This study investigates the application of the Gaussian Radial Basis Function Neural Network (GRNN), Gaussian Process Regression (GPR), and Multilayer Perceptron Optimized by Particle Swarm Optimization (MLP-PSO) models in analyzing the relationship between rainfall and runoff and in predicting runoff discharge. These models utilize autoregressive input vectors based on daily-observed TRMM rainfall and TMR inflow data. The performance evaluation of each model is conducted using statistical measures to compare their effectiveness in capturing the complex relationships between input and output variables. The results consistently demonstrate that the MLP-PSO model outperforms the GRNN and GPR models, achieving the lowest root mean square error (RMSE) across multiple input combinations. Furthermore, the study explores the application of the Empirical Mode Decomposition-Hilbert-Huang Transform (EMD-HHT) in conjunction with the GPR and MLP-PSO models. This combination yields promising results in streamflow prediction, with the MLP-PSO-EMD model exhibiting superior accuracy compared to the GPR-EMD model. The incorporation of different components into the MLP-PSO-EMD model significantly improves its accuracy. Among the presented scenarios, Model M4, which incorporates the simplest components, emerges as the most favorable choice due to its lowest RMSE values. Comparisons with other models reported in the literature further underscore the effectiveness of the MLP-PSO-EMD model in streamflow prediction. This study offers valuable insights into the selection and performance of different models for rainfall-runoff analysis and prediction.
Characterized by their high spatiotemporal variability, rainfalls are difficult to predict, especially under climate change. This study proposes a multilayer perceptron (MLP) network optimized by Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Firefly Algorithm (FFA), and Teleconnection Pattern Indices - such as North Atlantic Oscillation (NAO), Southern Oscillations (SOI), Western Mediterranean Oscillation (WeMO), and Mediterranean Oscillation (MO) - to model monthly rainfalls at the Sebaou River basin (Northern Algeria). Afterward, we compared the best-optimized MLP to the application of the Extreme Learning Machine optimized by the Bat algorithm (Bat-ELM). Assessment of the various input combinations revealed that the NAO index was the most influential parameter in improving the modeling accuracy. The results indicated that the MLP-FFA model was superior to MLP-GA and MLP-PSO for the testing phase, presenting RMSE values equal to 33.36, 30.50, and 29.92 mm, respectively. The comparison between the best MLP model and Bat-ELM revealed the high performance of Bat-ELM for rainfall modeling at the Sebaou River basin, with RMSE reducing from 29.92 to 11.89 mm and NSE value from 0.902 to 0.985 during the testing phase. This study shows that incorporating the North Atlantic Oscillation (NAO) as a predictor improved the accuracy of artificial intelligence systems optimized by metaheuristic algorithms, specifically Bat-ELM, for rainfall modeling tasks such as filling in missing data of rainfall time series.
In the present paper, we propose a new approach for monthly streamflow prediction based on the extreme learning machine (ELM) and the metaheuristic Bat algorithm (BAT-ELM). The performance of the BAT-ELM was compared to that of ELM, support vector regression (SVR), Gaussian process regression (GPR), multilayer perceptron neural network (MLPNN), and the generalized regression neural network (GRNN). The proposed models were applied using data from three hydrometric stations located in the Cheliff Basin, Algeria. It was shown that the BAT-ELM was more satisfactory than the standalone models. The numerical results showed that the BAT-ELM achieved the highest numerical performance with correlation coefficient and Nash-Sutcliffe efficiency ranging from 0.927 to 0.973 and from 0.846 to 0.944, respectively, much higher than the values obtained using the MLPNN, GRNN, SVR, GPR and ELM approaches, respectively. The obtained results demonstrated that the BAT-ELM presented an interesting alternative algorithm for predicting high extreme streamflow data.
28 April 2025 in Stochastic Environmental Research and Risk Assessment
This study aimed to improve daily streamflow forecasting by combining machine learning (ML) models with signal decomposition techniques. Four ML models were hybridized with five families of maximum overlap discrete wavelet transforms (MODWTs). The hybrid models were applied to predict daily streamflow at the Bir Ouled Taher station in northern Algeria. Model performance was evaluated using multiple statistical metrics and compared to standalone ML models. The hybrid MODWT-Gaussian process regression (GPR) model using Symlet wavelets (MODWT-GPR3 sym4) achieved the best performance, with R = 0.99 and NSE = 0.98 during validation. This significantly outperformed the standalone models tested and other hybrid combinations. The MODWT-GPR3 sym4 model demonstrated a superior ability to capture nonlinearities and predict peak flows. Hybridization of ML models with wavelet transforms, particularly the MODWT-GPR approach, can substantially improve daily streamflow prediction accuracy compared to standalone models. However, model performance may vary between watersheds due to differences in hydrological characteristics. Consideration of catchment concentration time when selecting model inputs could further enhance forecasting capabilities.