcomes. The results of sensitivity analysis showed that the precipitation parameter with three months lag was the most effective parameter in monthly precipitation modeling.