Geoscience Reference
In-Depth Information
by reducing reliance on crude assumptions regarding future precipitation. For pluvial
(surface water) flood forecasting, predictions of future precipitation are essential because the
time between the precipitation reaching the ground and any consequent flooding is very
short (Golding, 2009). In this section we review some of the key developments in the use of
radar for fluvial (river) flood prediction and warning.
6.2 Hydrological requirements for precipitation observations
It was the prospect of accurate, contiguous observations of precipitation over large areas
that first stimulated hydrologists to explore the use of radar data for the prediction of run-
off and river flow. Early assessments of the value of radar data were mixed (Anderl et al.,
1976; Barge et al., 1979). This is not surprising given the reliance of these early experiments
on deterministic precipitation estimates of variable accuracy, and the many factors known to
impact on hydrological forecast performance.
Hydrological requirements for precipitation observations and forecasts are a function of
catchment size, morphology and land use, and the hydrological model used (Hudlow et al.,
1981). Many operational, hydrological forecasting models are lumped conceptual models in
which the catchment response is modelled as a whole and the precipitation input is an areal
average estimate. A number of authors have emphasized that the benefits of radar derived,
spatially contiguous precipitation estimates can only be fully realized if used as input to
distributed, conceptual or physically-based hydrological models (e.g. Moore, 1987).
6.3 Impact of the spatial and temporal distribution of precipitation
Wilson et al. (1979) found that a failure to properly represent the spatial distribution of
rainfall, due to reliance on point observations from rain gauges, could produce significant
errors in the total volume, peak and time to peak of an estimated hydrograph, even when
the rainfall depth and its temporal evolution were accurately recorded at rain gauge sites.
Errors were largest in cases of localized convective storms. Bedient and Springer (1979)
demonstrated that the peak flows in a catchment could be enhanced when the precipitation
moved in the direction of the stream.
More recently, Bell and Moore (2000b) explored the sensitivity of lumped and distributed
catchment rainfall-run-off models to time series of rainfall observations from radar and rain
gauge, gridded to a range of spatial resolutions. For a small rural catchment, they confirmed
the sensitivity of distributed model run-off to the spatial variability of rainfall. A
comparison of the performances of lumped and distributed models showed similar levels of
predictive skill in stratiform rain, but superior distributed model predictions during
convective rainfall events.
Ball (1994) examined the impact of the temporal evolution of the precipitation pattern on the
time of concentration and peak discharge in a catchment. The time of concentration was
shown to be sensitive to the temporal evolution of excess rainfall over the catchment, where
as catchment peak discharge was not. Thus, timing errors in predicted flows can result if the
time interval between precipitation observations is too long. Collier (1996) suggests that a
radar scan cycle of no more than 5 minutes is required to capture the time evolution of most
convective precipitation fields.
