Climate models are tested by comparison of model predictions with real-world observations. For this purpose, climate models are run over a historical period, from around 1850 to near-present, using best estimates for the past forcings  during this time period (also see “What are the inputs and outputs for a climate model?”). These “hindcasts” of the past climate (e.g. surface temperatures)are then compared to actual recorded climate observations (Fig. 1c). The more precise the hindcast of past climate, the more reliable is the climate model, also in forecasting future climate.

These historical “hindcast” runs can also be used to determine human influence on climate change, the so called “attribution”. For this purpose, models are run with either only natural forcings (e.g. solar variation and volcanic activity) or anthropogenic forcings (e.g. greenhouse gasses and aerosols) as model inputs (Fig. 1a, b). These graphs show that natural forcings alone can’t explain climate’s behavior. Only when we also take anthropogenic forcing into account, we can explain the observed climate patterns.

Furthermore, big perturbation events like volcanic eruptions can be used to test climate model performance. Model projections can be compared to recorded short-term climate responses after an eruption. Studies focusing on the Mount Pinatubo eruption show that models can accurately project changes in temperature (Hansen et al., 1996) and atmospheric water vapor (Soden et al., 2002).

To produce more reliable estimates of twenty-first century climate, climate models are also tested against paleoclimate data (reaching back up to 21,000 years). This data shows larger climate changes than the observational record of the last 150 years, against which climate models are normally evaluated. Ice-core, marine (e.g. marine sediments) and terrestrial archives (e.g. tree rings) provide information about environmental responses to past climate changes. These records can be used to derive estimates of climate, i.e. provide paleo-proxies for past climate (e.g. paleotemperatures). Thus, the geologic record provides a unique opportunity to test model performance outside of the comparison with the short-term observational record. Evaluation of model simulations against paleodata shows that models reproduce the direction and large-scale patterns of past changes in climate, even though they tend to underestimate the magnitude of regional changes (Braconnot et al., 2012).

Sources

https://www.carbonbrief.org/qa-how-do-climate-models-work
https://skepticalscience.com/climate-models-intermediate.htm
Braconnot, P., Harrison, S. P., Kageyama, M., Bartlein, P. J., Masson-Delmotte, V., Abe-Ouchi, A., . . . Zhao, Y. (2012). Evaluation of climate models using palaeoclimatic data. Nature Climate Change, 2(6), 417-424.
Hansen, J., Sato, M., Ruedy, R., Lacis, A., Asamoah, K., Borenstein, S., . . . Campbell, M. (1996). A Pinatubo climate modeling investigation. In The Mount Pinatubo Eruption (pp. 233-272): Springer.
Soden, B. J., Wetherald, R. T., Stenchikov, G. L., and Robock, A., 2002, Global Cooling After the Eruption of Mount Pinatubo: A Test of Climate Feedback by Water Vapor: Science, v. 296, no. 5568, p. 727-730.