Section 11.4: Stochastic Precipitation Models
203
face variables (especially precipitation) to GeM free atmosphere variables.
Among these methods are Model Output Statistics (MOS) routines, which
are essentially regressions that adjust numerical weather 'predietions (produced on grid meshes smaller than those used for GeM climate simulations,
but still "Iarge" compared to the scale required for Iocal assessments). MOS
adjustments to precipitation are used, for instance, in quantitative precipitation forecasts (QPFs) for flood forecasting. One drawback of these routines
is that they attempt to produce "best estimates" in the least squares sense.
While this may be appropriate for forecasting applications, least squares estimates underestimate natural variability (especially when the forecasts are
most accurate), which can be critical for climate effeets assessments.
Other semi-empirical approaches have been developed to relate longer term
GeM simulations of free atmosphere variables to local precipitation. Among
these are the canonical correlation approach of von Storch et al. (1993), and
the regression approach of Wigley et al. (1990). The disadvantage of these
approaches is that they are most appropriate for prediction of local variables at a time scale much longer than the catchment response scale (e.g.,
monthly or seasonal). Therefore, there is no praetical way to incorporate
the predietions within a rainfall-runoff modeling framework from which effeets interpretations might be made. Moreover, in the case of the regression
approach of Wigley et al. (1990), the input variables for local precipitation
predietions include large-scale (e.g., GeM) precipitation, which is responsible for much of the predietive accuracy. Unfortunately, as indicated above,
GeM precipitation predictions are often badly biased, and this bias would
be transmitted to the Iocal predictions.
11.4 Stochastic Precipitation Models with
External Forcing
Several investigators have recently explored stochastic precipitation models
that operate at the event scale (defined here as daily or shorter) and incorporate, explicitly or implicitly, external large-area atmospheric variables.
The motivation for development of these methods has been, in part, to
provide stochastic sequences that could serve as input to hydrologie (e.g.,
precipitation-runoff) models. Most of the work in this area has utilized, either directly or via summary measures, large scale free atmosphere variables
rather than large area surface fluxes. In this respeet, their objective has
been to simulate stochastically realistic precipitation sequences that incorporate external drivers, rather than to disaggregate large area predietions, for
instance, of precipitation.
203
face variables (especially precipitation) to GeM free atmosphere variables.
Among these methods are Model Output Statistics (MOS) routines, which
are essentially regressions that adjust numerical weather 'predietions (produced on grid meshes smaller than those used for GeM climate simulations,
but still "Iarge" compared to the scale required for Iocal assessments). MOS
adjustments to precipitation are used, for instance, in quantitative precipitation forecasts (QPFs) for flood forecasting. One drawback of these routines
is that they attempt to produce "best estimates" in the least squares sense.
While this may be appropriate for forecasting applications, least squares estimates underestimate natural variability (especially when the forecasts are
most accurate), which can be critical for climate effeets assessments.
Other semi-empirical approaches have been developed to relate longer term
GeM simulations of free atmosphere variables to local precipitation. Among
these are the canonical correlation approach of von Storch et al. (1993), and
the regression approach of Wigley et al. (1990). The disadvantage of these
approaches is that they are most appropriate for prediction of local variables at a time scale much longer than the catchment response scale (e.g.,
monthly or seasonal). Therefore, there is no praetical way to incorporate
the predietions within a rainfall-runoff modeling framework from which effeets interpretations might be made. Moreover, in the case of the regression
approach of Wigley et al. (1990), the input variables for local precipitation
predietions include large-scale (e.g., GeM) precipitation, which is responsible for much of the predietive accuracy. Unfortunately, as indicated above,
GeM precipitation predictions are often badly biased, and this bias would
be transmitted to the Iocal predictions.
11.4 Stochastic Precipitation Models with
External Forcing
Several investigators have recently explored stochastic precipitation models
that operate at the event scale (defined here as daily or shorter) and incorporate, explicitly or implicitly, external large-area atmospheric variables.
The motivation for development of these methods has been, in part, to
provide stochastic sequences that could serve as input to hydrologie (e.g.,
precipitation-runoff) models. Most of the work in this area has utilized, either directly or via summary measures, large scale free atmosphere variables
rather than large area surface fluxes. In this respeet, their objective has
been to simulate stochastically realistic precipitation sequences that incorporate external drivers, rather than to disaggregate large area predietions, for
instance, of precipitation.