cc: Wolfgang Cramer , rik.leemans@rivm.nl, rounsevell@geog.ucl.ac.be, Dagmar Schroeter , Mike Hulme date: Wed, 21 Aug 2002 14:28:58 +0300 from: Timothy Carter subject: Re: ATEAM climate data-set to: Tim Mitchell , Markus.Erhard@pik-potsdam.de Dear all, Tim M. has done a fantastic job in preparing the climate data sets for ATEAM. Now we need to ensure that these data are used in the most effective way possible. In order to do that, we need to advise the modelling groups on a set of priority scenarios to adopt. Tim and I have had some correspondence on this but summer holidays have delayed our discussions somewhat. However, it seems that there are some alternative options for scenario prioritisation available. I think these really need to be resolved from the ATEAM impacts point of view rather than for pure climatological reasons. They are a little technical, but I hope you can all bear with me. There have already been several documents produced at different stages of scenario development, and while the scenario construction methodology is consistent throughout, there are some discrepancies between documents concerning the GCM information that was available for the analysis. THE DATA The final data set that is now available (for 5 different climatic variables) was presented by Tim in a table (document v1-02.pdf, dated 9 July 2002) which I reproduce here: PCM CGCM2 CSIRO2 HadCM3 A1FI low low low high A2 high medium medium high B2 low low low high B1 medium low low high This shows that there are 4 sets of GCM outputs available for 4 SRES emissions scenarios. Most of these are direct model runs, but a few are "scaled" from an SRES emissions scenario that has been run to represent an SRES scenario that has not been run (e.g. PCM did not conduct an A1FI simulation, so this is scaled from the A2 simulation by PCM). In addition, there are two gridded time series of observations for each climatic variable: (1) for 1901-2000 (expressed as anomalies from 1961-1990) (2) for 1901-2000 with the global temperature trend removed. Scenario data for the future, 2001-2100, requires combining the model outputs with a detrended version of the 1901-2000 observational time series referenced to 1961-1990. A scenario time series of "unforced" future climate simply repeats the detrended 1901-2000 series (not referenced to 1961-1990). The table entries are the priorities assigned by Tim M. to each run: High: All groups to use Medium: Strongly recommended to use if possible Low: Not mandatory, but desirable if computational facilities and time allow In addition, a high priority was assigned to an "unforced" scenario. THREE MAJOR ISSUES 1. Should we prioritise the emissions uncertainties over the inter-GCM uncertainties? In the Table this is done, by prioritising all four HadCM3 scenarios, which span the emissions range, but only two alternative GCMs (HadCM3 and PCM) for the A2 scenario (5 scenarios). The emissions uncertainties become important in the second half of the century (2050s and particularly 2080s), but for the 2020s time slice, the differences between emissions scenarios in a given region are quite small. Since we have 4 GCM patterns available, perhaps we should emphasise the difference between these for a given emissions scenario. Otherwise, all but one of the high priority future scenarios are based on HadCM3. This may be especially important for the early decades of the century. Alternative suggestion: to prioritise all four HadCM2 scenarios PLUS all four A2 scenarios, including HadCM3, giving 7 scenarios. 2. Should we attempt to embrace the fullest possible range of uncertainty? At present, the recommendation is to use the PCM B1 pattern as the lowest change and the HadCM3 A1FI as the highest change. HadCM3 A1FI is already high priority. However, PCM B1 is only allocated a medium priority by Tim to reduce the number of scenario runs. Should this be elevated to high priority? 3. What reference climate should we use to compare future impacts? Originally, we had thought to analyse the long, multi-century GCM control runs to look at modelled natural variability on long time scales. Unfortunately, these runs are not available for this project. Thus, Tim had to devise alternatives. There are two options so far on offer. OPTION 1: This uses the observations for 1901-2000 to define the reference climate, and then compares the impacts during that period with impacts occurring during the scenario period, 2001-2100. An obvious advantage is that all impact groups will anyway be estimating impacts for the 1901-2000 period based on the actual observations, both for model initiation and for model validation. It has the disadvantage that it may already contain an anthropogenic signal of warming, and therefore does not represent an unforced climate. It has a second disadvantage of not offering a reference for the future (i.e. to enable impact analysts to assess future impacts in the absence of climate change, but assuming other scenarios of land use change and air pollution to occur). OPTION 2: This uses a detrended observational record from 1901-2000 to represent an unforced climate, and repeats this into the future (2001-2100). The advantage of this option is that long term global long term climate changes that have occurred during the 20th century are detrended, and the only trends remaining in the observed data are multi-decadal and/or regional in origin. These can be assumed to be due to natural variability, and can be used to provide a reference for the future. A second advantage is that impacts can be estimated for the detrended, "unforced" historical period and compared to impacts estimated for the actual observations to evaluate the impact of historical anthropogenic forcing A disadvantage of using this detrended record is that the detrending may be removing natural variations, not anthropogenic change. Another disadvantage is that a detrended record will exhibit less absolute variability than the original observations, so the impacts could well be reduced relative to those that occurred during the 20th century. A third disadvantage is that use of the repeated, detrended series for 2001-2100, will not provide a directly comparable set of impacts. All future climate changes and their impacts must be considered relative to the present-day, which is commonly represented by the 1961-1990 climate. However, if the unforced climate is referenced to an unforced 20th century, this is not comparable to the 1961-1990 reference climate used in the other scenario runs. If there have been long-term trends during the 20th century, then ecosystems will have had to respond and to adapt to such changes. These trends might well be absent from the detrended climate. To address the last disadvantage, one additional possibility can be proposed: OPTION 3 - NEW: Use the 1901-2000 observed for the historical period, referenced to 1961-1990, and the detrended 1901-2000 record from 2001-2100, again relative to 1961-1990. [Note that Tim would have to provide additional guidance on how to do this.] This would offer a direct comparison between scenario impacts and impacts in the absence of ANY ADDITIONAL climate change relative to 1961-1990. At present, Tim has suggested that Option 2 be high priority. Option 1 is presumably being undertaken by all groups in any case. So the choice really lies between Options 2 and 3, if some measure of impacts for an unforced climate is to be obtained. These are the three issues to be explored. There may be others as well. I hope these explanations are intelligible. Tim may wish to add to them or correct them is they are incorrect. I know that he is finalising the full documentation, so we should try to ensure that everyone is happy with the scenarios and the scenario guidance. So, comments are welcome. Dagmar/Wolfgang: If you think this should go out to the wider ATEAM community, please forward it. Best regards, Tim