Climate modelling

Computer modelling of the climate permits theories to be tested and refined, the historical climate record to be explained, and possible future scenarios to be explored. We can't do experiments with the Earth (unless we call the emissions of the last 200 years an experiment.... a very badly planned one... experiments should be repeatable, and should not damage a planet irreversibly.). A climate model brings together mathematical representations of how parts of the climate system work. A climate model is based on equations for:

• fluids on a rotating sphere (air and ocean)
• incoming solar radiation and outgoing terrestrial radiation
• processes in atmosphere, ocean, land and ice and their interfaces.

These equations are inherently chaotic in that tiny changes in data can propagate into different systemic behaviours: a digital version of a butterfly flapping its wings causing a storm elsewhere, The information given the models is inevitably imperfect and partial. We cannot model every rain drop, for example. There is a mix of partial differential equations based on first principles, and parameterised processes, such as within clouds. Models take a defined state, including pressures, temperature, etc around the Earth then do the calculations to derive a new state for a short time later, then re-calculating from that new state.. These states are evaluated for grid cells, each corresponding to a part of the Earth's surface. A cell is typically ~100km square, at the equator. The data for a cell comprises surface and also column data for the ocean beneath or atmosphere above it. Simulations can be based on historical data or on hypothetical states for example to seek to isolate the impact of one physical process.

The basic science that increasing greenhouse gases will cause warming has been established for many decades. As models became more advanced and data for past climates accrued, so it became clear that modelling was a key to understanding the likely extent of climate change: "global warming" is a term that masks a diversity of effects - seasonal, regional and global, the intensity and frequency of extremes of heat and drought and rainfall,... Although atmospheric warming is the initial consequence of enhanced greenhouse gas concentrations, there are major consequences for ocean, land, ice and life of the planet and so more all-embracing Earth System models were developed, coupled to climate models.

The UN recognised the significance of this work and in 1988 established the Intergovernmental Panel of Climate Change. It in turn set up the Coupled Model Intercomparison Programme to compare results emerging from different research groups, there being a number of models in use across these. Scenarios were defined to allow comparison between models to be explored; standards were established for how the data were held (e.g. variable names and units; file formats for data). The IPCC includes CMIP in its gathering of research from around the world, into reports, the fifth of which was in 2014.

The models are thus used to do experiments in-silico, to explore:

• past climates - can we understand the past as inferred from diverse data sources (ice and soil cores, tree rings; observations from last century; recent observations from satellites and land/sea instruments
• future climate - what happens in different scenarios of greenhouse gas emissions
• imagined scenarios - to understand science, by asking unrealistic questions - like suppose all emissions stopped in 2020, what would the impact be on future climates, compared to other scenarios.
• fundamental understanding of the climate system - to enhance the modelling of the links between of land/ocean/atmosphere/ice; feedbacks such as consequences of warmer air holding more moisture,
• extreme events - are specific periods of drought/flood/high temperature indicators of climate change?

Modelled data can then be used to inform policy-makers, and to infer risks such as to infrastructure, to agriculture (in light of pest phenology), to the economy, etc.

The need to parallelise climate/weather models was recognised by L.F. Richardson around 1920. He imagined a huge number of human computers doing the calculations each for one grid cell, coordinated by a central controller.

That need for parallelism persists, and models of increasing complexity and resolution challenge the most powerful of computers available. At the same time much research can be done with coarser scale models - it depends on the research question.

In Edinburgh we make extensive use of a range of computers and data stores:

The international community shares CMIP data, and we generally access this via a UK service called CEDA, These data are big - even a current generation model will generate many TB; these data can then need to be processed along with the large volumes of satellite measurements, also held by CEDA. In the light of the large data volumes, the paramount principle is that processing should happen close to data (rather than large amounts of data being transferred across networks) We therefore make extensive use of the excellent JASMIN computer installed at CEDA.

Our most intensive model runs are done on the UK high performance computers at the Met Office, or on ARCHER. Mid-scale models can be run on computers like CIRRUS, sometimes termed the tier 2 (ARCHER being tier 1). Coarse resolution models are often run using the University's Eddie computer.

Post-processing of model data happens initially on the computer where the data were generated, to select data or generate statistical outputs. Then depending on the case, derived data might be brought back to Edinburgh and held on DataStore for final processing into plots for publications using a mix of Eddie, School and desktop computers according to need. Data for reuse by other researchers is usually uploaded with metadata into CEDA; some data are also loaded into DataShare. Long-term archiving of data not currently active is achieved using DataVault.

Why don't you explore featured projects demonstrating the use of similar resources and related training opportunities? Have a look at the carousels below.