An investigation of the reasons behind the biases in the tropical and extratropical atmospheric motion vectors derived from EUMETSAT Low Earth Orbit satellite data

Abstract

ABSTRACT Atmospheric Motion Vectors (AMVs) derived from Low-Earth-Orbit (LEO) images are essential to understand atmospheric dynamics, in particular in polar areas where other types of measurements are scarce. EUMETSAT computes AMVs from several sensors on board LEO satellites, including Metop – Advanced Very High Resolution Radiometer (AVHRR) and Sentinel-3 - Sea and Land Surface Temperature Radiometer (SLSTR). Wind observations from these sensors have a positive impact on the Numerical Weather Prediction (NWP) systems in which they are assimilated, but show strong speed biases when compared to the forecast model, especially in the tropics and the mid-latitude regions. Studies have tried to characterize these biases from a physical perspective and have hypothesized that a wrong height assignment could be responsible for these biases. In this article, we show however that the real culprit is the tracking algorithm. Indeed, our analysis of an SLSTR AMV dataset revealed that an approximation on the guess vector, combined with a lack of robustness of cross-correlation matching, is responsible for the patterns observed on our AMV bias world maps. As a side product of this finding, we elaborate on the necessity to investigate new algorithms to derive AMVs from LEO platforms.