Yes we have done this with our NASA research. See Yan Yun Li and Sang-Ki Lee as they have done this at NOAA AOML in Miami. Contact them for details.

Thank you Dr. Mitchell!. I will get connect to them.

In principle: yes, however, one should be confident that the predictor used in the SDSM is skillfullly reprsented in the CMIP model simulations. This is a non-trivial but necessary assumption.

The point made by Hans is very important. Be careful to pick up the model you are going to use.

A few guidelines to avoid common mistakes:

- Appropriate choice of the predictors: be sure that there is a strong physical relationship between the predictor and predictand. Check carefully by comparing to observations and/or reanalysis how this relationship is represented in the models that you are going to use to perform the SDSM (so check not just the predictor but also its relation to the predictand).

- A critical point: do use ***several*** models and reanalysis/observation data sets to construct your transfer function. You have to assess uncertainties as best as you can. For hydrological cycle variables, there can be huge differences between reanalysis data sets.

- Stationarity assumption: it can (and should !) be checked in a perfect model framework. If you take a CMIP5 model with historical and scenario runs, use one of the historical run to build your SDSM (as if it were observations), apply the SDSM to the model predictors in the scenario run and compare the output of the SDSM to the simulated data. Note that this is a necessary (not a sufficient one) condition for the validity of your method.

Best, Laurent

Important guide made by Laurent Terray! We were also trying to use SDSM to downscale CMIP5 model output, it worked properly until weather generator for baseline scenario. However, we have stuck/ unable to generate scenarios for future scenario and still could not understood what the error is? Any suggestion on this.

Yes, we can use CMIP5 model output in SDSM for downscaling.  I have done with CanESM2 model for precipitation and temperatue. Mr.A.Shawul please split those data into different periods. and check the number of days in a year for your GCM model output and change accordingly in your global settings of SDSM  and simulate for the future . I think this may help you. 

Yes, Alemayehu A. Shawul, Dr. Jeew is right. I am also doing it. But I think we have to use SDSM 4.2 version instead of SDSM 5.2 as later one has no option for generating future scenario. If we can do it in SDSM 5.2 as well, would you please let me know Dr. Jeew, where can I find a place to select GCM (CanESM2 model) for generating future scenario. And one more request to Dr. Jeew, would you mind briefing how to add the leap year days weather data in scenario generated dataset, as CanESM2 GCM output consider 365 days and we have to add leap year days data from 2006 to 2100.

We have to use SDSM 4.2 version  as SDSM 5.2 does not take any GCM out put for future simulation. The reason is the uncertainty involved in the GCMs. But SDSM 5.2 has the ability to find out missing data and if you have continuous observed data you can give different treatments to find out what is going to happen in different stress conditions without taking  any GCM outputs. The link for the CanESM2 data set is  "http://www.cccsn.ec.gc.ca/?page=download-intro". No need to add any extra day for leap year as CanESM2 is simulating for 365 days. please change the global settings to 365 days while performing scenario generate and you can compare both historical as well as scenario generated results after simulating. By the way one day won't make any change.

Thanks Jeew for your quick response. But our observed climate data have 366 days so there will be problem in weather generator while using CanESM2 historical data for calibration and validation. So I am using NCEP-NCAR reanalysis data for calibration and validation and only using CanESM2 for scenario generation. Is it o.k?

For GCM output you can use scenario generator for historical as well as for  future  and for NCEP you have to use weather generator. Because we are screening the predictors only using our NCEP predictors. But you have to change global settings before doing scenario generator.

Using

 http://www.cccsn.ec.gc.ca/?page=download-intro

I was able to obtain predictors for CMIP5 (CanESM2) but only for few sites. Other sites just gave me AR4 (CGCM) predictors. Any place I can find CMIP 5 predictors for sites located in different parts of the world? Thanks!

Can CMIP5 (CanESM2) be used for statistical downscaling of Temperature using SDSM? I have downscaled precipitation and shows satisfactory results but not same for Temperature. 

@Harshana, yes, you can use it. Infact, the efficiency of SDSM is better for temperature variables than precipitation.

Thanks, but i am getting very poor result. When i tried to downscale Tmax, it gives many negative values. There are no negative values in observed Tmax but in calibrated values, there are many negative values and also high fluctuation of temperature.

@Harshana, you can check methodology part. I have explained all the processes in detail. And, you can contact me via [email protected], if u have any problem.

Thank you very much. I will check it.

For temperature mark the unconditional process, model transformation-none and dual simplex optimization. try this and see the results.

It is really nice explanations, @ Rajesh I sent to you an email please check it.

I would appreciate if any body can tell me where I can get CMIP5 predictors of different GCM for SDSM for site located in East African country.

RCP scenarios of CanESM2 global model data to feed SDSM can be downloaded from this link:

http://www.cccsn.ec.gc.ca/?page=pred-canesm2

Thanks Iman, I got it

I have a question and I hope that somebody can help me. I am trying to make a calibration (using CanESM2, SDSM and NCEP data) for precipitation data but it does not work when I chose for 1961-1990. Whereas, it works when I applied for the whole period (1961-2005) could anybody assist me? I really need your advice.

I do not have this problem with Temperature data.

Thanks in advance 

Dear Sada you only need the predictors not the data. Predictors can be found in the http://www.cccsn.ec.gc.ca/?page=pred-canesm2. so far the predictors are available for one model CanESM2. Good Luck!!!

There are more than 30 GCMs available in CMIP5. Therefore it is not reliable to use only one GCM. Better to take multiple GCM and see the future projection with the ensemble mean of all the model.

Dear Ayesha Anwar yes the predictors are available for Pakistan, but you need to choose the best predictor that fits your observation.

@Ayesha Yes i guess they are suitable for Pakistan, since the GCM is applied globally; but IPCC highly recommends using an ensemble average of several model outputs. This is the main reason i have not been able to work with SDSM on my current project despite the fact it has a wonderful set of statistical tools. 

I recommend you check out MarkSim weather generator, it has about 17 GCM model predictors for IPCC AR5 (RCPs).

You have to take the projected mean not the predictor mean. There are several methodologies are available. and uncertainty analysis is also plays an important role in climate change analysis. you can go through Reliable Ensemble Averaging (REA) method.

You can use the CanESM2 GCM Predictors from this url: http://ccds-dscc.ec.gc.ca/?page=pred-canesm2. Use SDSM v4.2 to select the predictors to use in generating current and future climate scenarios.

 Greetings Everyone

I'm working on a site in USA for with I've got observation data of pcp, Tmax and Tmin for 1950-2007. I also downloaded all Climate Change models from this site:

http://gdo-dcp.ucllnl.org/downscaled_cmip_projections/

The CC projections are for 1900-2100 period and, as expected, the 1950-2007 part data doesn't match with observations, so I'll need to downscale them based on observed data.

The question is this: How can I downscale CC projections data based on observed data? I used SDSM but got profoundly sparse and unrelated answers.

Thanks a lot.

You may use the SDSM v4.xx software and canESM2 GCM (RCP2.6, 4.5 and 8.5)

Dear Mr.Chisanga, Thanks for your help.Unfortunately I can't use SDSM or any such software because I have to downscale too many datasets (300 to 400 sets) so I will need a MATLAB code to run that once. Do you have any idea how can I do this? Thanks a lot.

Dear Majid,

You should develop your downscaling method in MATLAB or R packages. For example there is a package for bias correction in R namely QMAP using Quantile mapping method. (parametric or nonparametric)

The procedure will be the same for other mathematical methods.

Hope it helps.

Yours,

Dear Farzad, Thanks for your assist. I appreciate that.

You can also try the R package: RMAWGEN, RClimMAWGEN in addition to qmap

Thanks all; since I'm not conversant with R, doesn't any MATLAB code exist regarding this problem? thanks again

Dear all, does anybody know how to get CMIP5 predictors? Is it available yet? I could not find it from ESGF website. Thank you.

Dear Rajesh Sada

Yes of course, you can use it

Dear Rajesh Sada,

that depends on the setting you use.

If you use SDSM as a (perfect prog) statistical downscaling method, it makes sense to directly use the GCM output. But you need to be sure that the so-called perfect prognosis assumption is valid, i.e., that the GCM simulates realistic predictors. Typically you need to have large-scale predictors from the free atmosphere such as T850, Z850 etc.

If you use SDSM as a change-factor whether generator you should better NOT use GCM output. I expect you are working over Nepal, where the topography is not at all resolved by the GCM. Here the climate change signal of the GCM might be extremely unrealistic. An RCM (under the condition that the GCM at least simulates the large-scale climate correctly over central and Eastern Asia) might provide a more realistic signal.

You may want to have a look at my review paper (Maraun et al., Rev. Geophys. 2010) or my forthcoming book: Maraun and Widmann, Statistical Downscaling and Bias Correction for Climate Research, CUP, 2017.

Best,

Douglas

Dear all, does anybody know how to get CMIP5 predictors ,other than canESm2 GCM model.Kindly share the link

Dear all, does anybody know how to get CMIP5 predictors ,other than canESm2 GCM model.Kindly share the link

Dear All

Is there anyone who has downscaled different GCMs decadal experiment data of CMIP5???

hello,

if anyone want to download CMIP5 data other than CanESM2 gcm please visit this website https://esgf-node.llnl.gov/projects/esgf-llnl/.

Hi Sonali Kadam,

I am checking the website but the number of parameter available for SDSM coming NCEP data are different to the parameters of other GCMs available from https://esgf-node.llnl.gov/projects/esgf-llnl/. Could you please guide how we can manage to use SDSM in this context?

Regards

I would recommend the Long Ashton Research Station Weather Generator. This stochastic weather generator uses CMIP5 GCMs indirectly as long as you have historical station data (e.g. 1980-2010) . url: https://sites.google.com/view/lars-wg/

Hi dear Sonali Kadam

I am checking the website( https://esgf-node.llnl.gov/projects/esgf-llnl/ ) but I can't find download path CMIP5 data. Can you guide me?

thanks

Hello Zahra Shirmohammadi Aliakbarkhani

If your problem hasn't solved yet, you are welcome to inbox me at [email protected] regarding downloading data from esgf website.

sent an email. to you