assessing soft computing techniques for river suspended sediment estimation

Objective: statistical and regression models are commonly used for this purpose, but they often yield inaccurate results due to the linear assumptions inherent in these approaches. hydraulic models, while useful, cannot always be fully trusted due to their requirement for extensive data, the potential unavailability of required data, and the possibility of human errors leading to inaccuracies in sediment simulation. to overcome these challenges, data-driven methods—specifically a subset of soft computing techniques—can be employed. material and methods: this study aims to evaluate and compare five methods for estimating the sediment load at the pol doab station on the qarachai river in markazi province, iran. the methods considered include the adaptive network-based fuzzy inference system (anfis), support vector machine (svm), gene expression programming (gep), multivariate adaptive regression splines (mars), and the group method of data handling (gmdh). the performance of these models in simulating the sediment load of rivers was assessed, and the results were compared both among the five methods and with those obtained from the sediment rating curve. results and discussion: the statistical indicators showed the following results for each method at the station: for the svr model: r2=0.98r2 = 0.98r2=0.98, rmse = 185, and mbe = -5.43. for the anfis model: r2=0.79r2 = 0.79r2=0.79, rmse = 24.3, and mbe = 1.9. for the gep model: r2=0.98r2 = 0.98r2=0.98, rmse = 0.74, and mbe = 0.00047. in the next step, the best-performing patterns from the anfis, svm, and gep models were used as inputs for the gmdh model. the results indicated that the gmdh model demonstrated the highest performance, with r2=0.99, 0.91, 0.98 r2 = 0.99, 0.91, 0.98r2=0.99, 0.91, 0.98, rmse = 83, 24, 73 (tons per day), and mbe = 3.2, 1.8, 1.2 respectively.conclusion: the findings suggest that the gep model, with r2=0.98r2 = 0.98r2=0.98, rmse = 0.74, and mbe = 0.00047, outperforms the svm, anfis, and gmdh models to some extent. moreover, the results demonstrate that all five data mining methods investigated in this study provide significantly better estimates than the traditional sediment rating curve. sediment load along with river flow causes irreparable damage to water development projects. estimation of river sediment load is an important and practical issue in the study and design of water and hydraulic projects. the purpose of this research is to evaluate and compare adaptive neural-fuzzy models (anfis), (svm), (gep), (gmdh) and (mars) and compare with the (src) method in estimating sediment load of pol doab station of qarachay river, markazi province. for this purpose, the performance of 5 types of data mining models in simulating river sediment load was investigated, then the results of the 5 methods were compared with each other and with the results of the scale curve method. the results indicate the acceptable performance of data mining models compared to the scale curve. the results also showed that the gep model with r2=0.98, rmse=0.74 and mbe=0.00047 has better performance than the svm, anfis, mars and gmdh models. the src method had the lowest r square value 0.61 and average rmse 75 and mbe 20. in general, all five data mining methods showed better performance than the src sediment ranking curve.