K. Sham Bhat - Scientist at Los Alamos National Laboratory
Current Position: Scientist in the Statistical Sciences Group at Los Alamos National Laboratory
I am currently working on calibration of computer models and uncertainty quantification (UQ) approaches with applications to climate models and carbon capture simulation models. We have explored the propagation of uncertainty in a multi-scale model using a “model-plus-discrepancy” approach. That is, we are not only propagating uncertainty upward due to small parameterizations but also incorporating the discrepancy between the model and reality. We are also beginning work on “intrusive” UQ; that is, incorporate scientific dynamics in our statistical models rather than assuming that the computer model is a “black box” with inputs and outputs.
Research Summary at Penn State:
At Penn State, I worked on research on calibration of large ocean models with large multivariate spatial output. We developed a computationally tractable approach to develop an emulator (surrogate model) for the computer model using a Gaussian processes. Using this emulator, which established the connection between the computer model and the calibration parameters of interest, we then inferred these parameters using a Bayesian approach while accounting for sources of uncertainty and spatial dependence. Computational gains were achieved by using kernel mixing and low-rank matrix identities.
I also worked on a project with the US Geological Survey to develop an approach to combine many Global Circulation models (with large spatial output) using Bayesian Model Averaging (BMA) in order to make future projections.
Why I chose Penn State:
Excellent program/reputation of statistics department and opportunity to do interdisciplinary research.