Questions on observations and how these may be used to evaluate climate models. If direct observations are lacking, other types of data may be used – such as reanalysis products.
How can climate models and downscaling techniques be evaluated?
Climate models are evaluated in their capacity to simulate the recent past climate and the evolution of the historical climate. Output from the climate models are compared to observations. Large discrepancies need to be explained and ideally lead to the development of improved model versions. As climate models are very similar to numerical weather prediction models, evaluation can benefit from such models that are routinely applied and evaluated on a daily basis.
In particular for regional climate models, evaluation is performed in two steps. In the first step, the RCM is evaluated in experiments when it is downscaling reanalysis data. In those, so called, hindcast experiments, the large-scale evolution of the climate system, including interannual variability, is as close to the observed state as it is possible. Consequently, model output can be more directly compared with observations, also in a time-series mode (i.e. the sequence of weather events should be realistic).
In the second step, each individual climate change projection can be evaluated for how well the historical climate is captured. An example is given for the EURO-CORDEX area in Vautard et al. (2020). In this case, it is essential that the underlying global climate model is evaluated as its representation of the large-scale conditions is key to the performance of the regional climate model. Also empirical downscaling methods need to be evaluated. A common approach is to perform cross-validation where data is split in two sets. One set is then used to find the empirical relationship, and the other set is used to evaluate it. Different ways to evaluate RCMs and statistical downscaling techniques are discussed in Maraun et al. (2015). For these types of evaluations, when downscaling of GCM output is considered, it is not meaningful to compare the actual time evolution of the climate with observations due to the internal variability in the climate system (i.e the sequence of weather events is not realistic and they can be reliable only in statistical sense for a longer time – typically 30 years – period).
It is recommended that users of downscaled climate information are aware of to what degree the downscaling methods can provide data that are of good enough quality for their respective questions.
What data can be used for evaluating climate models?
A wide range of observational data are used for evaluating climate models. This includes both in-situ measurements of near-surface variables as well as data from the atmosphere and oceans and remote sensing data from satellites and weather radars. It also includes so called reanalysis data that is a composite of observations as analysed by a weather prediction model.
Various data sets are used for model evaluation in different regions. This includes both global datasets and more detailed regional datasets. Many such datasets consist of information from near-surface observations that have been gridded but it could also be data from satellites. In addition to observations and remote sensing data, also reanalysis data are extensively used. These data are a blend of observations and weather forecast models. Particularly, for the evaluation of the EURO-CORDEX regional climate models a strong emphasize has been on the gridded data from E-OBS with daily data for temperature and precipitation for Europe extending back to 1950 (Cornes et al. 2018).
Users of EURO-CORDEX regional climate models should be aware that the observational material used for evaluation of regional climate models differ between different areas in Europe. For some areas, notably areas of complex topography, there are limitations in these data and, consequently, the skill of the climate models may not be fully known.
References
Cornes R, van der Schrier G, van den Besselaar EJM and Jones PD (2018) An Ensemble Version of the E-OBS Temperature and Precipitation Datasets, J. Geophys. Res. Atmos., 123. doi:10.1029/2017JD028200
Maraun D, Widmann M, Gutiérrez JM, Kotlarski S, Chandler RE, Hertig E, Wibig J, Huth R and Wilcke RAI (2015), VALUE: A framework to validate downscaling approaches for climate change studies. Earth's Future, 3: 1–14., https://doi.org/10.1002/2014EF000259
Vautard R, Kadygrov N, Iles C, Boberg F, Buonomo E, Bülow K, Coppola E, Corre L, van Meijgaard E, Nogherotto R, Sandstad M, Schwingshackl C, Somot S, Aalbers E, Christensen OB, Ciarlo JM, Demory M-E, Giorgi F, Jacob D, Jones RG, Keuler K, Kjellström E, Lenderink G, Levavasseur G, Nikulin G, Sillmann J, Solidoro C, Sørland SL, Steger C, Teichmann C, Warrach-Sagi K and Wulfmeyer V (2020) Evaluation of the large EURO-CORDEX regional climate model ensemble. J. Geophys. Res. DOI: 10.1029/2019JD032344
