FaIR

FaIR (Finite-amplitude Impulse-Response) is a simple climate model, also called an emulator: a small model that can be tuned to mimic the global behaviour of larger, more complex climate models.^1,4 Given a scenario, it produces projections of global mean temperature.^1,3 The scenario can be given as emissions of greenhouse gases and short-lived climate forcers, as concentrations of greenhouse gases, or as effective radiative forcing (the extra energy a gas or particle traps), and different gases can be given in different ways in the same run.^1,3 Users can define their own gases and every component is optional, which allows experiments such as the response to a single pulse of one gas.¹

Version 2 is described in a 2021 paper in Geoscientific Model Development by Leach and nine co-authors.^1,4 That version comes down to six equations.⁴ Five are the standard impulse-response model used for greenhouse-gas metrics in the IPCC's Fifth Assessment Report.⁴ The sixth lets each gas cycle's response time depend on the state of the system, which is needed to reproduce the carbon cycle's non-linear behaviour.⁴ The equations are simple enough to write into a spreadsheet such as Excel, yet can be tuned to imitate key processes in the CMIP6 climate models.⁴ The model runs quickly enough for large probabilistic ensembles, and the paper constrains a million-member ensemble against current estimates of the global warming trend.⁴ Earlier versions are described in papers by Smith and co-authors (2018, version 1.3) and Millar and co-authors (2017, version 1.0).¹

It is a Python package for Python 3.10 to 3.14, installed from conda-forge or with pip, under the Apache-2.0 licence.¹ A run is set up by naming its time span, its scenarios and its parameter sets, then filling in the data; many scenarios and parameter sets run side by side, and inputs and outputs are labelled arrays in xarray.³ Time steps need not be whole years.³ By default the gas cycle uses four boxes and the energy balance uses three ocean layers, and both can be changed.³ Example notebooks come with the code and can be tried online through Binder.¹ The documentation credits the fair development team, the code lives under the OMS-NetZero organisation on GitHub, and version 2.2.5 was released on 15 October 2026.^2,5

FaIR will run whatever climate it is given, even one that breaks the laws of physics, so its makers suggest plotting the historical temperature record over any result as a check.¹ For published work it points to calibrated, constrained parameter sets consistent with the IPCC's Sixth Assessment Report (Working Group 1), published on Zenodo.¹

FaIR answers one question about a place only indirectly: how much the whole world warms under a given emissions path.¹ A local model of heat, flooding or crops can take that global path as its starting point, choose the scenario that matches a policy being discussed, and then turn to regional projections for the local detail; see Weather and climate.