Uncertainty Quantification for Characterizing Spatial Tail Dependence under Statistical Framework
Abstract:Large-scale weather systems such as Atmospheric Rivers (ARs) may affect extreme climate events, in particular resulting in high spatial coherence across regions. We use methods from statistical extreme value theory to characterize the spatial dependence of extremes as a function of spatial distance. Our focus in this work is characterizing uncertainty in our understanding of how the spatial dependence of extremes in climate models
We investigate the influence of ARs on the spatial dependence structure of extreme precipitation from CMIP5 simulations under climate change. We fit statistical models that treat initial condition ensemble members as independent data replicates and uses bootstrapping (across yearlong blocks of data) to estimate uncertainty from having only limited model runs. We also focus on multi-model ensembles as random draws of CMIP5 model runs and approximate the uncertainty for simulating the behavior of tail dependence across models.
Preliminary results from four CMIP5 models show that projected AR events bring more severe rainfall with less dependent pattern between locations under high emissions scenario (RCP8.5) during 2076-2100 than for the historical run during 1981-2005. Within the UQ framework, spatial dependence between nearby locations is estimated more precisely, showing narrower confidence intervals, than the spatial dependence measure for locations further apart.