Validation of climate simulations
Climate simulations are produced by two main types of physical climate models
As their name suggests, global climate models (GCMs) cover the entire globe. Originally, when they could only take the atmosphere into account, they were called general circulation models. They have gradually been improved to integrate the other climate system components. Since the inclusion of the carbon cycle, they have been known as Earth system models (ESMs). These models do not require observations of the various climate variables that they are able to simulate, but they do rely on a description of the land surface (topography, bathymetry, soil and vegetation types, etc.) compatible with the resolution of their calculation grid. They also need the composition of the atmosphere, including greenhouse gas (GHG) concentrations and aerosols.
Regional climate models (RCMs) cover part of the globe and have a lot in common with ESMs. Covering a smaller area allows for increased horizontal resolution to capture more detail in land surface characteristics. However, in order to remain connected to global atmospheric and ocean circulations, an RCM needs to be driven, i.e. fed with data at the boundaries of its grid (FAQ 9). Thus, a climate simulation must always be identified by the name of the regional model as well as the driving GCM.
Models can produce historical and future simulations
Historical climate simulations are produced based on GHG concentration observations provided to the models. With these historical simulations, a model’s ability to properly represent the various characteristics of the climate can be assessed. However, their main role is to serve as a reference for estimating climate change.
Climate simulations covering a future time period, also called climate projections, are produced from the various GHG emissions scenarios. Some models may use emissions directly, while others use the concentrations corresponding to them.
Historical and future simulations are used to estimate climate change under a given GHG scenario by comparing the characteristics of the simulated future climate to those of the historical climate simulations.
The performance of climate models is evaluated based on their historical simulations
Except for GHG concentrations, note that no other observations are provided to ESMs to produce a historical simulation. Under these conditions, given the chaotic nature of the climate system, it’s impossible for an ESM to reproduce the same sequence of meteorological events (same time series or same specific events that occurred in reality) as what was observed at any point in the world. Instead, the role of an ESM is to reproduce characteristics of the observed climate like averages, variability and extremes. That makes it possible to compare the statistics over several decades of the historical simulation with those obtained from observations, but not to make temporal correlations between the simulated and observed data series.
For RCMs, the source of the data used to drive them determines how reasonable it is to compare one of their historical simulations with real observations:
1. If the regional model is driven by a reanalysis, a certain level of temporal correlation between simulated and observed events is expected, as the reanalysis is in fact based on a large number of observations from various sources. In addition to comparing the averages, variability and extremes, it’s possible to look at the level of temporal correlation between the simulated and the observed series. It’s also possible to assess whether the regional model succeeds in reproducing an event that occurred on a specific date. Performance expectations should be adjusted according to the level of compatibility between the size and duration of the observed event with the spatial and temporal resolution of the regional model. Consequently, better performance is expected from an RCM for climate variables or hazards that depend mainly on large-scale processes and those with a broader timescale, while poorer performance is expected for those that are strongly influenced by very local effects and those of very short duration.
2. If the regional model is driven by a simulation produced by an ESM, the possible comparisons with the observed data are the same. That means we can only compare the various moments of the simulated distribution to the observed distribution. Comparisons for specific dates or for the level of temporal correlation are not possible.
Summary of the permissible comparisons between observed data and simulated data for variables, indicators or hazards, depending on the origin of the climate simulations.
