Improved Statistical Models and Methods for Atmospheric Science Measurements

点击翻译按钮获取中文摘要

In atmospheric science, many remote sensing instruments make indirect measurements by observing readily-accessible phenomena and using mathematical models to infer the quantities of interest. A common example is the use of satellites to measure the spectrum of reflected sunlight and subsequent use of a procedure called optimal estimation to invert a physical model that relates radiances and atmospheric properties. This project focuses on developing statistical methods that more accurately capture the uncertainty in this inference. This research has the potential to greatly improve accuracy of current and future remote sensing efforts, particularly those that utilize spectrometers. Inferring the atmospheric state from spectra is called a retrieval. In statistical parlance, this is an estimation of parameters in a statistical model where a physical forward model defines the mean structure. Since the parameters have a particular physical meaning, it is essential that model error is properly accounted for. This project aims to extend the current framework through statistical advances that include efficient low-rank representation of model-discrepancy and targeted use of independent ground measurements for prior distributions. Developments will be performed within a test-bed created by an uncertainty quantification group. There a surrogate forward model has been developed that includes the essential physics but is simpler and computationally faster than the forward models currently used. The results of the project are anticipated to increase the accuracy of atmospheric measurements and enable advances in several areas of science.