Sensitivity of Satellite-Derived Wind Retrieval Over Cloudy Scenes to Target Selection in Tracking and Pixel Selection in Height Assignment

摘要

Satellite-derived atmospheric motion vectors (AMVs) are useful in weather analyses such as for identifying tropical lows, wind shears, and jet locations. AMVs are assimilated into numerical weather prediction models, particularly for ocean areas where wind observations are sparse. An AMV's accuracy is closely related to the processes of target tracking and height assignment (HA). The objective of this paper is to investigate the sensitivity of satellite-derived wind retrieval in cloudy scenes to the main components in these processes. AMVs are retrieved by identifying and tracking targets using advanced pattern-matching techniques based on cross-correlation statistics. In tracking targets, the main components of the AMV algorithm are the target selection methods such as the target box size, the grid size, the time interval between satellite images, and the method for determining the locations of targets. This study reveals that the optimal sizes of the target and grid could be determined differently according to the channel used for wind observation. The time interval between satellite images has a significant impact on the number of vectors with high quality and high accuracy. The HA method is also an important factor in determining the AMVs' accuracy. The heights of most vectors are assigned to cloud-top pressures using the representative radiances, and the current algorithm uses the coldest pixels to set these representative radiances. The template image used for feature tracking may contain various clouds with different movements and different heights. Therefore, without any information on feature tracking, the current approach may lead to HA errors. To mitigate these HA errors, a new approach using the individual-pixel contribution rate is tested. It tends to correct the heights of the AMVs using the water vapor channel and reduces the wind speed bias and root-mean-square vector difference.