Abstract
A general framework has been developed to study the predictability of space–time averages of mesoscale rainfall in the tropics. A comparative ratio between the natural variability of the rainfall process and the prediction error is used to define the predictability range. The predictability of the spatial distribution of precipitation is quantified by the cross correlation between the control and the perturbed rainfall fields. An upper limit of prediction error, called normalized variability, has been derived as a function of space–time averaging. Irrespective of the type and amplitude of perturbations, a space–time averaging set of 25 km2–15 min (or larger time averaging) is found to be necessary to limit the error growth up to or below the prescribed large-scale mean rainfall.
