Upgrading the United States Air Force Land Information System: Improvements to Global Precipitation Analysis

To meet operational requirements for the Department of Defense and other US Government agencies, the United States Air Force operates a land data assimilation system to produce near-real-time information on ground temperature, moisture, and surface characteristics. Since 2009, the Air Force has utilized the NASA Land Information System (LIS), a software framework providing state-of-the-art land surface models (LSMs) and data assimilation (DA) capabilities. NASA is in the process of upgrading LIS in several ways, with the goals of improving LSM output quality, expanding the domain to fully-global coverage (including Antarctica), and facilitating land initialization of the Air Force Global Atmospheric and Land Weather Exploitation Model.

This presentation will focus on improvements to the 3-hourly Air Force precipitation analysis, which provides a lateral boundary condition for the LSMs. The legacy operational product uses an empirical, hierarchical rules-based approach to combine rain gauge reports, bias-corrected CMORPH, IR-based and microwave-based rainfall retrievals, global NWP data, and climatology. NASA is replacing this with the Bratseth scheme, a successive correction algorithm that converges to the solution provided by Optimal Interpolation (OI). In the Bratseth scheme, a single-source background field (provided by global NWP) is adjusted to improve agreement with multiple observations (gauge, CMORPH, and satellite retrievals); error covariances for each data source are used to assign appropriate weights; and correlated errors in input satellite data are accounted for. The scheme also employs observation quality control checks and a Box-Cox transformed analysis variable (for quasi-Gaussian error distributions), leveraging work published by Environment Canada and ECMWF. Adding new data sources to the Bratseth scheme is much easier compared to the legacy algorithm, and the avoidance of explicit matrix inversion provides significant computational savings over a classic OI approach.

We will demonstrate sample results and improvements that the Bratseth scheme provides over the current operational product. We will also outline future work for assimilating NASA IMERG and/or NESDIS MIRS precipitation retrievals, and for modifying the Bratseth scheme to diagnose -- and correct for -- bias in the input data.