Monday, December 3 2018
11:00am - 12:00pm
Computational Mathematics Colloquium
CCM Colloquium: Ensemble Chemical Weather Forecasting/Data Assimilation with WRF-Chem/DART and Compact Phase Space Retrievals

Accurate chemical weather forecasts depend on chemical data assimilation. However, chemical data assimilation faces several challenges: (i) atmospheric composition profile observations are satellite-based retrieval profiles that have large data volume and low information density; (ii) retrieval profiles contain redundant information and include contributions from the retrieval prior; and (iii) the retrieval error covariance contains error cross-correlations.

WRF-Chem/DART is a regional ensemble chemical weather forecast/data assimilation system. In WRF-Chem/DART, we introduced “compact phase space retrievals” (CPSRs) to address the retrieval assimilation challenges. CPSRs are derived by applying compression and rotation transforms to the quasi-optimal form of the retrieval equation. The quasi-optimal form removes the retrieval prior’s contribution. The compression transform removes the redundant information and reduces the number of observations from the dimension of the retrieval vector to the rank of the averaging kernel. And the rotation transform diagonalizes the compressed retrieval error covariance to maximize the compressed retrieval error variance. We have applied CPSRs in several studies and found computational cost reductions ranging between 40% and 60% and forecast skill increases ranging between 5% and 10%.

In this seminar, I will introduce WRF-Chem/DART and the ensemble chemical weather forecast/data assimilation problem. I will review the mathematics underlying: (i) CPSRs; (ii) their extension to enable assimilation of truncated retrieval profiles as CPSRs, and (iii) a modification of the extension in (ii) to enable assimilation of retrieval profiles that extend above the model top as CPSRs. I will present results from several CPSR assimilation case studies and
discuss the outstanding research topics related to CPSRs.

This will not be a typical mathematics department seminar. I will not prove any theorems or derive any computational methods. The mathematics underlying CPSRs is relatively straightforward and comes from a first or second semester course on linear algebra. Despite that simplicity, CPSRs will likely revolutionize the way we assimilate and store retrieval profiles as we move toward the use and storage of the very high temporal/spatial resolution retrieval profiles from GEO-CAPE.

arthur.mizzi@state.co.us
Speaker:Dr. Arthur P. Mizzi
Affiliation:Colorado Department of Public Health and Environment, Air Pollution Control Division
Location:ACAD 4119


Download as iCalendar

Done