Proxy System Forward Modeling
The Problem: Paleo-Model Mismatch
Paleoclimatologists can use their records for "model verification"; in other words, they can compare paleoclimate records and climate model data to determine if climate models are accurately representing the range of variability in the pre-instrumental climate system. Some researchers have found that the frequency of temperature and rainfall variations often do not match in model and paleoclimate data. Recent studies [e.g. Ault et al., 2014, Laepple and Huybers, 2014] have suggested that there is at least an order of magnitude proxy-GCM mismatch in both temperature and precipitation variability at longer (decadal-centennial) time scales.
For example, the red lines below represent the spectra of CMIP5 Last Millennium precipitation and soil water content, and the green line represents the spectra of a variety of hi-resolution paleoclimate records from around the American Tropics that are sensitive to hydroclimatic variability (i.e., rainfall). Note the disagreement in the variance at longer time periods.
A possible explanation for the mismatch?
Importantly, part of this paleo-model mismatch can be explained by smoothing processes that occur as paleoclimate records form. For example, after rain falls to the ground, it often mixes with older water in a cave system before being deposited as a stalagmite. Using forward models, we can try to quantify the smoothing effects of the formation processes in paleoclimate records so we can better compare model data to paleoclimate data.
However, part of the proxy-model mismatch could also be explained by climate models that underestimate low-frequency variability in the climate system. This is important because low-frequency variations in the climate system give rise to extended dry (or wet) periods.
Working with Dr. Sylvia Dee, I compared paleoclimate and forward-modeled climate model data for more accurate comparisons among these two data sets.
PAGES2k coral data spectral scaling distribution (right) and forward modeled climate model data for the same locations (second from right). I obtain a spectral scaling ('beta') value of a given record by calculating its spectral slope (see spectral plots to left).