ERA-15 Project

<Prev   TOC   Next >

7. TOVS Quality Control and Bias Tuning

The CCRs constitute a consistent data set for the entire reanalysis period. Before 1D-Var retrievals can be calculated the observed radiances have to be tuned with the forecast model. This tuning corrects systematic errors in the forecast model and possible errors of the forward model. Monthly corrections are determined for each channel by comparisons with first guess forecasts in the vicinity of a selection of quality controlled radiosonde data in different parts of the world. These corrections are then applied during the following month of assimilation.

The first step of the quality control is carried out before the data are used in the analysis. For each satellite a monthly time series of the mean six hour brightness temperatures for each channel are plotted together with the corresponding number of data. This is important for several reasons. It is good to know beforehand if and when a channel drops out and when it comes back, both for the automatic bias tuning, and also for the calculation of a set of 'physical' 1D-Var retrievals. Information obtained from NOAA/NESDIS - Polar Orbiter Archived TOVS Sounding Data change and Problem Record - contains a list of the most important changes or errors (e.g miss-locations of data) as well as changes in the software used to calculate the CCR data. Most of these events are difficult to identify from brightness temperatures alone, since their effects are relatively small.

The derivation of 1D-Var is performed using the bias corrected brightness temperatures. For quality control purposes the mean corrected and uncorrected observation minus first guess departures for each six hour period are plotted as a monthly "radgram" for each satellite and channel. Since the first guess itself is independent of any CCR change, at least when a change is about to happen, these graphs reveal satellite problems that have occurred during the previous month. Often NESDIS has listed a change (e.g. a change in the water vapour attenuation coefficients), but it is only afterwards that it can be seen whether or not this change has caused a significant problem in the data assimilation. In practice full use of this information would require the bias tuning to be done separately during all the abnormal periods, and those periods subsequently re-run with the new coefficients.

<Prev   TOC   Next >