Geoscience Reference
In-Depth Information
propagation between sub-catchments were effectively averaged out over larger catchment
areas.
Despite continuing incremental advances in radar technology and performance during the
1970s and 1980s, a number of authors recognized that radar derived estimates of surface
precipitation rate and accumulation would remain subject to hydrologically significant
errors, particularly in hilly and mountainous areas. Collier and Knowles (1986) concluded
that the full benefits of radar to operational hydrology would only be realized when ways
could be found of accounting for these errors.
This thinking coincided with a growing awareness of the need to develop operational
systems integrating meteorological and hydrological forecast models (Georgakakos and
Kavvas, 1987). Early examples of such systems are the Integrated Flood Observing and
Warning System (IFLOWS) implemented in the USA (Barrett and Monro, 1981), and the
Regional Communication Scheme in north-west England, integrating operational weather
radar data with hydrological forecasting and warning under the North-West radar project
(Noonan, 1987).
More recently, a number of multi-national initiatives including HEPEX (e.g. Buizza, 2008;
Pappenberger et al., 2008), MAP-D-Phase (Zappa et al., 2008; Bogner & Calas, 2008) and
COST-731 (Rossa et al., 2011) have supported work to implement integrated flood
forecasting systems exploiting weather radar. A number of UK-based research programmes,
are relevant in this context, including HYREX (e.g. Mellor et al., 2000a,b; Bell & Moore,
2000a), the Natural Environment Research Council's Flood Risk from Extreme Events
(FREE), the Engineering and Physical Sciences Research Council's Flood Risk Management
Research Consortium (FRMRC) and FLOODsite.
Until Krzysztofowicz's pioneering work (Krzysztofowicz, 1983, 1993, 1998, 1999, 2001) to
develop and implement an integrated hydro-meteorological systems framework,
incorporating a Bayesian treatment of uncertainties in data inputs and deterministic model
forecasts, errors in the precipitation inputs to operational hydrological models tended to be
handled through the use of what-if scenarios (Haggett, 1986; Werner et al., 2009). The
derived distribution approach (Seo et al., 2000) developed by Krzysztofowicz exploits the
total probability law to derive by quasi-analytic means the conditional probability
distribution of river stage given the initial and boundary conditions, including future
precipitation parameterized in the form of a probabilistic QPF.
An alternative method for accounting for data input uncertainty entails the ingestion of
ensemble precipitation forecasts into a hydrological model, whilst taking separate account
of other sources of uncertainty contributing to the total uncertainty in the hydrological
forecast variable of interest (Krzysztofowicz, 2001). Probabilities of exceeding specific river
stage thresholds can then be estimated from the resulting ensemble of hydrographs
(Schaake & Larson, 1998; Pierce et al., 2004). During the past decade, this ensemble approach
has gained credence with the widespread implementation of operational ensemble NWP
models and the development of stochastic nowcasting and post-processing techniques for
the production of ensembles of high resolution precipitation nowcasts (e.g. Bowler et al.,
2006).
