ESRL/PSD Seminar Series
On the reliability of climate models: How well do they describe observed trends?
Geert Jan van Oldenborgh
KNMI (Royal Netherlands Meteorological Institute) , De Bilt, Netherlands
Climate models are widely used to construct local projections of future climate changes. For these to be used as "forecasts" the ensemble of climate models has to be reliable in the sense that the projected probability of outcomes should correspond with the realised probability. In weather and seasonal forecasts this is verified over a set of past forecasts. Since the local climate change signal is now emerging from the weather noise in many regions of the world, the reliability of climate model ensembles can be estimated by comparing the observed and modelled trends in temperature and precipitation over the past 50 to 100 years. The spatial dimension is used to gather the necessary statistics. Although global and continental trends are represented well, it is shown that in many regions of the world the observed local trends are not within the ensemble of modelled trends. These areas are larger than would be expected on the basis of chance fluctuations and are therefore a consequence of either misrepresentation of the trends or underestimation of low-frequency variability in climate models. Downscaling with regional climate models does not change this conclusion beyond the addition of orographic details. For European temperature and precipitation trend we have investigated the causes of the discrepancies. In winter, both temperature and precipitation have increased much faster than modelled due to an increase in westerly circulation associated with a significant increase in air pressure over the Mediterranean. In spring and summer the faster rise of temperature is over the land areas of southern Europe. In the Netherlands it is associated with a large increase in global radiation. The concomitant rise in East Atlantic SST causes an increase in coastal precipitation that is absent in the climate models. This is partially explainable by a wrong ocean current system in the North Atlantic Ocean, which is a well-known deficiency of coarse resolution ocean models. Finally, the decrease of mist and fog caused by decreased air pollution is not represented in climate models. None of these factors is associated with known modes of low-frequency variability, leading to the conclusion that the biases are more likely in the trend than in the variability. Time permitting, extreme hourly precipitation trends are discussed. Plotting these as a function of dew point temperature gives a common scaling behavior, between De Bilt and Hong Kong, two stations with long hourly time series. In the Netherlands this allows for an attribution of the increase of hourly extremes to local temperature rise. In Hong Kong this attribution cannot be made and other factors, such as possibly urbanisation, must be responsible for the observed increase.
Tuesday, Jan 31 2012
SECURITY: If you are coming from outside the NOAA campus, you must stop at the Visitor Center to obtain a vistor badge. Please allow 10 extra minutes for this procedure. If you are a foreign national coming from outside the NOAA campus, please contact Barbara Herrli (303-497-3876) for information on access requirements.