6 Precipitation
127
The representation heterogeneity within global numerical weather prediction and
climate models is important, and therefore sampling errors must be considered in both
the disaggregation of model rainfall and in the aggregation of surface fluxes measured
by satellites. Unfortunately global models use grids of characteristic length scale
50 Ian or larger and often fluxes from individual subgrid hydrological models which
are coupled to large scale meteorological models are arithmetically averaged over the
larger grid squares (see for example Molders and Raabe, 1997). Whilst this is
undoubtedly detrimental to both model predictions of convective rainfall and the
resultant runoff, there is little else that can be done. Conversely it is necessary to
ensure that model-derived rainfall is disaggregated over the hydrological model grid,
as suggested by, for example, Eagleson (1984), in a physically meaningful way, so
that subsequent hydrological model predictions of soil moisture and evaporation are
realistic when inputted to the meteorological model. Recently Collier (1993) has used
radar data to develop an improved method of disaggregation, but further work needs
to be undertaken to develop techniques which match the characteristics of observing
systems to numerical model grids.
6.4 The potential for improvement
6.4.1 Current performance levels
All satellite techniques suffer to a greater or lesser degree from errors arising from
sampling. Indeed, these errors can be greater than all the other errors if accumulations
are improperly computed. In tropical regions there can be a significant diurnal cycle
in rainfall activity, and the phase and intensity of the cycle may increase the errors due
to sampling. Nevertheless, since rain is a key component of the global hydrological
cycle, the only way to obtain better measurements on a global scale is from the use
of space-borne measurement techniques.
Satellite techniques are capable of estimating rainfall over areas in excess of
10 4 Ian 2 over periods of hours with an accuracy of around 50%. For much longer
integration times of order a month, this accuracy may be improved to around 10 -
20%. However, for much smaller areas errors are much larger, and indeed the
measurements are oflittle use for hydrological flow forecasting for catchment of size
around a few hundred square kilometres.
Ground-based radar offers higher accuracy than satellite techniques over small
areas (-1 02Ian 2 ) and small time periods (minutes to an hour or so). Indeed, the most
accurate accumulations are obtained with data collected every minute. Colour Plate
6.D shows a sequence of radar images at ten minute intervals. Note the highly
variable nature of the rainfall generally, but the organization of a small bond in one
location. Clearly then for most urban and many rural catchments radar offers the only
realistic approach to real-time measurement of rainfall and snowfall (see also Chap.
16).
Unfortunately radar data require significant real-time quality control if this level of
performance is to be consistently achieved. This has often lead to disappointment
127
The representation heterogeneity within global numerical weather prediction and
climate models is important, and therefore sampling errors must be considered in both
the disaggregation of model rainfall and in the aggregation of surface fluxes measured
by satellites. Unfortunately global models use grids of characteristic length scale
50 Ian or larger and often fluxes from individual subgrid hydrological models which
are coupled to large scale meteorological models are arithmetically averaged over the
larger grid squares (see for example Molders and Raabe, 1997). Whilst this is
undoubtedly detrimental to both model predictions of convective rainfall and the
resultant runoff, there is little else that can be done. Conversely it is necessary to
ensure that model-derived rainfall is disaggregated over the hydrological model grid,
as suggested by, for example, Eagleson (1984), in a physically meaningful way, so
that subsequent hydrological model predictions of soil moisture and evaporation are
realistic when inputted to the meteorological model. Recently Collier (1993) has used
radar data to develop an improved method of disaggregation, but further work needs
to be undertaken to develop techniques which match the characteristics of observing
systems to numerical model grids.
6.4 The potential for improvement
6.4.1 Current performance levels
All satellite techniques suffer to a greater or lesser degree from errors arising from
sampling. Indeed, these errors can be greater than all the other errors if accumulations
are improperly computed. In tropical regions there can be a significant diurnal cycle
in rainfall activity, and the phase and intensity of the cycle may increase the errors due
to sampling. Nevertheless, since rain is a key component of the global hydrological
cycle, the only way to obtain better measurements on a global scale is from the use
of space-borne measurement techniques.
Satellite techniques are capable of estimating rainfall over areas in excess of
10 4 Ian 2 over periods of hours with an accuracy of around 50%. For much longer
integration times of order a month, this accuracy may be improved to around 10 -
20%. However, for much smaller areas errors are much larger, and indeed the
measurements are oflittle use for hydrological flow forecasting for catchment of size
around a few hundred square kilometres.
Ground-based radar offers higher accuracy than satellite techniques over small
areas (-1 02Ian 2 ) and small time periods (minutes to an hour or so). Indeed, the most
accurate accumulations are obtained with data collected every minute. Colour Plate
6.D shows a sequence of radar images at ten minute intervals. Note the highly
variable nature of the rainfall generally, but the organization of a small bond in one
location. Clearly then for most urban and many rural catchments radar offers the only
realistic approach to real-time measurement of rainfall and snowfall (see also Chap.
16).
Unfortunately radar data require significant real-time quality control if this level of
performance is to be consistently achieved. This has often lead to disappointment