Support Vector Regression (SVR) combined with Invasive Weeds Optimization (IWO), standalone SVR, and Radial Basis Function Neural Networks are applied to estimate channel sinuosity in perennial rivers. With this aim, a dataset with 132 sinuosity data and related geomorphologic data, corresponding to 119 perennial streams, is considered. Bayesian Mutual Information theory is used to determine the parameters affecting channel sinuosity to reveal that bankfull depth affects sinuosity the most. Seven input parameter combinations for sinuosity prediction are considered, and in both training and testing stages, the SVR-IWO model \(\left( {R_{Train} = 0.959,RMSE_{Train} = 0.072, MAE_{Train} = 0.037, R_{test} = 0.892, RMSE_{Test} = 0.103, MAE_{Test} = 0.065} \right)\) shows the best prediction performance while the standalone SVR model generated the results with performances of \(\left( {R_{Train} = 0.792,RMSE_{Train} = 0.158, MAE_{Train} = 0.141, R_{test} = 0.704, RMSE_{Test} = 0.163, MAE_{Test} = 0.151} \right)\). Model prediction uncertainty is quantified in terms of entropy for the three models considered, further confirming that the sinuosity set predicted by the SVR-IWO model is the closest to the observed set.

支持向量回归 (SVR) 与入侵杂草优化 (IWO)、独立 SVR 和径向基函数神经网络相结合，用于估计多年生河流中的河道曲折度。为此，考虑了一个包含 132 个曲率数据和相关地貌数据的数据集，对应于 119 条多年生河流。利用贝叶斯互信息理论确定影响河道曲率的参数，揭示岸边深度对曲率的影响最大。考虑了曲率预测的七个输入参数组合，在训练和测试阶段，SVR-IWO模型\(\left( {R_{Train} = 0.959,RMSE_{Train} = 0.072, MAE_{Train} = 0.037, R_{test} = 0.892, RMSE_{Test} = 0.103, MAE_{Test} = 0.065} \right)\)显示最佳预测性能，而独立 SVR 模型生成的结果具有\(\left( {R_{Train} = 0.792,RMSE_{Train} = 0.158, MAE_{Train} = 0.141, R_{test} = 0.704, RMSE_{Test} = 0.163, MAE_{Test} = 0.151} \right)\)。模型预测的不确定性是根据所考虑的三个模型的熵来量化的，进一步证实了 SVR-IWO 模型预测的曲率集与观测集最接近。