1B.1 Global Flash Flood Prediction using the ECMWF Model

Monday, 23 January 2017: 11:00 AM
602 (Washington State Convention Center )
Tim D. Hewson, ECMWF, Reading,, United Kingdom; and F. M. Pillosu

In 2016 ECMWF (the European Centre for Medium range Weather Forecasts) embarked upon a new project to post-process gridbox rainfall forecasts from its ensemble prediction system, to provide probabilistic forecasts of point rainfall. The new post-processing strategy relies on understanding how different rainfall generation mechanisms lead to (a) different degrees of sub-grid variability in rainfall totals, and (b) different biases on the gridbox scale. We use a number of simple global model parameters, such as the convective rainfall fraction, to anticipate (a) and (b), and then post-process each ensemble forecast into a pdf (probability density function) for a point-rainfall total. The final forecast will comprise the sum of the different pdfs from all ensemble members. The post-processing is essentially a re-calibration exercise, which needs only rainfall totals from standard global reporting stations (and forecasts) to train it. High density observations are not needed.

This presentation will highlight which model and geographical parameters can provide “predictive skill”. By skill we mean an ability to associate different behaviour in (a) and (b) with different meteorological/geographical scenarios. Model parameters used include standard model output, and newly created diagnostics that focus on addressing known limitations of a global model. Standard output includes convective rainfall fraction, mid tropospheric wind speed and clear sky solar radiation, whilst new diagnostics include a cell drift parameter, to compensate for the non-advection of convective cells, and also topographic modulation parameters. Geographical parameters we have utilised include the land fraction and the standard deviation of the sub-grid orography. Reference will also be made to flash flood case studies which have informed the investigations.

The main practical benefit of this system will be a greatly improved capacity to predict extreme point rainfall, and thereby provide early warnings, for the whole world, of flash flood potential for lead times that extend beyond day 5. This is being incorporated into the suite of products output by GLOFAS (the GLObal Flood Awareness System) which is hosted at ECMWF. As such this work offers a very cost-effective approach to satisfying user needs right around the world. This field has hitherto relied on using very expensive high-resolution ensembles; by their very nature these can only run over small regions, and only for lead times up to about 2 days.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner