Estimation of AMV Height Assignment Errors Using Model Best-Fit Pressure and AEOLUS Height Data
The situation dependent observation error algorithm for AMVs, developed by Forsythe and Saunders (2008, IWW10), has been implemented in the ECCC data assimilation systems in 2018. In this algorithm, the observation error is the sum of two terms: one for the tracking error, the other for the height assignment error. The latter represents the largest part of the observation error. The height assignment error is estimated from differences between assigned AMV height and model best-fit pressure where the wind vector difference between AMV (O) and model background wind (B) is minimum. Standard deviations of O-B from winter and summer periods are used for estimating height error profiles for each AMV type (i.e. VIS, IR and WV) and for each satellite. This approach works reasonably well. However, it would be preferable to use independent sources of observations that provide both wind and cloud information, such as those from Aeolus, instead of B.
The Aeolus mission provides global measurement of horizontal light-of-sight (HLOS) wind profiles. There are two categories of winds usable for data assimilation: Rayleigh-clear and Mie-cloudy. The Mie-cloudy winds are particularly suitable for comparison with AMVs since these data are located near cloud tops and underneath, depending on the optical depth. The height and HLOS wind of collocated observations with AMVs can be used for estimating both the systematic and random height errors of IR and VIS AMVs from all GEO and LEO satellites.
With the implementation of new and improved AMV products, as well as the replacement of satellites, the height error profiles used in the situation dependent observation error algorithm need to be revised periodically. This revision was recently made at ECCC by using B and Aeolus data for the following satellites: Himawari-8, GOES-16/17, METEOSAT-8/11, SNPP/NOAA-20 and METOP-A/B. The reprocessed Aeolus observations (baseline 11) from July 2019 to March 2020 were used, giving a large number of collocations with AMVs. Both systematic and random height errors were examined. Various collocating distance and time difference criteria were tested. We found that the results are sensitive to these criteria and the distribution of the AMVs and Aeolus data in height. In addition, the spatial representativeness of these two types of observations is quite different, which makes the observation matching challenging. Nevertheless, the AMV height assignment errors estimated from B and Aeolus data agree reasonably well, after a careful selection of observation pairs.