Abstract Due to the complex interaction between the fluvial and tidal dynamics, estuarine tides are less predictable than ocean tides. Although the non-stationary tidal harmonic analysis (NS_TIDE) model can account for the influence of the river discharge, the predictive accuracy of the water levels in the tide-affected estuaries is yet to be improved. The results from recent studies using the NS_TIDE model in the lower reach of the Yangtze estuary showed the best root-mean-square-error (RMSE) between the predicted and measured water levels being in a range of 0.22 ~ 0.26 m. From the spectral analysis of the predictive errors, it was also found that the inaccurate description of tides in the sub-tidal frequency band was the main cause. This study is to develop a hybrid model in combination of the auto-regressive (AR) analysis and the NS_TIDE model in an attempt to further improve short-term (with time scale of days) water level predictions in the tide-affected estuaries. The results of the application of the hybrid model in the Yangtze estuary show a significant improvement for water level predictions in the estuary with the RMSE of 24 h prediction being reduced to 0.10 ~ 0.13 m.

中文翻译：

---

自回归（AR）分析在长江口水位短期预测中的应用

摘要 由于河流和潮汐动力学之间复杂的相互作用，河口潮汐的可预测性不如海洋潮汐。尽管非平稳潮汐谐波分析（NS_TIDE）模型可以解释河流流量的影响，但受潮影响的河口水位的预测精度仍有待提高。最近在长江口下游使用 NS_TIDE 模型的研究结果表明，预测和实测水位之间的最佳均方根误差 (RMSE) 在 0.22 ~ 0.26 m 范围内。从预测误差的谱分析中还发现，潮下频段的潮汐描述不准确是主要原因。本研究旨在开发一种结合自回归 (AR) 分析和 NS_TIDE 模型的混合模型，以尝试进一步改进受潮汐影响的河口的短期（以天为时间尺度）水位预测。混合模型在长江口的应用结果表明，长江口水位预测显着提高，24 h预测的RMSE降低到0.10～0.13 m。