Using Spectral Methods to Quantify Changes in Temperature Variability across Frequencies

Changes in future surface temperature variability are of great scientific and societal interest. Since the impact of variability on human society depends on not only the magnitude but also the frequency of variations, shifts in the marginal distribution of temperatures do not provide enough information for impacts assessment. Leeds et al (2014) proposed a method to quantify changes in variability of temperature at distinct temporal frequencies by estimating the ratio of the spectral densities of temperature between pre-industrial and equilibrated future climates. This spectral ratio functions well as a metric to quantify temperature variability shifts in climate model output.

In this study, we apply the method of Leeds et al (2014) to explore the temperature variability changes under increased radiative forcing. We compare changes in variability in higher-CO₂ climates across two different climate models (CCSM3 from the National Center for Atmospheric Research and GISS-E2-R from NASA Goddard Institute for Space Studies), and changes driven by two different forcing agents (CO₂ and solar radiation) within the same model (CCSM3). In all cases we use only the equilibrium stages of model runs extended several thousand years after an abrupt forcing change is imposed.

We find a number of results. First, changes in temperature variability differ by frequency in most regions, confirming the need for spectral methods. Second, changes are similar regardless of forcing agents. In experiments with abruptly increased CO₂ and solar forcing designed to produce the same change in global mean temperature, the distributions and magnitudes of spectral ratio changes are nearly identical. Finally, projections of variability changes differ across models. In CCSM3, temperature variability decreases in most regions and at most frequencies. Conversely, in GISS-E2-R, temperature variability tends to increase over land. The discrepancy between CCSM3 and the GISS-E-R highlights the need for further inter-model comparisons of variability projections. This study provides a potential framework for such comparisons.