• English 
  • Spanish 

A Comparison of Statistical, Dynamical and Combined MOS Downscaling Approaches in the Framework of the PNACC-2012 Spanish Program

In this work we present a comparison of dynamical and statistical (both perfect prognosis and Model Output Statistics, MOS) downscaling techniques focusing on maximum temperature and precipitation in the Iberian peninsula —a complex climate region with Atlantic, continental, and Mediterranean influences.— To this aim, the same forcing conditions (from ERA40 re- analysis data) are considered for all methods and a k- fold cross validation is conducted focusing on different validation scores. The methods are assessed not only in terms of accuracy and distributional similarity, but also considering their robustness/stationarity in changing climate conditions; to this aim, anomalous historical periods (warm/cold for maximum temperature and wet/dry for precipitation) are used as surrogates of possible future climate alterations, and differences with normal periods are tested using a standard Student t-test. As a result, the strengths and limitations of these techniques in terms of their robust applicability in climate change studies are reported.

As most important results, the statistical/dynamical downscaling approaches show the best/worst performance in terms of accuracy and distributional similarity; however, MOS techniques are able to correct the deficiencies of the dynamical models according to the referece observations. However, the known limitations of the analogue method to reach unobserved values affect the robustness of statistical downscaling methods, mainly in the case of the maximum temperature, which are less robust than the dynamical ones.

Finally, we compare the 21st century regional projections for Spain obtained with the above methods in the framework of the Spanish Plan for Regional Climate Change Scenarios (PNACC-2012) showing that, when discarding nonrobust methods, both dynamical and statistical projections are in good agreement.