WEATHER RADAR FOR URBAN PLUVIAL FLOOD FORECASTING Professor Chris Collier National Centre for Atmospheric Science, Head of Strategic Partnerships University of Leeds, UK
Oct 30, 2014
WEATHER RADAR FOR URBAN PLUVIAL FLOOD FORECASTINGProfessor Chris CollierNational Centre for Atmospheric Science, Head of Strategic PartnershipsUniversity of Leeds, UK
Impact of Flash Floods in Cities
Commercial district of Istanbul, September 2009At least 20 people died in Istanbul
7 drowned in a minibus going to work
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.
Flood protection and forecasting
Why radar?• Wide area measurements of precipitation from a single location.• High resolution.
19:35
mm/hr
< 0.125
0.25
0.5
1
2
4
8
16
32
> 64
High resolution (1 km) radar imagery 7 May 2000
Courtesy Met OfficeGreen on this map of Hull UK indicates areas that are prone to flooding
Rainfall totals measured at Ruislip, London and discharge from Yeading Brook West Branch on 8 May 1988
Bank full
1 in 100 yrs 1 in 25 yrs63.5 mm in 2.5 h 34.2 mm in 75 min
How radar works
Courtesy Met Office
The passage of line convection over London as observed by the Chenies C-band radar 7 December 2006, 1053UTC
Chenies
tornado
Courtesy Met Office
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.
Advantages Disadvantages• Attenuation through rain,
snow and ice (hail) but can correct if polarisation capability exists.
• Very limited clear air measurements.
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
Examples of mobile Doppler dual polarisation X-band radar
Selex Gematronik University of Auckland, NZ, Ardmore
Why do (some) hydrologists still distrust radar estimates of rainfall? Comparison with raingauges
Rainfall rate (mm h-1) on 21 June 2004, around 9:48 UTC, given by the Hameldon Hill C-band radar located some 24 km north
of the centre of Manchester, North West England.
Study domainStudy domain
Wind
Study domainStudy domain
Wind
Study domainStudy domain
WindWind
The red and white dot indicates Manchester city centre. This image is an example of the radar product used in this work (10 m in image, with 2 x 2 km2 spatial resolution).
Why we need to merge rainfall data?
0
5
10
15
20
25
00:05 01:05 02:05 03:05 04:05 05:05 06:05 07:05
Rain
Dep
th (
mm
)
Time (5 min)
Cumulative Rain Depth (23/08/2010 event)@Beal RG
Beal_RG Radar 1km
Amplification of radar errorsDischarge 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
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
1 1.1 1.2 1.3
Rainfall bias
Dis
char
ge b
ias
Croal August-September
January frontal
October squall line
gap
Fundamental Limitation of Widely Spaced Long Range Radars
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.
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]
The problem of issuing an alert under flood forecasting uncertainty
Expected value= ForecastProbability of overflowLevelCostsCross section
(courtesy E. Todini)
Expected value= Forecast
Probability of overflow
Lev
el
Costs
Cross section
(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)
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.