Ensemble downscaling methods

Downscaling ensembles to high resolution can require a lot of resources and memory. Here we consider the case where you want to downscale a coarse grid ensemble to a fine scale grid and compute an ensemble statistic across the members, outputting a single scalar field. There are three cases to consider:

• a) compute the required statistic on the coarse resolution and downscale this to the fine resolution. This doesn't require any special functionality.
• b) for each fine scale gridpoint, find the nearest coarse point and apply the ensemble operation where parameters to the method are specified on the fine grid. This requires a function that can apply the operation
• c) for each fine point, downscale all members to this point and compute the ensemble statistic.

This section considers case b). Gridpp contains several methods that can compute an ensemble statistic without downscaling all members.

Threshold exceedance probability

One example is to extract the fraction of members exceeding a threshold that a varies across the fine scale grid. This can be useful when computing the fraction of members with freezing temperatures and where a quantile mapping is needed in the downscaling from coarse to fine.

In this case we can compute a threshold on each fine scale gridpoints that will give us the freezing point if the nearest neighbour from the coarse field is used.