Invited Session III: Complex Computer Models
Saturday, October 6, 8:30 a.m. – 10:30 a.m.
Session Chair: Mikyoung Jun
Exploring the magnetosphere: Parameter estimation for the LFM model
Steve Sain
The magnetosphere is the region of the Earth's magnetic field that forms a protective bubble which impedes the transfer of energy and momentum from solar wind plasma. The Lyon-Fedder-Mobarry model is used for coupled magnetosphere-ionosphere simulation and to model the effect of electron storms on the upper atmosphere. This model is generally run in a high-performance computing environment and the output represents characteristics of the electron precipitation in the upper atmosphere. In this work, we outline an approach for estimating values of parameters that control this precipitation and quantifying the uncertainty in these estimates. Of particular interest is the challenge to combine output from a computationally expensive but higher-fidelity version of the model with output from a lower-fidelity but computationally inexpensive version of the model.
Mining spatial structure in regional climate.
Douglas Nychka and Tamara Greasby
The interest in the regional effects of climate change has motivated the analysis of large spatial and space-time data that are the result of numerical models. High resolution climate system models are grounded in physics and simulate weather based on different external factors such as increasing greenhouse gas emissions, changes in land use or increases in aerosols in the atmosphere. Climate is the distribution of weather. Thus the climate state associated with these models must be estimated from the realizations of weather and so statistical analysis is necessary to interpret numerical experiments. Typically the model output involves grids of several thousand points and standard methods of spatial statistics break when applied to these large data sets. The comparison of these model experiments to observational data is also problematic because one must account for differences in support and also the irregularity of the surface records in time and space. This talk will present a flexible spatial model based on fixed rank Kriging that can handle large number of spatial locations (LatticeKrig) and also include nonstationary spatial dependence. Using this method we estimate the change in the seasonal cycle of temperature over the USfrom climate simulations from the North American Regional Climate Change and Assessment Program (NARCCAP). Part of this analysis is to take into account topography and other covariates and to determine the effect of individual regional models on the results.
A comparison of methods for estimating uncertainties in model parameters and model-based predictions
D. Higdon, E. Keating, X. Dai, J. Gattiker
A number of approaches are available estimating uncertainties in model parameters and model-based predictions. These different approaches are typically developed in different scientific fields, motivated by features of their available computational models and physical data sets. This talk briefly explains and compares null-space Monte Carlo, Bayesian model calibration, and a simple one-step ensemble Kalman filter, using a simple synthetic example subsurface transport.
Blending and downscaling ensembles of climate model predictions
Bruno Sanso
We consider a general framework for the analysis of ensembles of simulations produced by climate models. The methods are based on hierarchical Bayesian models with space and time components. Our models provide summaries of the simulated variables that allow for sensible comparisons between model simulations and historical records, and quantify possible discrepancies. Such discrepancies are used as the basis for the averaging of the simulation ensembles. We are particularly interested in obtaining blended results for time varying factors that can summarize the variability of large spatio-temporal fields. Then use process convolutions and high resolution data to downscale those results in order to obtain predictions at small spatial scales. We consider predictions of oceanic indexes in the North Pacific and predictions of sea surface temperature along the California coast, at scales that are meaningful for coastal ecosystems. Joint work with Francisco Beltran, Ricardo Lemos and Roy Mendelssohn.