Zhu (Judy) Yao -- MS Thesis Defense

(Penn State, Department of Meteorology)

"A Statistically based Observation Operator for Mapping Columns of Atmospheric Variables to Satellite Microwave Radiances"

When Nov 23, 2021
from 01:00 pm to 04:00 pm
Where https://psu.zoom.us/my/judy8146990656
Contact Name Zhu (Judy) Yao
Contact email
Add event to calendar vCal

Advisor: Eugene E. Clothiaux


Assimilation of satellite passive all-sky radiances, especially microwave radiances or brightness temperatures (or Tbs, converted from radiances), improves estimates of humidity and hydrometeor concentrations in columns of weather forecast models. The assimilation procedure requires forward computation of model-equivalent Tbs for each of the model columns in order to create increments based on them. A commonly used scheme for computing Tbs is the Community Radiative Transfer Model (CRTM). However, large uncertainties exist when modelling the interactions of microwave radiation with hydrometeors as well as the resulting Tbs, especially at high microwave frequencies.

One approach to account for these uncertainties is use of appropriate statistical tools. In the current work, canonical correlation analysis is used to seek basic empirical relationships between model columns of atmospheric variables and their resulting microwave Tbs from a training database. These basic empirical relationships are subsequently used to develop a basic linear observation operator that maps model columns of variables to microwave Tbs with realtime model columns as the only input. With the implementation of a Gaussian mixture model (GMM), more precise empirical relationships are found and a GMM observation operator is developed.

By testing the operators on validation cases, the GMM operator outperforms the basic one by generally better capturing the microwave Tbs at 18.70 GHz-Vpol and 183.31 ± 7 GHz-Vpol. However, the GMM operator still tends to produce extreme Tb values in heavily precipitating regions. The performance of the operator is then improved by two approaches. The GMM 1st -approach observation operator is developed by tuning the number of components and the number of PCs upon which it is based to improve performance. For the GMM 2nd -approach observation operator, we first pre-classify model columns of atmospheric variables into a heavily precipitating category and a non-heavily precipitating category by total column ice water, and we train a GMM observation operator with a certain number of components, calculating first- and second-order statistics as well as the CCVs for each category. Both approaches substantially improved the GMM observation operator, indicated by high correlation coefficients and low root-mean-square errors. Overall, the GMM 2nd -approach observation operator has the best overall performance at both frequencies while requiring the fewest computing resources.