A Computationally Efficient Approach to Generate Large Ensembles of Climate Data with Coherent Spatial and Temporal Variability

Quantifying uncertainty in the climate system requires large ensembles of climate model output. Such ensembles are available for a handful of scenarios and models, but studies of climate impacts, adaptation, and vulnerability (IAV) need similar ensembles for arbitrary scenarios across multiple models. Computationally efficient, standard pattern scaling techniques only generate one realization and generally produce smooth scenarios with little interannual variability, making the outputs unsuitable for uncertainty and variability studies. In this study, we demonstrate a method for generating ensembles of climate data with spatially and temporally coherent variability, calibrated to model results from the Coupled Modeled Intercomparison Project Phase 5 (CMIP5). The method begins by applying a pattern emulation approach through use of principal component analysis to derive empirical orthogonal functions (EOFs) of local surface temperature and precipitation fields. These EOFs are scaled with an annual global mean temperature change field, derived from a reduced complexity climate model, to produce a 3-D field of expectation values. To incorporate internal variability we add another 3-D field of random normal deviates generated in such a way as to have the same spatial and temporal autocorrelation function as the target CMIP5 model. These autocorrelation properties are what ensure that the new emulated patterns reproduce the proper timescales of climate response and memory. We can repeat this process as many times as desired to produce any size ensemble of independent, yet space- and time-coherent, climate scenario realizations.