Abstract
Changes in internal variability of seasonal and annual mean 2-m temperature in response to anthropogenic forcing are quantified for a global domain using climate models driven by a twenty-first-century high-emissions scenario. While changes in variance have been quantified previously in a univariate sense, the field significance of such changes has remained unclear. This paper proposes a new field significance test for changes in variance that accounts for spatial and temporal relationships within the domain. The test proposed here uses an optimization technique based on discriminant analysis, yielding results that are invariant to linear transformations of the data and therefore independent of normalization procedures. Multiple significance tests are employed because spatial fields can differ in many ways in a multivariate space. All climate models investigated here predict significant changes in internal variability of temperature in response to anthropogenic forcing. The models consistently predict decreases to temperature variance in regions of seasonal sea ice formation and across the Southern Ocean by the end of the twenty-first century. While more than half the models also predict significant changes in variance over ENSO regions and the North Atlantic Ocean, the direction of this change is model dependent. Seasonal mean changes are remarkably similar to annual mean changes, but there are model-dependent exceptions. Some models predict future variability that is more than double their preindustrial control variability, raising questions about the adequacy of doubling uncertainty estimates to test robustness in detection and attribution studies.
