We present an approach for estimating physical parameters in nonlinear models that relies on an approximation to the mechanistic model itself for computational efficiency. The proposed methodology is validated and applied in two different modeling scenarios: (a) Simulation and (b) lower trophic level ocean ecosystem model. The approach we develop relies on the ability to predict right singular vectors (resulting from a decomposition of computer model experimental output) based on the computer model input and an experimental set of parameters. Critically, we model the right singular vectors in terms of the model parameters via a nonlinear statistical model. Specifically, we focus our attention on first-order models of these right singular vectors rather than the second-order (covariance) structure.
Abstract. A growing body of evidence indicates that anthropogenic greenhouse gases are changing Earth's climate, and that those changes may involve not only changes in climatic means but also in variability. Climate models may be informative about these future changes, but their use is complicated by the fact that they do not capture variability in current climate well. Many methods have therefore been developed to combine models and data in simulations of future climate, but current methods generally account only for changes in marginal variation and do not capture projected changes in correlation (spatial, temporal, spatiotemporal). We develop here a procedure to simulate future daily mean temperature that modifies climate observations based on changes in the mean and spectral density suggested by climate model output, and illustrate our methodology with projections from the CCSM3 (Community Climate System 3) climate model. We are able to simulate a future climate with changing temporal covariance while largely retaining non-Gaussian features of the observations. Our results suggest that in CCSM3, at most locations and most timescales, variability in daily mean temperature decreases under anthropogenic warming. The methodology presented here applies only to fully equilibrated future climate states, but may be extended to simulating transient states as well.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.