A new method for runoff prediction error correction based on LS-SVM and a 4D copula joint distribution.

• We combined the LS-SVM model with the 4D-copula model to jointly correct the errors in different forecast periods. • After preprocessing the initial error, the final correction result was better. • The Martingale model was used to statistically analyze the bias between before and after the correction. The nonlinear relationship between runoff and time is a major challenge in hydrological forecasting. It is important to improve the prediction accuracy of models for disaster mitigation, thus allowing decision makers to make decisions in advance. This paper proposes a method called joint error correction. First, we cluster the initial prediction errors. Then, the moving average method is used to smooth the classified error series. Finally, we establish a 4D copula function model for errors in different forecasting periods. After a sampling test, we compared the errors before and after the correction. We found the system corrected the initial error of the Least Squares Support Vector Machines (LS-SVM). The overestimation of flood in LS-SVM model is reduced. Taking the runoff data from the Fu Yu hydrological station from 2004 to 2018 as an example, the following conclusions were drawn: 1) The correction effect of errors after pretreatment is better than that without pretreatment. Compared with the initial error, the NSE values of the revised models in four different forecasting periods increased by 8%, 5%, 5% and 16%, and the decreases in the RMSE values were 39.37%, 12.25%, 10.38% and 32.6%. 2) After the initial error was corrected, the errors in ω t,- were correlated with each other, and the variables in ω -,h were independent of each other. After preprocessing the initial values, the joint error improvement value groups ω t,- and ω -,h displayed independent characteristics. 3) The revised forecast was unbiased. Therefore, the overestimated actual runoff predicted was corrected to some extent. [ABSTRACT FROM AUTHOR]