River basin flood forecasting by coupling quantitative precipitation prediction with distributed hydrological model: a case study in Pearl River Basin in southern China

Improving flood forecasting capability, including extends lead time and increases accuracy has long been the goal of both hydrological and meteorological communities all over the world. Recent developments in distributed hydrological models and quantitative precipitation estimation and prediction provide a new way for fulfilling this goal. This paper has done some preliminary works on coupling the WRF predicted precipitation with Liuxihe Model for the flood forecasting of the Pearl River basin in Southern China. Pearl River is the largest river in southern China and the second largest river in China based on the annual water quantity, which has a more than 2000km length and a nearly half million km2 drainage area, and flooding is continuously a serious natural disaster threatening the basin area, particularly the downstream area where stands the most developed area in China.

The WRF weather model is first developed to predict the precipitation over the whole basin with a spatial resolution at 20km grid, and a temporal resolution at 1 hour with 72 hour lead time. Liuxihe model, a physically based, distributed hydrological model is then developed to predict the flood over the basin by having the WRF predicted precipitation as the model driver. Three different scale Liuxihe Models have been proposed to validate the capability of the method. The first Liuxihe Model covers the whole basin area with a spatial resolution at 20km grid to couple the WRF predicted precipitation, which is called large scale coarse Liuxihe Model. The second Liuxihe Model only covers a first order branch of Pearl River, the Liujiang River with a drainage area of around 60,000km2, and the spatial resolution is at 1km grid, while the temporal resolution is also at 1 hour, and the precipitation coupled is assimilated from the 20km grid WRF predicted precipitation using the most nearest interpolation method, and this model is called the medium scale Liuxihe Model. The third Liuxihe Model only covers a first order branch of Liujiang River with a drainage area of around 1,000km2, the spatial resolution is at 100m grid, while the temporal resolution is still at 1 hour. Two different precipitation have been tested, the one is assimilated from the 20km grid WRF predicted precipitation, while the other is interpolated from the raingauge observed precipitation, this model is called as small scale fine Liuxihe Model.

The model parameters of the small scale fine Liuxihe Model have been optimized by using the observed flood events with rangauge observed precipitation, and validated with raingauge interpolated precipitation and WRF predicted precipitation. The results show that the flood simulation accuracy with raingauge interpolated precipitation is acceptable and higher than that with WRF predicted precipitation, although with a shorter leading time, which implies that the accuracy of WRF precipitation prediction is still need to be improved. The simulation results of the medium scale Liuxihe Model show that the flood simulation accuracy with both precipitation are not so good but reasonable, which implies that not only the accuracy of WRF predicted precipitation needs to be improved, but the Liuxihe Model's performance in large river basin also needs to be improved, and the coarse spatial resolution may be the key factor to have impacted the model performance. The results in the large scale coarse Liuxihe Model is not so good and much improvement needs to be done before it is used for flood forecasting.

This study has made some preliminary studies on coupling the quantitative precipitation forecasting with hydrological model for real-time flood forecasting, the results show it is promising to couple the WRF predicted quantitative precipitation with Liuxihe Model for real-time flood forecasting, but there are still lots to do for scientific communities.