|
Abstract:
|
There are many stakeholders interested in the behavior of future climate. E.g., insurance companies need to assess how future climate may impact the cost and risk of potential claims. Future climate behavior is typically projected using large-scale computer-based simulations, which are generated for both historical and future time periods under different scenarios. It is natural to assess the trustworthiness of these projections by determining whether the simulated historical climate matches what was observed. Comparisons of observed and modeled climate behavior typically focus on central tendencies, which overlooks other important distributional characteristics related to extreme quantiles and variability. We present a nonparametric bootstrap approach for assessing the accuracy of climate models. This approach allows us to identify potential model deficiencies over space and time for a variety of distributional characteristics, providing a more comprehensive assessment of climate model accuracy. Application will be made to state-of-the-art data from the North American Coordinated Regional Climate Downscaling Experiment (NA-CORDEX).
|
