Advanced genetic programming models for precipitation forecasting

Rain forecasting is an intricate task as hydrological processes tend to be nonlinear and dynamic. This thesis formulates an improved hybrid model technique for forecasting short-term precipitation by integrating genetic programming (GP) and wavelet-based signal decomposition. The meteorological parameters of Antalya, Turkey, were employed here and analyzed between 2016 and 2023, covering variables such as temperature, humidity, sea-level pressure, wind speed, solar radiation, and precipitation. The raw data were initially preprocessed by performing normalization, treating missing values, and eliminating outliers. Next, meteorological variables were used for wavelet decomposition to form the multi-scale components of the raw and lagged data. The description of principles and the development of several GP models were completed. These GP models were tested with different combinations of inputs, lagged data, and wavelet-cellulated data. The model assessment measures were calculated for each model, including the RMSE and NSE scores, to decide which model outperforms the others. The outcome clearly shows that the GP model integrating temporal features but excluding the wavelet transformation provided the most precise prediction, while the models with wavelet components portrayed lower performance due to the domination of the complexity of the inputs. The model's efficiency was significantly increased via the feature selection based on the correlation analysis approach without any loss of model accuracy. In general, the study underlines the possibility of GP-based models to be used in the case of rainfall forecasting tasks, and the importance of temporal information and feature selection is discussed. In the future, further studies should be conducted to test different hybrid strategies and feature engineering methods that might improve forecasting accuracy. KEYWORDS: Genetic programming, wavelet decomposition, rainfall forecasting, symbolic regression, feature selection, Antalya