A hybrid groundwater level prediction model using signal decomposition and optimised extreme learning machine

The estimation and prediction of groundwater levels (GWLs) are key to water resource management and directly linked to the socio-economic growth of sub-Saharan Africa. This current study proposed three novel hybrid denoised artificial intelligence (AI) GWL prediction models, namely: wavelet transform-self adaptive differential evolutionary-extreme learning machine (WT-SaDE-ELM), empirical wavelet transform-self adaptive differential evolutionary-extreme learning machine (EWT-SaDE-ELM), and variational mode decomposition-self adaptive differential evolutionary-extreme learning machine (VMD-SaDE-ELM). First, input hydrometeorological data (rainfall, temperature and evaporation) were denoised (noise filtered) using wavelet transform (WT), empirical wavelet transform (EWT), and variational mode decomposition (VMD). The noise filtered hydrometeorological data then served as the input in the SaDE-ELM to improve GWL prediction accuracy. To verify the potency of the proposed WT-SaDE-ELM, EWT-SaDE-ELM and VMD-SaDE-ELM denoised models, the undenoised (original) hydrometeorological data was applied directly to SaDE-ELM, particle swarm optimisation-artificial neural network (PSO-ANN) and genetic algorithm-artificial neural network (GA-ANN). Statistical indicators such as root mean square error (RMSE), scatter index (SI), mean absolute error (MAE) and Bias were used to assess the model's performance. The comparative statistical analysis revealed that among all the developed models, the denoised hybrid AI models achieved the best performance in GWL prediction for all the 13 boreholes considered. Out of the thirteen (13) boreholes, the WT-SaDE-ELM achieved optimal results for six, VMD-SaDE-ELM had five whilst the EWT-SaDE-ELM had two respectively. To this end, the study has demonstrated that denoising the input parameters can improve the GWL prediction efficiency of machine learning models.