In the late 50ies and the 60ies the limit of predictability of weather has been shown to be 5 to
14 days by theoretical and numerical studies as consequence of sensible dependence on initial
conditions. The simulation of an ensemble is the standard approach to address this
uncertainty. Climatological means over time scales of 30 years are regarded as “certain”
under constant climate forcing conditions. A systematic analysis of predictability limits on
different space and time scales in the earth system is still missing. It is relevant in
particular for time scales between the time scales of weather and climate and helps
avoiding misinterpretation of the results and/or to find an optimal configuration for the
ensemble.
In terms of statistics, the predictability of weather can be associated with the
predictability of 6h and 100 km mean values. In mid latitudes it reaches a saturation value at
the time scale of baroclinic instability Ti of 3-5 days. In the case of a purely stochastic
process, this uncertainty is decreasing with -N– where N is the number of instability time
periods.
One of the open questions is, which field variables exhibit a purely stochastic behavior
and where. The application of Earth System Models is computationally demanding. Chaotic
behavior may occur in some regions at certain conditions affecting the analysis. A huge
number of degrees of freedom makes very long simulation times necessary. The application
of limited area modeling opens the opportunity to analyse the behavior in different regions
independently und thus to investigate the stochastic properties in different climates.
Three different regions Europe, Africa and Central America have been simulated
twice (reference and disturbance run) at standard grid resolution of 18 to 25 km
using the community model COSMO-CLM. In Europe a purely stochastic behavior
was found for the momentum, pressure and precipitation. A strong memory effect
was found for soil moisture and temperature and a weak memory effect for the
atmospheric temperature. The analysis of the simulation for Africa and Meso-America is
ongoing.
Predictability limits for different thresholds and variables can be calculated in terms of
the ensemble size needed to keep the uncertainty below the threshold. Such maps
will be presented and the differences between the variables and regions will be
discussed. |