Can sampling techniques improve the performance of decomposition-based hydrological prediction models? Exploration of some comparative experiments

Abstract The development of sequence decomposition techniques in recent years has facilitated the wide use decomposition-based prediction models hydrological forecasting. However, usually overall (OD) sampling technique to extract samples. Some studies have shown that OD causes abnormally “high” performance owing utilization future information, and this cannot be applied practice. Several researchers also proposed novel techniques, such as semi-stepwise (SSD), fully stepwise (FSD), single-model SSD (SMSSD). Moreover, an improved FSD (SMFSD) is study. Four methods, namely discrete wavelet transform (DWT), empirical mode (EMD), complete ensemble with adaptive noise (CEEMDAN), variational (VMD), are introduced A systematic investigation developed using conducted, applicability SSD, FSD, SMSSD, SMFSD reasonably evaluated. application monthly runoff five four methods at representative stations Poyang Lake, China, shows (1) EMD CEEMDAN (including EMD-based method) used construct because implementation strategy leads a variable number sub-series. (2) develop convincing for practical information into samples model training. (3) Models based on SMSSD do not training process, but suffer from severe overfitting inferior performance. (4) can produce results, combination VMD develops superior significantly enhances efficiency models.