IHACRES, GR4J and MISD-based multi conceptual-machine learning approach for rainfall-runoff modeling

Abstract

As a complex hydrological problem, rainfall-runoff (RR) modeling is of importance in runoff studies, water supply, irrigation issues, and environmental management. Among the variety of approaches for RR modeling, conceptual approaches use physical concepts and are appropriate methods for representation of the physics of the problem while may fail in competition with their advanced alternatives. Contrarily, machine learning approaches for RR modeling provide high computation ability however, they are based on the data characteristics and the physics of the problem cannot be completely understood. For the sake of overcoming the aforementioned deficiencies, this study coupled conceptual and machine learning approaches to establish a robust and more reliable RR model. To this end, three hydrological process-based models namely: IHACRES, GR4J, and MISD are applied for runoff simulating in a snow-covered basin in Switzerland and then, conceptual models’ outcomes together with more hydro-meteorological variables were incorporated into the model structure to construct multilayer perceptron (MLP) and support vector machine (SVM) models. At the final stage of the modeling procedure, the data fusion machine learning approach was implemented through using the outcomes of MLP and SVM models to develop two evolutionary models of fusion MLP and hybrid MLP-whale optimization algorithm (MLP-WOA). As a result of conceptual models, the IHACRES-based model better simulated the RR process in comparison to the GR4J, and MISD models. The effect of incorporating meteorological variables into the coupled hydrological process-based and machine learning models was also investigated where precipitation, wind speed, relative humidity, temperature and snow depth were added separately to each hydrological model. It is found that incorporating meteorological variables into the hydrological models increased the accuracy of the models in runoff simulation. Three different learning phases were successfully applied in the current study for improving runoff peak simulation accuracy. This study proved that phase one (only hydrological model) has a big error while phase three (coupling hydrological model by machine learning model) gave a minimum error in runoff estimation in a snow-covered catchment. The IHACRES-based MLP-WOA model with RMSE of 8.49 m3/s improved the performance of the ordinary IHACRES model by a factor of almost 27%. It can be considered as a satisfactory achievement in this study for runoff estimation through applying coupled conceptual-ML hydrological models. Recommended methodology in this study for RR modeling may motivate its application in alternative hydrological problems.

Assessment of hybrid machine learning algorithms using TRMM rainfall data for daily inflow forecasting in Três Marias Reservoir, eastern Brazil

“…The Multilayer Perceptron (MLP) is a widely used artificial neural network consisting of multiple interconnected layers of neurons. It includes input, output, and hidden layers and is trained using backpropagation [ 6 ]. The input layer receives signals, the output layer performs tasks such as prediction [ 38 ], and the hidden layers serve as the primary computational engine for the MLP [ 39 ].…”

“…R –R models can be categorized differently based on their spatial and structural processing methods, including physical, empirical, conceptual, distributed, lumped, and semi-distributed models. Each category possesses distinct advantages and drawbacks [ [6] , [7] , [8] , [9] ]. Moreover, climate change has complicated the rainfall-runoff relationship even further, leading to more nonlinear and intricate processes along with increased frequencies and severity of floods and droughts.…”

Modeling Various Drought Time Scales via a Merged Artificial Neural Network with a Firefly Algorithm

“…Consequently, machine learning (ML) systems for drought forecasting have attracted considerable interest [14,15]. In addition, several types of research have shown that AI algorithms outperform conventional approaches [16][17][18][19], such as artificial neural network (ANN) [20], support vector machines (SVMs) [21], random forests [22], and the adaptive neuro-fuzzy inference system (ANFIS) [23], which are examples of these ML systems. In recent decades, different developed types of ML approaches have been widely used for drought modeling tasks; for example, Inoubli et al [24] investigated the ability of the LSTM model for SPEI forecasting in East Azerbaijan province in Iran.…”

Short-term runoff forecasting in an alpine catchment with a long short-term memory neural network

“…Even though these modern data-driven approaches are showing promising results, they are only slowly tested on catchments in Switzerland. For instance, Mohammadi et al (2022) recently combined the outputs of three existing conceptual models used in Switzerland with a MLP neural network to estimate the river runoff in the Emme watershed in central Switzerland. Also, only a handful of studies have focused on leveraging the predictive power of neural networks for application in glacier-influenced catchments.…”