Direct use of satellite horizontal gradients in variational analysis

The Local Analysis and Prediction System (LAPS) is a product of the Forecast Systems Laboratory. LAPS primary function is to produce an analysis suitable for local-scale model initialization, the subjective monitoring of state variables, and special product generation including precipitation type and threat indices (i.e., aircraft icing, and road weather). LAPS has its foundation in the local weather forecast office where it is traditionally the only place where high resolution Doppler radar data is available along with other local observations such as local mesonet data. LAPS uses these data along with widely disseminated data such as satellite radiances and imagery, large-scale forecast model output for background information, RAOBS, ACARS, and pilot reports to produce sub-hourly analyses on modest computer workstation technology.

Gradient matching is not a new technique to LAPS but was sidelined in the recent past due to the focus on direct assimilation of satellite radiances exploiting radiative transfer forward models (such as RTTOVS and OPTRAN) along with direct assimilation of satellite derived products and conventional data using variational methods. Variational techniques have the advantage of using all observations directly in the analysis process with reliance on error information to derive the best fit to all measurements. Recent studies using moisture data from the International H2O Project (IHOP–2002) demonstrated that at least in the moisture context, there may be significant bias problems using satellite product data since in many cases, products appear “tuned” to specific hour RAOB data (00 or 12UTC), or reflect a-priori model first-guess dependencies that might not be obvious. This can result in uncharacterized high bias errors at intervening hours (we have observed a 6-fold increase in such error during the diurnal cycle in the satellite moisture product studied during IHOP–2002). One means to solve this problem is the introduction of a horizontal gradient term in the minimized variational functional. Using horizontal gradients of the satellite product field have the advantage of ignoring bias that may be present and enabling the inclusion of satellite structure, the primary advantage of satellite data. Other data sources such as ground-based and highly dependable (and high quality) GPS and RAOB data can be used to anchor the analysis since these sources are more immune to bias problems. The combination of both gradient and direct data types in the functional render a means to achieve better accuracy while benefiting from the structure and high spatial information offered in satellite data.

This paper demonstrates the technique using first synthetic data to illustrate the mechanics of the algorithm and the means by which it was derived. Then it moves on to a genuine application of the technique using GOES moisture product data. A comparison of analyses made using conventional direct assimilation of the GOES product data vs. GOES product gradients is made.