• English 
  • Spanish 

VALUE perfect predictor validation results, part 1: Marginal, extremal and temporal aspects

VALUE is an open European network to validate and compare downscaling methods for climate change research (http://www.value-cost.eu). A key deliverable of VALUE is the development of a systematic validation framework to enable the assessment and comparison of both dynamical and statistical downscaling methods. This framework is based on a user- focused validation tree, guiding the selection of relevant validation indices and performance measures for different aspects of the validation (marginal, temporal, spatial, multi-variable). Moreover, several experiments have been designed to isolate specific points in the downscaling procedure where problems may occur (assessment of intrinsic performance, effect of errors inherited from the global models, effect of non-stationarity, etc.). The list of downscaling experiments includes 1) cross-validation with perfect predictors, 2) GCM predictors and 3) pseudo reality predictors (see Maraun et al. 2015, Earth’s Future, 3, doi:10.1002/2014EF000259, for more details). The results of these experiments are gathered, validated and publicly distributed through the VALUE validation portal, allowing for a comprehensive community-open downscaling intercomparison study.
As a result of the open call for contribution to the experiments, over 40 methods representative of the main approaches (MOS and PP) and techniques (linear scaling, quantile mapping, analogs, weather typing, linear and generalized regression, weather generators, etc.) were submitted. This constitutes the largest and most comprehensive to date ensemble of statistical downscaling methods. In this contribution, we present an overall validation of the first of the three mentioned experiments, the cross-validation with perfect predictors, analyzing marginal and temporal aspects of mean and extreme values to assess the intrinsic performance and added value of statistical downscaling methods at both annual and seasonal levels. This validation takes into account the different properties/limitations of different approaches and techniques (as reported in the provided metadata) in order to perform a fair comparison. Thus, it is pointed out that this experiment alone is not sufficient to evaluate the limitations of (MOS) bias correction techniques. Moreover, it also does not fully validate PP since we don't learn whether we have the right predictors and whether the PP assumption is valid. These problems will be analyzed in the subsequent VALUE experiments.

Palabras clave: