1. W EATHER R ADAR FOR URBAN PLUVIAL FLOOD FORECASTING Professor Chris Collier National Centre for Atmospheric Science, Head of Strategic Partnerships University of Leeds, UK

2. Impact of Flash Floods in Cities Commercial district of Istanbul, September 2009 At least 20 people died in Istanbul 7 drowned in a minibus going to work

3. Urban drainage In many urban areas of England the UDS is complex, and in parts old and in need of refurbishment. Sewage discharges to natural water courses Accurate high resolution (1km x 1km) rainfall measurements and forecasts needed. Changes in rainfall patterns and amounts may cause problems in UDS management.

4. Flood protection and forecasting

5. Why radar? Wide area measurements of precipitation from a single location. High resolution. Courtesy Met Office Green on this map of Hull UK indicates areas that are prone to flooding

6. Rainfall totals measured at Ruislip, London and discharge from Yeading Brook West Branch on 8 May 1988 Bank full 1 in 100 yrs1 in 25 yrs 63.5 mm in 2.5 h34.2 mm in 75 min

7. How radar works Courtesy Met Office

8. The passage of line convection over London as observed by the Chenies C-band radar 7 December 2006, 1053UTC Chenies tornado Courtesy Met Office

9. X-band radar Ease of siting Cost Mobility Less ground clutter provided one degree beamwidth used. Can detect smaller particles including the detection of light precipitation such as snow. AdvantagesDisadvantages Attenuation through rain, snow and ice (hail) but can correct if polarisation capability exists. Very limited clear air measurements.

10. Polarization techniques offer increased accuracy for measuring heavy rain 60 dBZ core could be torrential rain or hail Conventional Radar Reflectivity Differential Phase Shift Phase shift indicates torrential rain Rain gauge confirmed 250mm/hour

11. Examples of mobile Doppler dual polarisation X-band radar Selex Gematronik University of Auckland, NZ, Ardmore

12. Why do (some) hydrologists still distrust radar estimates of rainfall? Comparison with raingauges

13. Why we need to merge rainfall data?

14. Amplification of radar errors Discharge bias using radar data input to a stochastic model of the urban River Croal, UK catchment compared to a model of the Baron Fork River USA catchment

15. gap Fundamental Limitation of Widely Spaced Long Range Radars

17. High resolution numerical forecasts 1-2 km grid lengths now beginning to be used operationally. Realistic forecasts now being produced, but problems remain e.g. Representing sub-grid scale processes, although grid lengths of less than 1 km are also possible. These forecasts are expected to replace radar-based nowcasts for lead times beyond two hours or so. However the assimilation of radar data is likely to become an essential part of operational procedures.

18. Illustrating Cobbacombe radar 5 hour total rainfall (mm) (left panel) and 1 km UM forecast rainfall (mm) for 12-17 UTC 16 August 2004 (from Golding et al, 2005) [performance due in part to the dynamic impact of the sea breeze with orography which introduced a level of stationarity to the convection]

19. The problem of issuing an alert under flood forecasting uncertainty (courtesy E. Todini)

20. (a) (b) (c) Example of the exploitation of new data sources, data assimilation and ensemble techniques for storm and flood forecasting Boscastle storm Case study: Boscastle storm (a) a ‘pseudo-ensemble’ of high-resolution 1 km NWP rainfall, (b) an ensemble of distributed hydrological model simulations of river flow using the Grid-to-Grid (G2G) model, (c) comparison of G2G ensembles with observations for the River Tamar at Gunnislake (location and 1 km catchment boundary is given in (a) and (b)). (courtesy R. Moore and S. Cole)

21. Concluding remarks Radar data are likely to be the basis of forecasts for 1-2 hours ahead. However for longer lead times high resolution NWP forecasts assimilating radar data, offer the best hope of improvement to hydraulic and hydrological forecasts. It will be necessary to constrain uncertainty using both rainfall and hydrological model ensembles with statistical procedures. Rain Gain will produce a significant step forward in using X- band radar data within the context of operational radar networks.