IPCC Fourth Assessment Report: Climate Change 2007
Climate Change 2007: Working Group I: The Physical Science Basis

11.3.2 Skill of Models in Simulating Present Climate

Atmosphere-Ocean General Circulation Models show a range of performance in simulating the climate in Europe and the Mediterranean area. Simulated temperatures in the MMD models vary on both sides of the observational estimates in summer but are mostly lower than observed in the winter half-year, particularly in northern Europe (Supplementary Material Table S11.1). Excluding one model that simulates extremely cold winters in northern Europe, the seasonal area mean temperature biases in the northern Europe region (NEU) vary from –5°C to 3°C and those in the southern Europe and Mediterranean region (SEM) from –5°C to 6°C, depending on model and season. The cold bias in northern Europe tends to increase towards the northeast, reaching –7°C in the ensemble mean in the northeast of European Russia in winter. This cold bias coincides with a north-south gradient in the winter mean sea level pressure that is weaker than observed, which implies weaker than observed westerly flow from the Atlantic Ocean to northern Europe in most models (Supplementary Material Figure S11.22).

Biases in simulated precipitation vary substantially with season and location. The average simulated precipitation in NEU exceeds that observed from autumn to spring (Supplementary Material Table S11.1), but the interpretation of the difference is complicated by the observational uncertainty associated with the undercatch of, in particular, solid precipitation (e.g., Adam and Lettenmaier, 2003). In summer, most models simulate too little precipitation, particularly in the eastern parts of the area. In SEM, the area and ensemble mean precipitation is close to observations.

Regional Climate Models capture the geographical variation of temperature and precipitation in Europe better than global models but tend to simulate conditions that are too dry and warm in southeastern Europe in summer, both when driven by analysed boundary conditions (Hagemann et al., 2004) and when driven by GCM data (e.g., Jacob et al., 2007). Most but not all RCMs also overestimate the interannual variability of summer temperatures in southern and central Europe (Jacob et al., 2007; Lenderink et al., 2007; Vidale et al., 2007). The excessive temperature variability coincides with excessive interannual variability in either shortwave radiation or evaporation, or both (Lenderink et al., 2007). A need for improvement in the modelling of soil, boundary layer and cloud processes is implied. One of the key model parameters may be the depth of the hydrological soil reservoir, which appears to be too small in many RCMs (van den Hurk et al., 2005).

The ability of RCMs to simulate climate extremes in Europe has been addressed in several studies. In the Prediction of Regional scenarios and Uncertainties for Defining European Climate change risks and Effects (PRUDENCE) simulations (Box 11.2), the biases in the tails of the temperature distribution varied substantially between models but were generally larger than the biases in average temperatures (Kjellström et al., 2007). Inspection of the individual models showed similarity between the biases in daily and interannual variability, suggesting that similar mechanisms may be affecting both.

The magnitude of precipitation extremes in RCMs is model-dependent. In a comparison of the PRUDENCE RCMs, Frei et al. (2006) find that the area-mean five-year return values of one-day precipitation in the vicinity of the European Alps vary by up to a factor of two between the models. However, except for too-low extremes in the southern parts of the area in summer, the set of models as a whole showed no systematic tendency to over- or underestimate the magnitude of the extremes when compared with gridded observations. A similar level of skill has been found in other model verification studies made for European regions (e.g., Booij, 2002; Semmler and Jacob, 2004; Fowler et al., 2005; see also Frei et al., 2003).

Evidence of model skill in simulation of wind extremes is mixed. Weisse et al. (2005) find that an RCM simulated a very realistic wind climate over the North Sea, including the number and intensity of storms, when driven by analysed boundary conditions. However, most PRUDENCE RCMs, while quite realistic over sea, severely underestimate the occurrence of very high wind speeds over land and coastal areas (Rockel and Woth, 2007). Realistic frequencies of high wind speeds were only found in the two models that used a gust parametrization to mimic the large local and temporal variability of near-surface winds over land.

Box 11.2: The PRUDENCE Project

The Prediction of Regional scenarios and Uncertainties for Defining European Climate change risks and Effects (PRUDENCE) project involved more than 20 European research groups. The main objectives of the project were to provide dynamically downscaled high-resolution climate change scenarios for Europe at the end of the 21st century, and to explore the uncertainty in these projections. Four sources of uncertainty were studied: (i) sampling uncertainty due to the fact that model climate is estimated as an average over a finite number (30) of years, (ii) regional model uncertainty due to the fact that RCMs use different techniques to discretize the equations and to represent sub-grid effects, (iii) emission uncertainty due to choice of IPCC SRES emission scenario, and (iv) Boundary uncertainty due to the different boundary conditions obtained from different global climate models.

Each PRUDENCE experiment consisted of a control simulation representing the period 1961 to 1990 and a future scenario simulation representing 2071 to 2100. A large fraction of the simulations used the same boundary data (from the Hadley Centre Atmospheric Model (HadAM3H) for the A2 scenario) to provide a detailed understanding of the regional model uncertainty. Some simulations were also made for the B2 scenario, and by using driving data from two other GCMs and from different ensemble members from the same GCM. More details are provided in, for example, Christensen et al. (2007), Déqué et al. (2005) and http://prudence.dmi.dk.