An introduction to the PRECIS system
Introduction to PRECIS, its benefits, applications and the science behind the model.
As of January 2020, the PRECIS team are no longer taking on new PRECIS users, nor able to provide technical support to existing PRECIS users. We are taking this time to regroup and redevelop what our training and modelling capabilities can offer.
Please refer to the PRECIS user forum for solutions to common technical problems.
PRECIS has been utilised in a wide range of climate-related studies and impacts research projects across the world, including analysis of droughts in the Horn of Africa, climate change impacts in Bangladesh, and building resilience to climate extremes in the Philippines. Case studies regarding the application of PRECIS can be found in the PRECIS projects and publications page.
The science behind PRECIS
PRECIS is a regional climate model (RCM) that takes large scale atmospheric and ocean conditions from observations or global climate models (GCM) where horizontal resolutions vary from 100 to 300kms, and downscales it over a region of interest to resolutions of 25 or 50km. This allows for a more realistic representation of the climate over the region of interest, accounting for complex surface features such as mountains, coastlines and islands which are not resolved in the global models.
Click the headers below to find more information on climate modelling and PRECIS.
What is downscaling?
Downscaling is the process of adding high resolution information to large scale projections. The two main downscaling techniques are statistical and dynamical, which can be combined to produce a third statistical-dynamical technique.
Statistical downscaling assesses the relationship between large-scale and local climate variables to estimate local and regional climate characteristics. This technique is computationally inexpensive and provides climate information at point locations. A disadvantage to this technique however is the long time series of relevant data needed to form relationships between variables, and the uncertainty as to whether the relationship will be applicable to the future climate.
Dynamical downscaling uses models of the physical climate system to directly model physical processes down to regional scales. This technique produces high resolution information on a large set of climate variables, including a better representation of climate extremes. Due to the model's complexity and the range of output produced however this method is computationally expensive and requires large amount of input data as boundary conditions. The other main limitation is that with any model, the representation of the physical system will be incomplete resulting in output bias which must be accounted for when the model is interpreted or applied.
Both techniques also suffer from the limitation that any biases in the driving data will be passed on to the downscaling model.
Why use a regional climate model?
A regional climate model (RCM) is a downscaling tool that adds higher resolution (50 or 25km in the case of PRECIS) information to large-scale projections observations of the climate or simulations of current or future climates generated at coarser horizontal resolution (typically 100 to 300km). RCMs are full climate models, and as such are physically based. They represent most if not all of the processes, interactions and feedbacks between climate system components as represented in GCMs. They produce a comprehensive set of output data over the model domain.
Values of a regional climate model:
- accounting for local topographical features affecting the climate subsequently producing more detailed climate change projections;
- better representing certain regions such as small islands which have previously been represented by the surrounding ocean; and
- better representation of meso-scale weather features and extreme events.
Limitations and uncertainty:
Uncertainties in regional climate modelling can be associated with the downscaling process, the model itself, and the climate scenarios used to produce climate change projections. Many of these uncertainties are accounted for through producing ensembles using different climate scenarios, multiple realisations and driving GCMs. Inherent uncertainty passes from the driving GCM due the coarse resolution leading to errors in the data when downscaled for a regional climate, for example for coastal points and inland seas, variables must be interpolated or extrapolated leading to error.
- Uncertainty in future emissions through assumptions of the future socio-economic and technological situation.
- Uncertainty in the response of the climate to various drivers of change, due to an incomplete understanding of climate system processes and their representation in the model.
- Uncertainty due to natural variability on temporal scales. Natural variability may add to or take from changes due to human activity at any point in time.This can be quantified by running multiple realisations of the climate of a given period (e.g. 30 years) to account for climate responses to natural variability.
Understanding these benefits and limitations leads to more accurate interpretation of model output. As such adaption decisions should be based on a range of climate scenarios accounting for uncertainties associated with projections. This range of scenarios can be collected through collaboration with neighbouring scientific groups and countries assessing a similar regional scale.
Formulation of the PRECIS model
The scientific formulation of PRECIS is based on the atmospheric component of the Met Office Hadley Centre's coupled climate model, HadCM3.
What hardware and operating system is required to run PRECIS?
PRECIS has been developed to run on PCs under 64-bit Intel (x86-64) compatible Linux-based systems. No particular Linux operating system is recommended; several distributions of SuSE, RedHat, Ubuntu and Debian have been tested successfully. Dual-boot PCs can be used, but bear in mind that the Windows partition will be unavailable unless the experiment is interrupted and the machine re-booted.
At least 512MB physical memory is required, and 1GB or more is recommended.
A minimum dual core CPU is required - more cores means faster run rate as PRECIS can run in parallel mode across cores.
Depending on the output options chosen, a typical 30-year PRECIS experiment generates between 130 and 540GB of output data, whilst the boundary conditions for a thirty-year experiment occupy 40GB. A hard disk size of at least 500GB is therefore recommended, although with careful data management by the user, smaller disks can be used.
If power supplies are unreliable, PRECIS can be shut down cleanly (ready for a re-start) given approximately a 1 hour notice. If shut down without warning, the model can be re-started from the beginning of the model month in progress, which might involve repeating up to typically twelve (real) hours of simulation. A UPS (uninterruptible power supply) may therefore be appropriate if power outages are frequent.
How fast is PRECIS?
A typical experiment, covering a 100-by-100 grid box domain run on 2 cores in parallel, takes 1 month to complete a 30-year simulation. With 4 cores it will take about 2 weeks and 8 cores about 1 weeks.
What output does PRECIS produce?
A comprehensive set of variables has been chosen for PRECIS's default output - for a listing, please see Appendix C of the PRECIS Technical Manual.
Can PRECIS be parallelized or run on a cluster?
The current version of PRECIS runs on multi-core cluster systems in parallel MPP mode.
PRECIS will not run as installed on cluster systems using scheduling, queuing or job control software without extra effort from the PRECIS team which will incur additional costs. It could be much easier and cheaper to run PRECIS on a multi-core desktop rather than a dedicated cluster.