1. Introduction

Atmospheric motion vectors (AMVs) are one of the most important meteorological products extracted from satellite imagery because they are assimilated every day in numerical weather prediction (NWP) models. They have been derived operationally from geostationary satellites [GOES, Meteorological Satellite (Meteosat), Multifunctional Transport Satellite (MTSAT)] for more than two decades by tracking clouds or water vapor tracers through successive images. Polar wind extraction was initially tested with the Advanced Very High Resolution Radiometer (AVHRR) on NOAA satellites (Turner and Warren 1989) and then with the Moderate Resolution Imaging Spectroradiometer (MODIS) on Terra and Aqua satellites (Key et al. 2003). However, despite the operational production of AMVs over polar regions, a gap of AMVs observations still exists in the 55°–70° mid- to high-latitude bands north and south between the coverage areas of geostationary and polar AMVs.

The European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) has been deriving AMVs operationally over polar regions from AVHRR/3 on board MetOp-A satellites (Klaes et al. 2007) since 2011. Unlike the standard polar wind extraction method, which uses image triplets for tracking clouds (Key et al. 2003), the MetOp-A polar winds extracted at EUMETSAT rely on image pairs taken by the same satellite (Dew 2010). This strategy has resulted in the loss of the temporal consistency test between the two consecutive intermediate vectors usually obtained from image triplets. But it has three major advantages: first, it decreases the tracking time from two orbit periods to one; second, it increases the wind retrievals poleward of 50° latitude north and south; and last, it decreases product latency. The increased coverage area of the EUMETSAT MetOp polar winds helps to fill the severe lack of data in the 55°–70° latitude bands at north and south, but it is not enough to fill it completely.

In 2012, EUMETSAT launched the MetOp-B satellite that took over primary operations from MetOp-A in April 2013. The tandem configuration with the two MetOp satellites in the same orbital plane provides an interesting opportunity to create global AMVs from satellites with a significant overlap in imagery data. The minimum overlapping area seen consecutively by the two MetOp satellites is half of the AVHRR swath width (~1500 km) at low latitudes. This is an area wide enough to extract wind information from two consecutive AVHRR images. The AVHRR AMV extraction algorithm initially developed at EUMETSAT for polar wind extraction from a single satellite has therefore been modified to ingest the AVHRR data from the two MetOp satellites. The latest version of the AVHRR AMV extraction algorithm developed in 2015 allows for the extraction of three different AVHRR AMV products (EUMETSAT 2015) that use a pair or a triplet of AVHRR images, taken from one or two MetOp satellites, with different coverage and a different temporal gap between the successive images used to extract the motions. O. Hautecoeur and R. Borde (2016, unpublished manuscript) make a detailed comparison of these products over polar regions, discussing their respective performance, the importance of the temporal gap between the images used for the tracking, and the impact of the temporal consistency check in the calculation of the quality index (QI).

This paper presents the new EUMETSAT global AVHRR wind product that became operational on 30 January 2015. Global coverage is ensured by two complementary products: one considering MetOp-A as the first image and MetOp-B as the second image of the pair (noted MetOp-A/MetOp-B in the following), and another one considering MetOp-B as the first image and MetOp-A as the second image of the pair (noted MetOp-B/MetOp-A in the following). The temporal gap between the two consecutive images is about 50 min. The first part of this paper illustrates the potential advantages of the product, such as the global coverage and the better characterization of the polar jets located in the mid–high-latitude bands. The second part presents comparisons of the global AVHRR wind product against collocated Meteosat First Generation (MFG) and Meteosat Second Generations (MSG) AMVs. Such a test is a good quality check because MFG and MSG AMV products have been used and evaluated in NWP for a long time and are known to be reliable products. The last part discusses the potential benefits of global AVHRR wind product for assimilation into NWP models and prospective for future developments.

2. Global AVHRR wind product performances

Upper plots of Fig. 1 show wind speeds extracted on 5 January 2015 by the global AVHRR wind extraction algorithm for AMVs that have a QI > 60 on the range scale from 0 to 100. The results are presented for the Arctic (top left) and the Antarctic (top right), and midlatitude and tropics coverage (middle). The quantities are averaged on an equal-area sphere pixelization with a resolution of about 150 km. These plots illustrate the global coverage of the global AVHRR wind product, allowing for a homogeneous retrieval of the winds over the whole globe. Global AVHRR wind production is better resolved over polar areas because the MetOp-A or MetOp-B satellites fly over these areas every 50 min, and only twice per day over the same location in tropical areas. The global AVHRR wind product helps fill the gaps of observations between 55° and 70° latitude north and south. As illustrated in Fig. 1, the fast winds of the polar jets are frequently located in these latitude bands. The bottom plot of Fig. 1 shows the zonal distribution of speeds vs altitude for wind speeds extracted on 25 January 2015. The fast winds of the polar jets (>60 m s⁻¹) are well described at around 60° latitude north and south, at an altitude close to 300 hPa. This retrieval matches very well with the average location and altitude of the synoptic-scale phenomena in January.

View Full Size

Wind speed of the global AVHRR wind product extracted on 5 Jan 2015 over the globe: (top left) Arctic and (top right) Antarctic, and (middle) global coverage. (bottom) Zonal distribution of the speeds vs altitude (25 Jan 2015). Size of the dots represents the number of AMVs extracted.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0155.1

• Download Figure
• Download figure as PowerPoint slide

Figure 2 shows the time series of global AVHRR wind product statistics over a 2-month period, from May to July 2015. Plots show the daily average of the total amount of extracted winds (top), speed biases (middle), and speed root-mean-square (RMS) (m s⁻¹) against collocated forecast fields from the ECMWF Integrated Forecast System (IFS) model. Results are split into five latitudinal bands: the North Pole (NP) between 60° and 90°N, the northern midlatitudes (NH) between the Tropic of Cancer (23.47°N) and 60°N, the intertropical region (TR), the Southern midlatitudes (SH) between the Tropic of Capricorn (23.47°S) and 60°S, and the South Pole (SP) between 90° and 60°S. Only AMVs having a QI > 60 have been considered in the bias and RMS statistics. AMV production appears very stable over the whole period for all the studied areas. It can be noted that the number of extracted AMVs for SP is less than that for NP for this period. Beyond the geographical characteristics of the two areas, this difference is also explained by the poor performance of the cloud mask scheme actually implemented on the operational AVHRR chain (EUMETSAT 2013) and used in global AVHRR wind extraction. Indeed, the lack of information from the visible channel during nighttime and the simple thresholding methods used for the pixel identification do not allow for separating clearly clouds from ice or snow in the polar regions. So, the amount of cloudy pixels drops dramatically during nighttime. Between May and July, fewer AVHRR pixels are classified as cloud than during daytime conditions over the South Pole as it is the polar night. Consequently, the number of AMVs produced is considerably reduced.

View Full Size

Time series monitoring from 8 May to 8 Jul 2015 of global AVHRR statistics: (top) total amount of AMVs, (middle) speed biases against forecast (m s⁻¹), and (bottom) RMS speed difference (m s⁻¹). Gray and black solid lines represent statistics for the NP (lat > 60) and SP (lat < −60), gray and black dashed lines for the NH (23.47 < lat < 60) and SH (−60 < lat <−23.47), and the dotted line for the TR (−23.47 < lat < 23.47).

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0155.1

• Download Figure
• Download figure as PowerPoint slide

Averaged speed biases over the studied period are found around 0.35 m s⁻¹ for NP, −0.22 m s⁻¹ for NH, 1.69 m s⁻¹ for TR, −0.52 m s⁻¹ for SH, and −0.28 m s⁻¹ for SP. The corresponding averaged speed RMS differences are equal to 3.32 m s⁻¹ for NP, 4.06 m s⁻¹ for NH, 4.02 m s⁻¹ for TR, 4.00 m s⁻¹ for SH, and 4.23 m s⁻¹ for SP. These statistics are quite stable over the whole period. Speed biases are small, between −0.55 and 0.35 m s⁻¹, except in TR, where a large positive bias is observed. The geographical coverage plot in Fig. 3 shows that large positive biases match well to the intertropical convergence zone (ITCZ). It is presently not clear how to explain why such large positive speed biases occur especially in this area. However, internal studies are currently ongoing to identify the role played by semitransparent clouds, which are very frequent in this area. Indeed, the lack of water vapor and CO₂ channels on the AVHRR instrument does not allow for properly correcting the cloud-top height (CTH) retrieval for semitransparent clouds in the EUMETSAT AVHRR AMV algorithm. Being calculated only from AVHRR brightness temperatures that appear artificially too warm due to infrared radiation from a lower level below the cloud, the AMV altitude appears set too low in the troposphere, leading to fast positive speed bias.

View Full Size

As in Fig. 1, but for the corresponding speed bias against the forecast field.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0155.1

• Download Figure
• Download figure as PowerPoint slide

A closer comparison of Figs. 1 and 3 shows that speed biases are quite small in the polar jet areas too, meaning that the fast speeds detected by the global AVHRR wind algorithm are in fair agreement with the corresponding forecast fields. This is shown both on the coverage pictures and on the zonal plots of Fig. 3. A potential benefit to NWP may come from such better characterization of the polar jets’ fast winds.

3. Comparison against collocated geostationary AMVs

The global coverage of the global AVHRR wind product allows for a direct comparison with geostationary AMVs that are usually assimilated in NWP models. EUMETSAT produces AMVs operationally from several geostationary satellites. Meteosat-10 is the current EUMETSAT primary operational geostationary satellite operating at 0°; it extracts AMVs hourly considering four consecutive images taken every 15 min (Borde et al. 2014). Meteosat-7 is the last satellite of the MFG series; it takes an image every 30 min over the Indian Ocean (57°E longitude) and extracts AMVs every hour considering three consecutive images. Both Meteosat-7 and Meteosat-10 extract AMVs from the infrared channel at 10.8 μm, allowing for a direct comparison with global AVHRR wind extracted at the same location. Collocations between MFG or MSG and the global AVHRR wind [EUMETSAT Polar System (EPS)] have been considered only for AMVs having a QI > 80, a horizontal distance < 0.25° between the two collocated AMVs, and a temporal gap shorter than 45 min. It must be noted that the EPS subset could have slightly different statistics to the global AVHRR wind product statistics presented in Fig. 2 because of the differences in collocation criteria and geographical areas used.

Figure 4 shows scatterplots of EPS wind speeds, directions, and altitudes, against collocated Meteosat-10 AMVs (left column) and Meteosat-7 AMVs (right column). Results show a very good agreement between EPS winds and the collocated geostationary AMVs for speeds and directions, the Pearson correlation coefficients being between 0.90 and 0.96. The two instruments mostly detect the same motion at the same location and at the same time despite the different spatial, temporal, and processing characteristics used to extract the AMVs (see Table 1). The agreement is slightly better for the directions than for the speeds, and it can be noted that EPS wind speeds are generally a bit faster than MSG and MFG wind speeds extracted at the same location. According to García-Pereda and Borde (2014) and Bresky et al. (2012), the difference in target box sizes used in the extraction process, close to 80 × 80 km² for MSG and 30 × 30 km² for AVHRR, and the difference in spatial pixel resolution, around 3 km at nadir for MSG versus 1 km for AVHRR, may explain the slightly faster winds extracted from AVHRR instrument. Indeed, winds derived using a geometrical small target window more accurately reflect the motion on a local scale, whereas the winds derived using a large target window reflect the mean synoptic-scale motion. So, a large target window is more likely to contain motions on varying spatial and temporal scales and possibly at different levels.

View Full Size

Global AVHRR (top) wind speeds, (middle) directions, and (bottom) altitudes vs collocated (left) Meteosat-10 AMVs and (right) Meteosat-7 AMVs for the months of January–March 2015. Only AMVs that have a QI > 80 have been considered.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0155.1

• Download Figure
• Download figure as PowerPoint slide

Table 1.

Differences between EPS, MSG, and MFG AMV extraction scheme characteristics.

View Table

Altitude scatterplots show more differences, especially at high levels, where the EPS winds are found at lower altitude than corresponding MSG and MFG ones. Statistics (not shown) indicate that the average EPS wind pressure is found to be around 35 hPa larger than MSG AMVs and 50 hPa larger than MFG AMVs. As already mentioned, the lack of semitransparent cloud correction method to set the CTH with AVHRR can explain why AVHRR winds are found at lower altitude in the troposphere. This correction is done for both MSG and MFG AMVs.

The size of the target box used to extract the AMVs roughly defines the scale of the structures tracked by the algorithm (García-Pereda and Borde 2014). In addition, Borde et al. (2014) have noted that the coldest pixels present in the target box, which correspond to the highest clouds, contribute significantly to the correlation process and must be used to set the altitudes of the winds. The likelihood of high-level cloud in the target box is greater when a larger target box size is used. Therefore, we would expect a greater proportion of AMVs at a high level with MSG or MFG compared to EPS.

Differences between the AMVs altitudes also occur at low levels, where the methods used to set the altitudes are slightly different for EPS, MSG, and MFG. The criteria for applying the inversion method, which sets the AMV altitude to the level of the temperature inversion, are slightly different for the MFG, MSG, and EPS AMV extraction schemes. It explains the vertical stripes present in the lower-left panel of Fig. 4, for which the inversion method has been applied to MSG and not to EPS AMVs.

Tables 2 and 3 present the speed biases and speed RMS statistics of EPS winds and collocated MSG and MFG AMVs, respectively, against forecast fields from January to March 2015. Only AMVs that have QI > 80 have been considered. Results are presented by altitude levels—high level (HL), above 400 hPa; midlevel (ML) between 400 hPa and 700 hPa; and low level (LL), below 700 hPa—for the same geographical regions as used in section 2. The strict criteria with a large QI threshold used for collocation did not allow getting large dataset with MFG. EPS winds tested during this period show performances comparable to the Meteosat-10 and Meteosat-7 AMVs that are already assimilated into NWP models. EPS winds get even smaller speed biases and RMS at the HL and ML in NH and SH, but they present slightly worse statistics at the LL. The fast HL positive bias in TR appears specific to the EPS wind product and needs more investigations. However, existence of hourly geostationary AMVs reduces the requirement for assimilation of the global AVHRR wind product in TR.

Table 2.

Comparison of collocated MSG and EPS AMVs against forecast fields for the 3-month period from January to March 2015. Collocations criteria: QI > 80, a horizontal distance smaller than 0.25°, a temporal gap shorter than 45 min, and a pressure difference smaller than 25 hPa.

View Table

Table 3.

As in Table 2, but for comparison of collocated MFG and EPS AMVs.

View Table

4. Conclusions and discussion

This paper presents the new global AVHRR wind product developed at EUMETSAT. Two complementary products—MetOp-A/MetOp-B and MetOp-B/MetOp-A—are extracted from the tandem configuration of MetOp satellites, ensuring a global operational extraction of AMVs over the whole globe. Speed bias and speed RMS difference performances of the product are good except in the tropics. The performances are even a bit better than those of collocated Meteosat-7 and Meteosat-10 AMVs in NH and SH at high and midlevels. However, because of the poor twice-per-day occurrence of AMV extraction over the same areas in the tropics and midlatitudes, it is not expected that this product can compete with the hourly geostationary winds that have been used for assimilation in NWP. The most important expected impact of the global AVHRR wind on the forecast score might come from a better characterization of the polar jets in the 55°–70° latitude bands in NH and SH, where there is still a lack of good wind observations. The preliminary results, which show a good characterization of the polar jets, are very encouraging, and they should be hopefully confirmed by the future impact studies done in NWP centers.

Beyond the potential use in NWP models, the global coverage of a homogenous AMV dataset provides also possibilities to investigate deeply several scientific questions that are still open in the AMV extraction work. For example, the global AVHRR wind product can be compared to AMVs extracted from all geostationary satellites in operation—Meteosat, GOES, MTSAT, Fengyun (FY), and the Indian National Satellite–3D (INSAT-3D)—in order to better understand the differences induced by the use of different AMV extraction schemes by the AMV producers. The global AVHRR wind product has very bad statistics in the tropics, especially in the ITCZ area. Future investigation of this large positive bias found in the tropics should help to understand the long-standing problems of AMV extraction in this area.