Geoscience Reference
In-Depth Information
communities to react, since the rain may not fall on the catchment that would affect them. In Vaison,
179 mm were recorded in 24 hours, with higher amounts elsewhere on the catchment and intensities of
up to 200 mm/hr during six-minute periods. The time between the start of the rainfall and the peak of
the flood was just 3.5 hours. The estimated peak discharge was 600-1100 m
3
/s, with a peak stage in the
town of Vaison of 21 m. The ancient Roman bridge in the centre of Vaison, was overtopped but survived
the flood, although there is impressive video footage of floating caravans being crushed beneath it. The
majority of the fatalities in this event were tourists at a campsite situated in the valley bottom upstream
of Vaison. There was much discussion after this event about whether development on the flood plain
upstream of the town had made the depth of flooding and the impact of the flood much worse (Arnaud-
Fassetta
et al.
, 1993). What is clear is that many lives could have been saved if an accurate flood warning
had been possible.
To make adequate warnings requires knowledge of the rainfalls as they occur or, even better, accurate
forecasts of potential rainfall intensities ahead of time, which would allow an increase in the forecast lead
time that might be important for small catchments and flash floods, such as that at Vaison-la-Romaine.
The advantage of weather radar in these situations is that the radar will often pick up the most intense
cells of rainfall in the weather system (e.g. Smith
et al.
, 1996). Such cells may be smaller than the spacing
between telemetering raingauges and might therefore be missed by ground-based systems. The problem
with radar, as an input to predicting the resulting flood discharges, is that the relationship between the
radar signal and the rainfall intensity may not always give an accurate estimate of the absolute intensity
(see Section 3.1), particularly when there are attenuation effects due to heavy rainfall close to the radar
masking the signal from further afield. Thus, for both radar and raingauge systems, it may be possible to
recognise that intense rainfall is occurring over a catchment area but not exactly how intense.
Forecasting rainfalls generally now involves a combination of projecting radar rainfall images into
the future and ensemble weather predictions of future precipitation (for example the UK Met Office
NIMROD radar and MOGREPS forecast products). Projection of radar images is easier with newer
Doppler radar systems that also provide information about local wind speed within the image but the
degradation of the radar projections is usually quite rapid because of the way in which individual rain
cells grow and decay in complex ways. This has also led to attempts to increase lead times using radar
by relating the development of the current pattern of rainfalls to historical analogues (e.g. Obled
et al.
,
2002). The MeteoSwiss Nowcasting Orographic Rainfall in the Alps (NORA) system is based on this
approach (Panziera and Germann, 2010).
A number of weather forecasting agencies are also now producing ensemble numerical weather pre-
dictions of precipitation that are being used for flood forecasting (see the recent review by Cloke and
Pappenberger, 2009). The European Centre for Medium Range Weather Forecasts (ECMWF), for ex-
ample, provides a control run and an ensemble of 50 “stochastic physics” runs up to 10 days ahead at a
grid scale of 40 km. The ensemble is re-initialised using a 4D variational data assimilation methodology
and re-run every day. The ECMWF have also run reanalyses of past periods that can be used in “hind-
casting” studies. A past reanalysis, ERA40, covers the period from 1957 but the most recent reanalysis,
ERA-Interim, which uses an updated forecasting model, started in 1989.
The ECMWF forecasts are used in the European Flood Alert System (EFAS) that is run at the European
Community Joint Research Centre in Ispra, Italy (see http://floods.jrc.ec.europa.eu/efas-flood-forecasts).
This system is run operationally to provide flood alerts for larger basins in Europe with lead times from
three to 10 days. The ECMWF flood rainfalls are processed through a version of the LISFLOOD rainfall-
runoff model to provide forecasts of river flow on a 5 km grid (see Thielen
et al.
, 2009; Bartholmes
et al.
,
2009). Although some calibration of the LISFLOOD model has been carried out, the EFAS system
recognises that there may be uncertainty in predicting runoff and uses a system of alerts based on the
current situation at any grid square relative to the ERA40-reanalysis-driven LISFLOOD simulations. This
gives an estimate of the number of ensemble members that suggest extreme flows in a river relative to the
reanalysis period. The US National Weather Service is making more direct operational use of ensemble
