Uncertainties in Observed Changes in Climate Extremes

Abstract

Some of the sources of uncertainty in evaluating observed changes in climate extremes include the highly non-linear, chaotic behavior of the dynamical climate system; inhomogeneities in the climate station records, and incomplete sampling of the climate system. There is no known theoretical basis for characterizing uncertainties arising from the chaotic nature of the climate system. Ensemble simulations as well as long control simulations from climate models offer an opportunity to characterize this source of uncertainty, realizing that climate models introduce their own layer of uncertainty. Prominent examples of inhomogeneities in climate station records include changes in instrumentation, station moves, changes in measurement methodology, and changes in time of observation. Examples of the potential effects of an instrument change in the U.S. cooperative observer network and a shift in the predominant time of observation are provided. Sampling uncertainties are greater for precipitation than temperature extremes, because precipitation extremes are smaller in scale. A Monte Carlo approach to quantify sampling uncertainty is described, based on station distribution in the U.S. cooperative observer network. Also, the use of statistical tools (such as Kendall's tau, generalized extreme value theory, and bootstrap resampling) and climate model ensemble experiments are discussed.