X. Wang, G. Burgers and F. Selten Ensemble prediction system (EPS) has been used widely in weather prediction for many years. Since more and more computing resources are available, in recent years the EPS also used in climate predictions. The use of the EPS for climate predictions reflects the fact that there are many uncertainties in the weather and climate prediction models. These uncertainties include the parameterization of the sub-grid process, clouds, clouds feedbacks, solar irradiance and atmospheric compositions etc.. Use of EPS makes it possible for us to make a chance prediction for the climate and the range of possible climate change if the EPS has enough spread. An ensemble with 62 members has been made by the Dutch Challenge Project at SARA. All the 62 runs differ only in the initial states in the atmosphere. Because the chaotic character of the climate system model simulations starting from a slightly different initial conditions evolute independently after a number of months, hence after a number of years of integration different realizations of climate will be produced.

From observations it is known that the global mean temperature has increased by about 0.5°C in the period of 1970 to present. This temperature increase can be seen both in ncep/ncar Reanalysis and ERA40 Reanalysis data. It is accepted that this temperature trend is mainly produced by the increase of CO2 concentration in the atmosphere since 1940. Locally the temperature has increased even stronger in some areas. For instance in Northern America some areas temperature has increased upto 2.0°C/decade in the winter months. In North-West Europe and Asian continent more or less the same trends has been observed. Is this temperature change also simulated by the ensemble? If yes. How are the ensemble members different from each other? How are they compare with the observations? Can we make probability prediction base on the ensemble? These are the questions that we try to answer by comparing the temperature change simulated by the ensemble and the observations (ncep/ncar Reanalysis) both locally and globally.

The Dutch Challenge Project uses the Community Climate System Model version 1.4 with resolution of T31. It was developed by NCAR in the U.S.. It contains atmospheric, ocean, land and sea-ice, and coupler components. The model was run from 1900 with initial condition of 1900. From this state the model was integrated for 40 years with prescribed time dependent atmospheric composition, solar irradiance and sulphate and volcanic aerosols. The atmospheric state in 1940 is randomly perturbed and used as an initial state for 62 parallel integrations. The initial state of the other components of the system are not perturbed. Historical estimates of atmospheric composition, solar irradiance and sulphate and volcanic aerosols are used upto the year 2000. From 2000 until 2080 only the atmospheric composition changes according to the BAU(Business As Usual). The output data that we used here for this study are the monthly mean surface temperature from year 1970 till 2003. The temperature trend is defined as the mean temperature in period of 1987-2003 minus mean temperature in period 1970-1986. The temperature trends are computed for each member of the ensemble and compared with ncep/ncar Reanalysis. First of all we have computed the temperature trend (as defined above) for each grid box (10° x 5°) (lon x lat) on the earth for the ncep/ncar Reanalysis for all the seasons. In figure.1 the results for summer and winter are shown.

Figure.1a shows the temperature trend in summer (JJA). From this figure it is clear that the most striking temperature increase in the northern hemisphere is in West Europe with maximum magnitude around 0.8-1.0°C/decade. Over North-American continent the largest warming occurred in the region of Alaska with maximum value about 0.6-0.8°C/decade. Over the Eure-Asia continent the temperature trend show positive values around 0.2-0.4°C/decade with some of the grid-boxes 0.6. The maximum of temperature trend in the southern hemisphere concentrated in a narrow band extending from 60°S-90°S. The maximum value reaches 3°C/decade. We must bury in mind that the data in these area are less reliable. This large trend could be spurious. The only region in the rest of southern hemisphere with noticeable temperature increase is the western coastal area of Angola with a value of 0.8-1.0°C/decade.

Looking at figure.1b for winter case (DJF) it seems that the warming trend extends over much wider area of the northern hemisphere. Both continents of North-America and Eure-Asia show large area of warming. In North-West Europe a warming of 1.0°C/decade has been observed. Large area of Asian continent experienced a warming trend varies from 0.2 to 1.2°C/decade. The trend in the North Atlantic is warming in the order of 0.4-0.6°C/decade. In the tropical region the temperature trend is much less. On the African continent north of the equator shows a cooling trend, north of it a slight warming trend. In the southern hemisphere the temperature trend is rather weak. The south ocean is largely covered by blue color indicating a cooling trend. Also the south Atlantic shows a cooling trend.

In figure.2 the 62 members mean temperature trend are shown for summer and winter. From figure.2a we see that in the summer (JJA) the temperature trend is very smooth comparing with the ncep/ncar case. The large warming in West Europe is not simulated by the ensemble mean. Also the warming trend in area of Alaska is not show up. In the southern hemisphere the maximum trend appears in the southern ocean with magnitude of 0.6°C/decade. The cooling trend in this area in the ncep/ncar Reanalysis is not shown in the ensemble mean.

In winter case (DJF) figure.2b the trend is much weaker compare with the ncep/ncar Reanalysis. The strong warming in North-West Europe, Eure-Asian continents and North-American continent are not reproduced by the ensemble mean. Although it shows a general warming trend on most part of the earth. The other important feature is that the warming is most pronounced at higher latitude of northern hemisphere which is in agreement with other people's results. The maximum trend occurs in the region of east Greenland and west of Greenland sea which exceeds 2.3°C. But it concentrated on a very limited small area and it could be due to model problem since as the model integrates further this trend disappeared. On both Eure-Asian and North-American continents the warming trend is in the order of 0.2 to 0.4°C/decade which is less than half of the magnitude found in ncep/ncar Reanalysis.

The spread of the ensemble in term of standard deviation is also computed and shown in figure.3. We see that the standard deviation in the summer is quit small, larger values appear in the higher latitudes. Figure.3b shows winter season, we see that the standard deviation is relatively larger in northern higher latitudes.

In figure.4 the ratio between mean temperature trend of the 62 members and standard deviation of it are shown for summer and winter seasons. From this figure it's clear that the signal to noise ratio is most pronounced in the tropical region. This is true for both summer and winter cases. In figure.5 the results of the Student t-test are show. This test is done to see whether the difference in the temperature of the two periods are significant different statistically. From the figure it's clear that the difference is significant almost everywhere on the globe. Only in very limited areas not significant.

In figure.6 the temperature trend of global mean, tropical area (30N-30S) mean, north and south hemispheric mean as a function of month for 62 members mean and ncep/ncar Reanalysis are shown. From this figure we see that the reanalysis shows a larger global mean temperature trend in March till October. Only in winter months (November till February) the ensemble mean shows a more or less same trend as reanalysis. For tropical area the two cases agree with each other very well. For northern hemisphere the member mean shows a slightly smaller trend comparing with ncep/ncar Reanalysis both in summer and in winter. But the general feature: larger winter trend and smaller summer trend is in agreement. For southern hemisphere the member mean shows very little variability with time whereas the ncep/ncar Reanalysis gives clear seasonal variabilities: in the summer (southern winter) large temperature increase and in the winter (southern summer) even negative temperature trend.

(1) The temperature trend found in the ensemble mean is not as large as in ncep/ncar Reanalysis. The large warming trend in northern hemisphere in the winter and the warming trend in the summer Europe are not simulated by the ensemble mean. Because the ensemble mean cancels out internal natural variabilities it may conclude that the large warming in the ncep/ncar Reanalysis is due to natural variability. Increasing CO2 concentration does have some warming effect but it's much smaller than what was observed.

(2) The ensemble mean shows warming mainly in winter in northern higher latitudes. Also there shows the most variability in the warming.

(3) The signal to noise ratio is most pronounced in the tropical region for both summer and winter indicating the warming trend in there is more predictable than in the higher latitudes.

(4) The North Atlantic shows warming trend at the same time South Atlantic shows a cooling trend. It could be interesting to investigate whether this is related to the change of thermohaline circulation.

Contact e-mail wang@knmi.nl tel.: 030-2206757 or burgers@knmi.nl, 030-2206714

© KNMI