Hi why you want to select the GCMs,? if you decide to use GCMs use them all. Why can't you use regional climate models from CORDEX?. I advice you use several GCMs or RCMs to characterize the extremes

If you expect answers about the best model delivering precipitation, I would like to ask, what do you mean by best? What is your benchmark? Common ones are

• The annual total precipitation amount
• Reproduction of the overall timing in terms of seasonal or even daily cycles (timing statistics of precipitation)
• Reproduction of probability function between low and high precipitation events (the right weighting between high and low events)
• Frequency of occurance of extreme events (comes back to the question of @Kenneth M Towe)
• Reproduction of precipitation type: drizzle, rain, snowfall, convective vs. large scale precipitation (the processes leading to precipitation).

I would like to know what do you have actual in mind, because if you are more clear about that, an answer could be found with more easily.

Hi Ahmed, to select the best model from many, you have to correlate the historical data from each GCM with the station observational data & pick the model with the highest coefficient. Normally a coefficient greater than 0.75  is good.

When you  plot time series curves for the observed station data & the best model you have chosen, the two curves will assume a similar pattern. 

Dear Awais,

This is a deep subject. Simply, draw a Taylor diagram and/or use multi-model ensemble of GCMs.

Best,

Ahmed. You need to decide first if you are studying weather or climate. By definition, GCMs cannot reveal anything on an annual basis. That is not what GCMs are for.

Monsoon rainfall can be studied for annual variations. A research who is a member of Researchgate claims there is a strange attractor related to chaotic monsoon activity. There are variations in ENSO that modulate rainfall. There is a biennial oscillation.

Lots of ways to approach extreme rainfall events. But GCMs won't get you anywhere.

Multi-model ensembles will not help you either, because the main feature of the ensembles is (1) they are based on parameterizations (2) the modeling groups do not agree on the values of the parameters. This is why the CGMs have not converged for 30 years. 

Have a look at Belda et al (2014)., Climate classification revisited: from Köppen to Trewartha

Free access to PDF file: URL: http://www.int-res.com/abstracts/cr/v59/n1/p1-13/

Soon after Belda published the CRU at the UEA revised the rainfall dataset to correct for wet bias. As a consequence, fewer of the global geographic cells changed climate category compared to Belda's estimate (8%). 

I interpret this to mean that you need to find data that has not passed through a low-pass filter that does not smooth the annual variations. Also you do not want the output from models that do not attempt to project annual variations.

Dear Ahmed,

you should first spell out precisely what you mean by "annual basis". Do you mean annual total precipitation, and the extremes thereof? Or do you mean annual maxima of, e.g., daily precipitation? Because this very much determines the answer to your question. GCMs do not resolve processes at the meso and small-scales. They might pretty well simulate a cyclone, they will do badly with organised convection and they will fail to simulate local thunderstorms. Thus the suggestion to maybe use CORDEX RCMs is a good one - but note that also these have a limited resolution of typically between 12 and 50km.So for extremes of total annual rainfall - if this is not dominated by small-scale processes, a GCM might do well. For regional scale processes an RCM might help, but it is not clear yet whether it will give you a correct picture of anything related to convection.  Note that statistical downscaling may get you the effects of local orography in present climate, but it will essentially inherit the large-scale climate change signal of the dynamical model - which might be wrong at the local scale. E.g., there is evidence that the climate change signal is different on different sides on a mountain.

In any case, for sampling as much uncertainty you need ensembles of models! Anything else would not be defensible.

There are no simple metrics to select models for your ensemble - similarity of precipitation in present climate with observations is not an indicator of a good performance in a future climate. You will have to choose models that get the large-scale (e.g., Monsoon circulation, Westerlies, ENSO,...) and local-scale processes (e.g., influence of orography, convection) sufficiently well simulated.

Some literature:

Stainforth et al., Phil Trans Roy. Soc A, 2007 (on uncertainties)

O'Gorman, Curr. Clim. Change Rep., 2015 (review on extreme precip under CC)

Meredith et al., J. Geophys. Res., 2015 (on difficulties of standard RCMs to simulate the response of extreme precip to warming)

Maraun and Widmann, Camb. Univ. Press, 2017 (forthcoming, in depth discussions on all the issues presented above)

Best,

Douglas

Extreme weather events on an annual basis are still weather events and cannot be predicted by any GCM.

The World Bank has a climate website covering many countries, such as Cambodia. The WB says that, for Cambodia, precipitation cannot be predicted with confidence. This is probably true for most countries.

The CRU at the UEA has recently revised their precipitation model to correct for precipitation bias (too wet).

The 3rd Assessment report of the IPCC stated that climate is non-linear and chaotic.

If you are looking for an area of research that has a future try the oceanic oscillations and their relationship with weather and climate.

I am referring to the AMO, PDP, ENSO, etc. A senior researcher has posted on Researchgate a paper suggesting that the Indian Ocean monsoon is governed by a strange attractor, i.e. is chaotic.

But since GCMs do not account for cyclical natural climate fluctuations, you cannot use GCMs for this. You will actually have a chance to do something original science because this is a field that has been under-researched.

Dear Ahmed there is no best single model especially on predicting precipitation. Different models behave differently. I would advice you to use several models to see their variability. 

Dear

Awais Ahmed

in my poster ,Title:Left (Right)-Tail L Probability Distribution Function and Its Application in Extreme Drought (Wet) Index,just uploaded recently in my contribution,

the  methods of "a new extreme meteorological wet index from right-tail L distribution" may be suitable for your "Criteria for selection of GCMs for extreme precipitation events on annual basis", in detail, XA may be averaged precipitation value of some month of long term(long years), for example, for June, for July, or for August, may be also for JJA(summer).........

yours

wanli wang

21-Sep-2017

Poster Left(Right)-Tail L Probability Distribution Function and its...