基于不同物理参数化方案的中小流域降雨径流集合预报

Ensemble rainfall-runoff forecasting based on different physical parameterization schemes for small and medium catchments

气象水文耦合预报能够延长洪水预报预见期,针对预报结果不确定性大的问题,选取东南沿海地区的梅溪流域为研究区,以2012年8月3日“苏拉”台风和2014年6月17日“海贝思”台风引发的降雨洪水为例,开展气象水文耦合下的集合预报研究。依托WRF(weather research and forecasting)模式建立基于36种物理参数化方案组合的降雨集合预报集,并通过耦合WRF模式和梅溪流域分布式水文模型,实现降雨径流集合预报。结果表明:在不同物理参数化方案下,数值降雨预报结果有一定差异,且对降雨空间分布的预报效果优于降雨时间分布,更容易准确描述时空分布均匀的降雨,很难捕捉短历时强降雨;采用集合预报的方式能够有效降低洪水预报的不确定性,当预见期超过6 h时,对于时空分布均匀的降雨,相应洪水过程的洪峰流量预报误差R_(f)为11.30%,能够准确反映洪峰量级,峰现时间提前2 h,相比基于“落地雨”开展的洪水预报有一定优势;基于异方差扩展型Logistic算法对预报降雨进行处理后,能够有效提高降雨预报精度,但对于时空分布不均匀的降雨,洪峰流量误差R_(f)由处理前的-86.89%降低至-48.95%,仍有较大的提升空间。

The ensemble method can efficiently reduce the uncertainty of rainfall forecast.The ensemble rainfall forecast is established through random disturbance of the initial field.Different numerical weather prediction models are used to form the forecasting ensemble.The ensemble rainfall forecast is build based on a combination of different physical parameterization schemes,which is regularly used for rainfall forecasting under unknown weather conditions.The selection of the physical parameterization scheme has a significant impact on forecasting results.A single physical parameterization scheme is difficult to adapt to different rainfall processes,which brings great uncertainty in the forecast.Ensemble forecast based on physical parameterization scheme can effectively reduce the uncertainty of rainfall forecast,which can provide reliable rainfall information for flood forecast.Thirty-six physical parameterization schemes based on the WRF model are used to establish the ensemble rainfall forecast.The relative error(E_(R)),critical success index(I_(CS)),and the root mean square error(E_(RMS))are used to comprehensively evaluate the rainfall forecast.Meixi distributed hydrological model is constructed based on China flash flood hydrological model(CNFF-HM).The peak flood discharge error,peak present time error,and Nash efficiency coefficient are used to evaluate the flood forecast.The coupled meteorological and hydrological system is formed by the WRF model and Meixi distributed hydrological model.The research also uses a statistical model that is developed based on the heteroscedastic extended Logistic algorithm to post-process the rainfall ensemble forecast results.For rainfall storms caused by Saola typhoon,the E Rs based on 36 schemes are between 0.88%and 21.00%.In spatial dimension,the I CS s are between 0.7368 and 0.7582,and the E RMS s are between 0.1331 and 0.2216.In the time dimension,the I CS s are both 0.6875 and the E RMS s are between 0.5924 and 0.7600,respectively.The error of peak flow discharge based on coupled meteorological and hydrological systems is 11.3%.With rainfall forecasting post-process,the error of peak flow discharge is 3.97%.Likewise,for rainfall storms caused by Hagibis typhoon,the E Rs based on 36 schemes are between 24.32%and 68.51%.In spatial dimension,the I CS s are between 0.3470 and 0.4879,and the E RMS s are between 0.5216 and 0.8451.In the time dimension,the I CS s are between 0.3292 and 0.4356,and the E RMS s are between 1.3001 and 1.6349,respectively.The error of peak flow discharge based on coupledmeteorological and hydrological systems is-86.89%.With rainfall forecasting post-process,the error of peak flow discharge is-48.95%.Forecasted rainfall in spatial dimension performs better than that in time dimension with different physical parameterizations schemes.The ensemble rainfall forecast is appropriately used for flood forecast with the coupled meteorological and hydrological system,which can efficiently reduce the forecasting uncertainty.Reasonable post-processing methods should be used to process the numerical rainfall forecast.For the rainfall with even spatiotemporal distribution,flood forecast with the coupled meteorological and hydrological system has certain advantages compared to flood forecast based on observed rainfall.For the rainfall with uneven spatiotemporal distribution,the forecast still has room for improvement.