Rainfall modeling using two different neural networks improved by metaheuristic algorithms

• Main Authors: Sammen S.S., Kisi O., Ehteram M., El-Shafie A., Al-Ansari N., Ghorbani M.A., Bhat S.A., Ahmed A.N., Shahid S.
• Other Authors: 57192093108
• Format: Article
• Published: Springer 2024
• Subjects: Markov chain; MLP; Probability matrix; Rainfall modelling; RBFNN; Malaysia; Forecasting; Markov processes; Multilayer neural networks; Particle swarm optimization (PSO); Soft computing; Water management; Gas solubility; Mean absolute error; Multilayers perceptrons; Network functions; Optimisations; Probability matrixes; Radial base neural network function; Radial basis neural networks; Testing process; climate modeling; climate prediction; optimization; rainfall; Rain

Rainfall is crucial for the development and management of water resources. Six hybrid soft computing models, including�multilayer perceptron (MLP)�Henry gas solubility optimization (HGSO), MLP�bat algorithm (MLP�BA), MLP�particle swarm optimization (MLP�PSO), radial basis neural network function (RBFNN)�HGSO, RBFNN�PSO, and RBFGNN�BA, were used in this study to forecast monthly rainfall at two stations in Malaysia (Sara and Banding). Different statistical measures (mean absolute error (MAE) and Nash�Sutcliffe efficiency (NSE) and percentage of BIAS (PBIAS)) and a Taylor diagram were used to assess the models� performance. The results indicated that the MLP�HGSO performed better than the other models in forecasting rainfall at both stations. In addition, transition matrices were computed for each station and year based on the conditional probability of rainfall or absence of rainfall on a given month. The values of MAE for testing processes for the MLP�HGSO, MLP�PSO, MLP�BA, RBFNN�HGSO, RBFNN�BA, and RBFNN�PSO at the first station were 0.712, 0.755, 0.765, 0.717, 0.865, and 0.891, while the corresponding NSE and PBIAS values�were 0.90�0.23, 0.83�0.29, 0.85�0.25, 0.87�0.27, 0.81�0.31, and 0.80�0.35, respectively. For the second station, the values of MAE were found 0.711, 0.743, 0.742, 0.719, 0.863 and 0.890 for the MLP�HGSO, MLP�PSO, MLP�BA, RBFNN�HGSO, RBFNN�BA, and RBFNN�PSO during testing processes and the corresponding NSE�PBIAS values were 0.92�0.22, 0.85�0.28, 0.89�0.26, 0.91�0.25, 0.83�0.31, 0.82�0.32, respectively. Based on the outputs of the MLP�HGSO, the highest rainfall was recorded in 2012 with a probability of 0.72, while the lowest rainfall was recorded in 2006 with a probability of 0.52 at the Sara Station. In addition, the results indicated that the MLP�HGSO performed better than the other models within the Banding Station. According to the findings, the hybrid MLP�HGSO was selected as an effective rainfall prediction model. � 2023, The Author(s).