Improving the reliability of AMV height assignments: a study using ABI cases of the IWWG intercomparison and OCA cloud properties retrievals
A key component of Atmospheric Motion Vectors (AMV) based on cloud targets detected in successive
geostationary images is the height assigned to the tracked target area. Height assignment (HA) is
generally based on results of a cloud property retrieval algorithm and the quality and usefulness of the
resulting wind vectors for data assimilation is highly dependent on the reliability of these retrievals.
Here we examine the use of Optimal Cloud Analysis (OCA) products for AMV HA. The products were
derived from GOES-16 observations from 20th October 2019, as chosen for the latest International
Winds Working Group intercomparison study. The OCA product has not previously been
systematically used for AMV HA and it has some novel features that potentially impact how the
retrieved cloud top pressures (CTPs) should be treated in this particular application.
All OCA products (excepting cloud phase) are available with an associated expected per-pixel error
estimate and each pixel has an associated ‘solution cost’ that indicates the overall suitability of the
modelled cloud parameters to represent the real conditions. In this study we examine the impact of
applying a quality filter based on thresholds of acceptable cost and CTP error, before considering pixels
for AMV HA.
A second feature of the OCA algorithm to be considered for HA, is its capability to detect and retrieve
cloud properties in situations with overlapping cloud layers. We examined the AMV characteristics in
such cases compared to single layer clouds, and verified the likelihood that the tracked feature is
associated with the lower cloud layer when the upper is optically thin.
Standard AMV metrics, i.e. Quality Indicator, speed bias and r.m.s. against Numerical Weather
Prediction model (Global Forecast System, GFS) data, were used to assess impacts of the filtering
steps. While just three image repeat cycles from the single day were available, this is sufficient to gain
some significant insight. In addition to the overall statistics, we examined the metrics against the
imagery to correlate them when possible with specific geographic areas, cloud types and known
retrieval characteristics.
