Geoscience Reference
In-Depth Information
made using average past rainfall and an assumption of no more rain. In most cases, they
found that nowcast driven hydrological forecasts outperformed the alternatives, although
on one occasion they showed the former to be poor.
Several decades later, a similar, case study orientated evalution of the utility of Nimrod
(Golding, 1998) extrapolation nowcasts for rainfall-run-off modelling in Scotland (Werner
and Cranston, 2009) drew similar conclusions: although errors in nowcast driven
predictions of river flows could be substantial, they were smaller than those of flow
forecasts made assuming zero future rainfall.
Mecklenburg et al. (2001) found that COTREC-based radar extrapolation nowcasts
(Lagrangian persistence) produced superior hydrological forecasts to Eulerian persistence
using a lumped conceptual model. In a similar vein, Berenguer et al. (2005) compared
hydrological forecasts made with the S-PROG model (Seed, 2003) with those produced
using a simpler, extrapolation-based precipitation nowcast in a Mediterranean environment.
S-PROG utilizes a scale decomposition framework and associated hierarchy of auto-
regressive models to smooth the advected precipitation field at a rate that is consistent with
its loss of predictive skill on a hierarchy of scales. This approach is intended to minimize the
root mean squared forecast error. Berenguer et al. (2005) concluded that radar-based
precipitation nowcasts in general could extend the lead time of useful hydrological forecasts
from 10 minutes to over an hour in a fast response responding Mediterranean catchment.
However, the results obtained with S-PROG were not significantly better than those
obtained with a simpler Lagrangian persistence technique.
Since one of the key benefits of radar is its ability to provide contiguous, instantaneous
observations of precipitation over a wide area, other studies have focused their efforts on
demonstrating the benefits of precipitation nowcasts when input to distributed hydrological
models. In these models, the run-off response can vary within a catchment according to the
temporal and spatial variability of the rainfall, surface properties and antecedent wetness
(Ivanov et al., 2004; Vivoni et al., 2005). Amongst other things, this capability allows time
series of run-off to be generated at ungauged sites (Moore et al., 2007).
Sharif et al. (2006) explored the potential of the National Center for Atmospheric Research's
Auto-Nowcaster to improve the lead time and accuracy of hydrological forecasts made with
a physically-based distributed parameter model. Rain gauge and radar observation driven
simulations were used as a baseline. Results confirmed that the use of precipitation
nowcasts could significantly improve flood warning in urban catchments, even in the case
of short-lived events in small catchments. Similar conclusions were drawn by Vivoni et al.
(2006) in relation a set of small, mixed land-use catchments in Oklahoma, in this case using
NEXRAD-based extrapolation nowcasts and a distributed hydrological model.
6.7 Treatment of nowcast errors in hydrological forecasts
Vivoni et al. (2007) explored the impact of errors in deterministic precipitation nowcasts on
errors in flood forecasts using a distributed hydrological model and a range of catchment
sizes. Their investigations showed that increases in nowcast error with lead time produced
larger errors in the resulting hydrological forecasts. They demonstrated that the effects of
nowcast errors could be simultaneously enhanced or dampened in different locations
depending on forecast lead time and precipitation characteristics. Differences in error
