Evaluation of Quantitative Precipitation Predictions by ECMWF, CMA, and UKMO for Flood Forecasting: Application to Two Basins in China

Abstract

Numerical weather predictions (NWPs) are very useful in hydrological modeling, including for river flow forecasting and flood warning in river basins. However, uncertainties in NWPs also significantly impact the accuracy of streamflow forecasting. Therefore, evaluating the accuracy of NWPs is crucial to achieve reliable streamflow forecasts. In this study, rainfall prediction skills of three NWP models [developed by the European Centre for Medium-Range Weather Forecasts (ECMWF); the U.K. Meteorological Office (UKMO); and the China Meteorological Administration (CMA)] are evaluated in two basins (Linxian and Jiuzhaigou) in China, which have different hydroclimatic, topographic, and other characteristics. The evaluation is made by comparing the model predictions with measurements of ground-based rain gauges during the flood seasons (May to October) during 2011-2013. Four different evaluation measures are used: the confusion matrix, correlation coefficient, Nash-Sutcliffe efficiency, and root-mean square error. The influence of rainfall station representativeness (i.e., location and density of rain gauges in the basin) is also analyzed in detail. The results show that ECMWF has the highest skill in precipitation forecast over the two studied basins, followed by UKMO and CMA. The performance of UKMO is also found to be very close to that of ECMWF. The results also indicate that the precipitation prediction of each of the three models is better for the Linxian Basin when compared to that for the Jiuzhaigou Basin. The present results have important implications for the use of NWP data in hydrological modeling, especially for flood forecasting.

An overview of the ORACLES (ObseRvations of Aerosols above CLouds and their intEractionS) project: aerosol–cloud–radiation interactions in the southeast Atlantic basin

“…The ORACLES team decided on separate deployments in September 2016, August 2017, and October 2018, a decision that was aided by a relative lack of interannual variability in meteorology. This variability was predominantly linked to SST variations known as Benguela Niños that mainly occur in boreal spring, not fall, and are much less frequent than the better-known Pacific El Niños (Rouault, 2012). Interannual variability in fire emissions was expected to be low as well (van der Werf et al, 2010).…”

An overview of the ORACLES (ObseRvations of Aerosols above CLouds and their intEractionS) project: aerosol-cloud-radiation interactions in the Southeast Atlantic basin

“…the radiation measurements of some instruments (e.g., 4STAR, SSFR). Verification studies prior to the ORACLES 365 deployments showed that the European Centre for Medium Range Weather Forecasts (ECMWF;Pappenberger et al, 2008;ECMWF Newsletter, 2012;Ye et al, 2014) and United Kingdom Meteorological Office (UKMO; Ran et al, 2018) global forecast models provided the best performance for cloud forecasts. ECMWF digital data were available at 0.125° longitude x latitude resolution, and included the primary meteorological variables (relative humidity and horizontal winds at 925, 850, 800, 700, 600, 500, 400, 300, 150, and 100 hPa levels; 1000-500 hPa layer thickness; 370 surface wind speed; mean sea level pressure; boundary layer height; precipitation; convective available potential energy) as well as 3-D ice and liquid water mass.…”

Flood Forecasting System Based on Integrated Big and Crowdsource Data by Using Machine Learning Techniques

“…These characteristics thus called for big and not conventional data handling. It would be later shown in the subsequent experiments that the rainfalls big data forecasted by ECMWF model yielded the most accurate results [61].…”

“…In terms of mathematical models, it was found that current frameworks made their flood forecast based on three models, i.e., meteorological, hydrological, and ML ones. The first two models were used with many algorithms, e.g., HEC-HMS [57], HEC-RAS [59], Mike11 [60], Mike21 [60], Mike-Flood [60], and ECMWF [61], etc., while those using ML model were implemented with various methods (as discussed in the section I. E). Thus far, these systems were only experimental and, to our knowledge, not yet publicly distributed.…”