Abstract:In this study, a novel nonlinear segmental runoff ensemble forecast model based on the Bayesian model averaging (BMA) algorithm (NLTM-BMA m (P-III)) is proposed based on multimodel ensemble prediction for forecasting monthly runoff and quantifying forecast uncertainty. Four nonlinear time series models were used as ensemble members, and runoff segmented intervals were divided based on P-III type hydrological frequency curves. On this basis, the BMA algorithm was used to obtain the weight sets of each interval after the Box‒Cox transformation. Finally, the mean and probability forecasts were obtained using the weighted average method and the Monte Carlo method. The model was applied to monthly runoff forecasts at eight hydrological stations in the Hei River Basin and two hydrological stations in the Wei River Basin; and compared with the whole-segment simple averaging model NLTM-SMA, the whole-segment Bayesian averaging model NLTM-BMA 1 and the segmented Bayesian averaging model with normal distribution partitioning NLTM-BMA m (Normal). The results show that (1) the BMA algorithm yields more reliable forecasts than the SMA algorithm, (2) Segmentation criteria appropriate for the runoff distribution can improve the forecasting accuracy, which would otherwise be reduced, and (3) Compared with the NLTM-SMA and NLTM-BMA 1 models, the NLTM-BMA m (P-III) model yields a higher CR value, demonstrating that the segmented ensemble forecasting model can improve the accuracy of probability prediction by considering the diversity of ensemble members. Additionally, the BMA algorithm has good applicability in the segmented ensemble model. The model provides a new method for medium- and long-term runoff forecasting.

On buffered moving average models

Smooth Transition Moving Average Models: Estimation, Testing, and Computation

A Note on Moving‐Average Models with Feedback

Simulation and Application of Subsampling for Threshold Autoregressive Moving-Average Models.

A Vector Autoregressive Moving Average Model for Interval-Valued Time Series Data

Penalized Time-Varying Model Averaging

Comparative Analysis of Properties of Weakening Buffer Operators in Time Series Prediction Models.

Threshold Autoregressive Models for Interval-Valued Time Series Data

Temporal Aggregation of Markov-switching Financial Return Models

A Class of Time-Varying Vector Moving Average Models: Nonparametric Kernel Estimation and Application

On a Threshold Double Autoregressive Model

Shrinkage Estimation of Multiple Threshold Factor Models

On Moving-Average Models with Feedback

Penalised inference for autoregressive moving average models with time-dependent predictors

Improving on the Markov-Switching Regression Model by the Use of an Adaptive Moving Average

Robust Bayesian model averaging for the analysis of presence–absence data

Functional Coefficient Moving Average Model with Applications to Forecasting Chinese CPI

Application of Bayesian model averaging for modeling time headway distribution

Best Subset Selection for Double-Threshold-Variable Autoregressive Moving-Average Models: The Bayesian Approach

Adaptive Dynamic Model Averaging with an Application to House Price Forecasting

Nonlinear Segmental Runoff Ensemble Prediction Model Using BMA