A Computationally Efficient Data Driven Statistical Emulation for Large-Scale Remote Sensing Observing System: An Application to OCO2
Conference
64th ISI World Statistics Congress - Ottawa, Canada
Format: IPS Abstract
Keywords: dimensionreduction, fpca, gaussian-process
Session: IPS 448 - New Statistical Methods for Surrogate Modeling and Inverse Problems
Tuesday 18 July 2 p.m. - 3:40 p.m. (Canada/Eastern)
Abstract
The Gaussian process emulator is a frequently used surrogate for computationally expensive simulators to quantify uncertainty and improve process understanding. We propose a joint framework for constructing low-dimensional approximations of complex simulators by combining Gaussian process emulation technique with dimension reductions for both input and output spaces in a data-driven way: Functional principal component analysis (FPCA) procedure via a conditional expectation method which provides the best linear prediction of functional principal component scores is incorporated to reduce dimension of output space. The gradient-based kernel dimension reduction method is applied to reduce the dimension of input space when the gradients of the complex simulator is unavailable or computationally prohibitive. Theoretical properties of the resulting statistical emulator are explored, and the performance of our approach is illustrated with numerical studies.