UNCOVER is a tool to help explore uncertainty and trade-offs within ensemble model simulations.
UNCOVER implements a search procedure to discover ensemble member simulations that maximise performance against a set of one or more preferences.
The resulting set, called
a Pareto set,
contains the model output ensemble members for which your data contains no better-performing
simulation when considering all of the preferences.
The Pareto set is found using a
based on block-nested-loop and lattice algorithms.
The method weights each of an ensemble of behavioural (i.e. potentially acceptable) models by a measure of
its performance over a calibration period. The performance metrics are interpreted analagously to likelihoods.
The resulting set of predictions are used to construct a weighted cumulative distribution function (CDF)
as an expression of the uncertainty for any predicted variable of interest. The method offers flexibility to assess uncertainty
conditionally on a range of measures, although this flexibility can lead to ambiguity without a clear account of the judgements
made by the analyst in a particular application. Future revisions of UNCOVER will offer tools to expose and record these judgements.
This web page is currently a demonstration prototype.