linked in logo | IS-ENES | Contact

Exploring climate model data

Generic processing of climate data for use in impact assessments

The following diagram lists all the neccessary and optional processing steps that you need to do to get the dat you need and make them usefull for your purposes. Most of these steps are facilitated by this portal. Be aware that some of these steps require considerable climate expertise and you may ask advice through our Contact the Expert form. The diagram is rather complete: some steps are almost trivial and do not take much of your time, others can be significant tasks in themselves.

In the diagram continuous boxes indicate neccessary steps, dashed boxes optional steps ansd dashed-dot boxes alternative options. Blue line represent primary data flow, red lines derived information flow. Some steps require sector specific impact expertise, others climate expertise (coloured arrows).

Please realise that generally you need three types of data:

  1. observed climate data for your region of interest
  2. climate model data simulated for the same period covered by your observed data 
  3. climate model data simulated for the future

The first two you need to assess how good the choosen model is able to reproduce the relevant weather variable in the current climate, its skill, for the area of your interest. Model skill may be different for variables; e.g. generally skill for temperature is better than skill for precipitation, or phrased altenatively: generally bias for temperature is smaller than the bias for precipitation. Model skill will definitely be different from one area to the next; e.g. skill is often less in mountainous areas compared to flat areas. At some stage you may need to correct your future climate data for such biases.

Generic processing workflowVariable and Domain selection

Depending on your research question you must define the area that you need data for (e.g. the Alps or the Iberian peninsula, or the the city of London, etc), the time horizon that you need data for (e.g. 2070-2100) and of course the neccessary variable your impact model needs (e.g. temperature, precipitation, sea level pressure, etc). We recommend that you do not choose too small an area, it generally should cover a number of gridcells of the climate model you choose, so for GCM data on the order of 1,000,000 km2 or for RCM data about 10,000km2. Similarly, do not choose too narrow a time slice. To get a statistically representative period you need at least 30yrs of data. Alternatively you may use ensemble data to increase your temporal sample.

Scenario selection

You need to make a choice which type of global development pathway your impact study will be placed in. You can choose from the different SRES scenarios or the different RCP scenarios (see Scenarios for more information).

Model selection

You need to make a choice which climate model produced the data you need. That choice may depend on model skill, on sampling the spread in climate sensitivities, institutional preferences, etc. (see Which GCM to select for more information).


You may need to match the grid between the model data and the observed data or between data from different models. This may involve re-projection from e.g. a sterographic projection to a regular lat/lon grid, interpolationTools to do so are provided on this portal [to be implemented] or can be found and used offline, e.g. CDO (later on more information).

(Dynamical) downscaling

Climate model data are often of a too coarse resolution to make them directly usefull in impact studies. You may need to downscale them. In dynamical downscaling you use a regional climate model to do so. THIS IS NOT A TASK AN IMPACT RESEARCHER NORMALLY DOES. Instead you can use RCM data produced by others, e.g. from the CORDEX project (see Regional models for more information).

Skill assessment

For each climate variable that you need in you impact model you need to assess any systematic biases generated by the GCM of your choice. You need climate data simulated for the past/present by the same model and compare them against observations. Your climate model data may be e.g. 2K to warm, or producing 30% to low precipitation, in the region of your choice. The quality indicator you get in this way can be used either for model weighting (see below) or can be used to correct future data (see Bias correction methods for more information).

Skill assessments for seasonal and decadal predictions differ from those for climate projections in two aspects. The first is that systematic biases are not constant in time, but generally increase with Forecast time, due to model drift. The second is that since we are dealing with real forecasts, skill should also be quantified in terms of how often the forecast is right or wrong. This type of analysis is generally called verification. (see Verification methods for more information).

Model weighting

Not all climate models perform equally well in reproducing (any aspect of) the current climate. When you create stochastic climate data you may combine data from various models and weigh each according to its skill. This is a complicated task (see e.g. Weigel, 2010), please ask advice from a climatologist with the relevant expertise (Contact the Expert) and more information will follow.

Bias correction

With systematic biases quantified you should now correct for these. In its simplest form you simply add or substract a bias (delta correction, e.g. for temperature) or you add or substract a relative bias (scaling, e.g. for precipitation), generally on a month-by-month, gridcel-by-gridcel basis. Much more complicated methods exist (see Bias correction methods for more information). Though the basics are simple, experience helps in the process, don't hesitate to ask advice (Contact the Expert). Statistical downscaling implicitly includes also a form of bias correction; so when you perform this type of downscaling (but NOT when you use dynamical downscaling results) you can skip bias correction as a seperate processing step (see Statistical downscaling for more information).

For seasonal forecast, bias correction takes a slightly (but crucial!) different form in that biases not only vary by month, but also by Forecast time. Otherwise the methods are similar.

Indices calculation

From these climate data you may finally need to calculate any indices (percentiles, averages, extremes indicators) as needed by you own impact model.


You may need to match the final climate date processed as above to the specific imnput format requirements of you impact model.


Weigel, Andreas P., Reto Knutti, Mark A. Liniger, Christof Appenzeller, 2010: Risks of Model Weighting in Multimodel Climate Projections. J. Climate, 23, 4175–4191. doi: .



The ENES3 project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 824084.