do not change with time), uncertainties due to the fact that growth and decay of
precipitation is not accounted for and uncertainties related to the type of nowcasting
system used to produce the forecasts [137, 138]. It is therefore common practice for
nowcasting models to account for the uncertainty in the predictions using ensemble
forecasts, which is a set of equally likely forecasts. Ensemble radar-based forecasts
can be generated by adding spatially correlated noise to the deterministic forecast.
The ensemble of forecasts have in common the more predictable large-scale precipitation patterns but will differ in the small-scale patterns that are less predictable.
More details on the implementation of ensemble forecasts can be found in Seed
[139] and Berenguer et al. [140].
Although NWP models perform better than radar nowcasting for longer time
scales, NWP models do not generally capture well the initial precipitation conditions
in comparison to radar-based precipitation forecasts. Consequently, in order to
combine the strengths of NWP forecasts and radar nowcasts, new blending methods
have been developed such as the Short-Term Ensemble Prediction System (STEPS)
[141]. With the increase in computer power, new NWP-based nowcasting
approaches based on 4D-Var data assimilation have been developed and look
promising [142]. However, for very short-term forecasts (60-min or so), radarbased nowcasting can provide better forecasts than NWP-based approaches. This
is particularly important for radar rainfall applications in urban catchments.
4.2 Hydrological Applications
Hydrological forecasting is one of the most important applications of radar rainfall
observations [14, 143–145]. The hydrological processes in river catchments can be
modelled using rainfall–runoff models (also known as hydrological models). These
models have a variety of applications that include flood forecasting. Depending on
the application, hydrological models can have different levels of complexity and can
be classified into lumped, semi-distributed and distributed models. In the UK, the
national flood forecasting system platform uses several hydrological models including lumped (the probability-distributed model) and distributed (the grid-to-grid
model) models [146, 147]. These models use radar rainfall measurements and forecasts (nowcasts, NWP forecasts or a combination of both) to simulate river flows for
any catchment in the UK. River flow simulations and forecasts are used to issue
flood warnings several hours in advance. In fact, radar-based hydrological forecasts
are extremely important especially during flash floods. Hydrological forecasts with
longer lead times are also available when radar nowcasts are blended with NWP
forecasts. It is obvious that the quality of the hydrological forecasts depends on the
quality of the rainfall measurements [148] and also on the quality of the hydrological
models used to simulate the hydrological processes in a particular catchment.
On the other hand, urban catchments usually have shorter response times [149–
151] and weather radar provides unique information on the dynamics of precipitation
events in space and time at high spatial and temporal resolutions, which is very
250
N. Nanding and M. A. Rico-Ramirez
precipitation is not accounted for and uncertainties related to the type of nowcasting
system used to produce the forecasts [137, 138]. It is therefore common practice for
nowcasting models to account for the uncertainty in the predictions using ensemble
forecasts, which is a set of equally likely forecasts. Ensemble radar-based forecasts
can be generated by adding spatially correlated noise to the deterministic forecast.
The ensemble of forecasts have in common the more predictable large-scale precipitation patterns but will differ in the small-scale patterns that are less predictable.
More details on the implementation of ensemble forecasts can be found in Seed
[139] and Berenguer et al. [140].
Although NWP models perform better than radar nowcasting for longer time
scales, NWP models do not generally capture well the initial precipitation conditions
in comparison to radar-based precipitation forecasts. Consequently, in order to
combine the strengths of NWP forecasts and radar nowcasts, new blending methods
have been developed such as the Short-Term Ensemble Prediction System (STEPS)
[141]. With the increase in computer power, new NWP-based nowcasting
approaches based on 4D-Var data assimilation have been developed and look
promising [142]. However, for very short-term forecasts (60-min or so), radarbased nowcasting can provide better forecasts than NWP-based approaches. This
is particularly important for radar rainfall applications in urban catchments.
4.2 Hydrological Applications
Hydrological forecasting is one of the most important applications of radar rainfall
observations [14, 143–145]. The hydrological processes in river catchments can be
modelled using rainfall–runoff models (also known as hydrological models). These
models have a variety of applications that include flood forecasting. Depending on
the application, hydrological models can have different levels of complexity and can
be classified into lumped, semi-distributed and distributed models. In the UK, the
national flood forecasting system platform uses several hydrological models including lumped (the probability-distributed model) and distributed (the grid-to-grid
model) models [146, 147]. These models use radar rainfall measurements and forecasts (nowcasts, NWP forecasts or a combination of both) to simulate river flows for
any catchment in the UK. River flow simulations and forecasts are used to issue
flood warnings several hours in advance. In fact, radar-based hydrological forecasts
are extremely important especially during flash floods. Hydrological forecasts with
longer lead times are also available when radar nowcasts are blended with NWP
forecasts. It is obvious that the quality of the hydrological forecasts depends on the
quality of the rainfall measurements [148] and also on the quality of the hydrological
models used to simulate the hydrological processes in a particular catchment.
On the other hand, urban catchments usually have shorter response times [149–
151] and weather radar provides unique information on the dynamics of precipitation
events in space and time at high spatial and temporal resolutions, which is very
250
N. Nanding and M. A. Rico-Ramirez