Simulating Future Transient Climates By Combining Observational Data with Climate Model Information Using Time-Varying Spectral Methods

Abstract:

Emulation methods typically capture patterns of regional mean climate change predicted by a general circulation model (GCM), but impacts assessments depend on changes in variability in addition to changes in mean climate. GCMs do predict future changes in variability, but do not reproduce the means or variability of observed climate. Our work addresses the ensuing need for climate simulations that combine observational data with GCM projections of changes in mean and variability. This work is part of a developing framework for simulating future climates, which includes statistical methods described in Castruccio et al. (2014) for mean emulation of GCMs. Leeds et al. (2014) additionally introduced a spectral-based methodology for modifying existing temperature observations based on GCM projections of changes in means and variability in future equilibrium climates. Here, we extend the methodology in Leeds et al. (2014) to account for GCM projections of transient climates. Other existing methodology can account for changes in mean climate (the “Delta method”) or model-observation discrepancy in marginal probability distributions (quantile-based methods), but a key advantage of the methodology proposed here is that it accounts for changes in temporal dependence, and therefore in different changes in variability at distinct timescales. Our methodology can be combined with emulation to study changes in unmodeled forcing scenarios and impacts under these scenarios, assuming that the changes in covariance structure can be described by a simple function of the forcing scenario. We illustrate our methodology with runs from the NCAR CCSM3 model, comparing daily temperatures under continuously increasing CO2 forcing scenarios with temperatures from a run forced at preindustrial conditions. In most regions, temperature’s overall temporal variability decreases as temperature increases, and furthermore, its temporal dependence structure changes in time.