In response to CGMS Recommendations 39.26

This paper reports on the recent status of JMA AMVs from MTSAT-2and MTSAT-1R.
In accordance with WMO CBS-XIII Recommendation 4 "Amendments to the Manual on Codes (WMO-No. 306) Volume I.2," JMA plans to switch the MTSAT-AMV BUFR format from FM-94 BUFR edition 3 to edition 4 by 6 November, 2012.
Rapid-scan images taken at five-minute intervals by MTSAT-1R are utilized to derive rapid-scan AMVs. JMA is currently considering the use of these AMVs in typhoon analysis as an ancillary dataset, and is also conducting observing system experiments for such AMVs using its meso-scale NWP system.
JMA/MSC is developing AMV software for the current MTSAT-1R and MTSAT-2 operational satellites and their follow-on units, Himawari-8 and Himawari-9. Recent development activity has included work on improving cloud tracing based on small target boxes with sizes such as 5 x 5 pixels. The results of an AMV derivation experiment based on maximum likelihood estimation have shown that wind speed values are pushed up and IR negative wind speed bias values are mitigated using this method.
In response to CGMS recommendation 39.26, JMA’s resent activity and plan to introduce the stand-alone portable AMV software developed by the NWCSAF is presented.


This paper reports on Atmospheric Motion Vectors (AMVs) produced by the Meteorological Satellite Center (MSC) of the Japan Meteorological Agency (JMA) using images from MTSAT-1R and MTSAT-2. The status of the AMV production and dissemination plan is covered in Section 2, AMV quality is detailed in Section 3, Rapid-scan AMVs are dealt with in Section 4, and JMA’s ongoing development for MTSAT AMVs and the Himawari-8 and Himawari-9 follow-on satellites are outlined in Section 5. In final section 6, examination to the stand-alone AMV software package from the NWCSAF and plan to use MTSAT imagery are introduced.


Table 1 lists the details of current MTSAT-2 AMV dissemination. JMA generates four types of AMVs from MTSAT-2 infrared (IR: 10.8 micrometers), water vapor (WV: 6.8 micrometers), visible (VIS: 0.63 micrometers) and short-wave infrared (IR4: 3.8 micrometers) images (referred to below as IR AMVs, WV AMVs, VIS AMVs and IR4 AMVs, respectively). IR, VIS and WV AMVs are disseminated via the Global Telecommunication System (GTS) in Binary Universal Form for data Representation (BUFR) format. Short-wave infrared AMVs are computed exclusively for JMA’s NWP system.

JMA plans to switch the MTSAT-AMV BUFR format from BUFR edition 3 to edition 4 by 6 November, 2012.

Table 1 MTSAT-2 atmospheric motion vector products generated by JMA


IR high-level and low-level AMVs

This section reports on the monthly quality of six-hourly IR and WV AMVs produced from July 2005 to June 2012 based on standard CGMS AMV statistics. AMVs are compared with sonde observations for quality evaluation.

Figures 1 and 2 show time-series representations of monthly statistics (number, root mean square vector difference (RMSVD) and wind speed bias (BIAS)) for high-level (above 400 hPa) and low-level (below 700 hPa) IR AMVs. In these statistics, AMVs with quality indicator (QI; Holmlund 1998) values above 0.85 are applied. To match sonde observation times, AMVs from 00 and 12 UTC are used.

Figure 1 shows that high-level IR AMVs settled to show relatively good quality after the algorithm upgrades of 2007 (Imai and Oyama 2008) and 2009 (Oyama 2010), although slightly slow BIAS values still remain in the winter hemisphere.

Figure 2 shows that low-level IR AMVs have relatively small RMSVDs and BIAS values (around 4 – 5 m/s and 0 – 1 m/s, respectively). One reason for this is that low-level winds move more slowly than high-level winds (not shown).

There were two operational switchovers from MTSAT-2 to MTSAT-1R in August 2011 and November 2011. Figures 1 and 2 indicate no significant change in the statistical quality and characteristics of IR AMVs in relation to the switchovers.

WV AMVs over cloudy areas

Figure 3 is similar to Figures 1 and 2, but shows only extracted high-level WV AMVs over a cloudy region. With respect to AMV quality change, the magnitude of the higher-wind speed BIAS for WV high-level AMVs is slightly larger than that seen before the upgrades of 2009. However, these upgrades reduced periodic fluctuations of BIAS values, especially in the winter season over the Northern Hemisphere. MSC/JMA is currently working on the development of an algorithm to mitigate high-speed BIAS.

Figure 1: Long-term time series of numbers (upper panel), RMSVDs and BIAS values (lower panel) of high-level IR AMVs (QI > 0.85) over the Northern Hemisphere (between 20°N and 60°N; blue lines), the tropics (between 20°S and 20°N; red lines) and the Southern Hemisphere (between 60°S and 20°S; green lines). RMSVDs (solid lines) and BIAS values (dashed lines) are shown.

Figure 2: As per Figure 1, but for low-level IR AMVs

Figure 3: As per Figure 1, but for cloudy-region WV AMVs

4 Rapid-scan Operation and Rapid-scan AMVs

4.1 Rapid-scan AMVs in the vicinity of typhoons

In 2011, JMA began to conduct Rapid Scan Operation (RSO) during the daytime (00 UTC to 09 UTC) from June to September. AMVs derived from short-interval images are considered useful for capturing short-lived phenomena such as rapidly developing cumulus clouds or rapidly deforming typhoon cloud systems.

Figure 4: Rapid scan wind data from MTSAT-1R visible imagery superimposed onto ASCAT wind data in the vicinity of Typhoon Ma-on (01 UTC, 23 July, 2011). ASCAT wind data are utilized as an ancillary resource to support estimation of the force wind area of typhoons, but cover a relatively narrow area and have poor temporal resolution (data are retrieved twice a day). RS-AMV has relatively coarse spatial resolution but good temporal resolution (5 min) compared to ASCAT winds. Such short-interval images can capture short-lived cloud systems that cannot be tracked by operational MTSAT-2 images taken at 15- or 30-min intervals.

Currently, ASCAT surface wind data are utilized as a main ancillary resource in JMA typhoon analysis, especially for estimation of force wind areas over data-sparse sea regions. However, as ASCAT coverage is temporally sparse (ASCAT observes the same points twice a day), data-sparse regions must be compensated for by other observations. Rapid-scan AMVs (RS-AMVs) are obtained every five minutes for the area around Japan, and the ability of rapid-scan images to capture short-lived typhoon cloud systems supports their use for such compensation in data-sparse areas around typhoons. Accordingly, MSC/JMA conducted a study to compare ASCAT wind and AMV data from rapid-scan visible imagery covering the vicinity of a typhoon to validate the use of RS-AMVs in typhoon analysis.

Figure 5 shows the results of comparison between RS-AMVs from visible imagery and ASCAT wind data from around the vicinity of Typhoon Ma-on. It indicates a close correlation between sea-surface winds and AMVs, as suggested by previous work involving the use of sea-surface wind data from in-situ observations (e.g., Ohshima et al. 1991; Dunion and Velden 2002). Wind speeds and directions show close correspondence between RS-AMVs and ASCAT data. However, the high-speed AMV BIAS values seen in the high-speed regime (highlighted by the blue circle in Figure 5) require further investigation.

Figure 5: Scatter plots of MTSAT-1R rapid scan wind (visible) and ASCAT wind data

Wind speed differences (left) show close correspondence, but the mean speed of ASCAT winds is about 0.8 times that of AMVs up to 15 m/s. Wind direction differences (right) are small compared to speed differences, but ASCAT winds are directed about 10 degrees inward to the typhoon center with AMV data as a reference. The statistical period is from 17 to 23 July, 2011 (roughly corresponding to the period during which Typhoon Ma-on was in the rapid-scan area).

4.2 MTSAT Rapid-scan AMVs and their impact on NWP

The results of a JMA observing system experiment (OSE) on how MTSAT RS-AMVs impact NWP are provided on the IWWG website at

The outcomes of the experiment indicate that RS-AMVs show a small observation error correlation against first-guess data and have a smaller standard deviation against first-guess figures above 400 hPa, while the values is larger below 400 hPa. An OSE was also performed on MTSAT RS-AMVs using the meso-scale operational NWP system for 2010, with results indicating smaller wind speed forecast errors at all levels against sonde observations. In this experiment, precipitation predictions in 15-hour forecasts were also improved (Yamashita 2012).


JMA/MSC is currently developing a new tracking method based on probabilistic estimation. The algorithm uses two target boxes and one search box for each wind vector. The target boxes are large and small, respectively.

In the new algorithm, wind vectors are derived as usual. The only difference is the use of a new correlation surface, which is the mean of two such surfaces derived from the large and small target boxes. Averaging operation for the correlation surfaces is based on the idea that the correlation value on the surface should be regarded as the likelihood to match target and search features. Feature tracking using the large target box is considered to support observation of atmospheric motion for large-scale phenomena, and correlation values computed in the tracking process are only minimally contaminated by noise because the large target box contains many pixels. Conversely, with the small target box, wind vectors are expected to capture atmospheric motion in detail. However, many noisy wind vectors are generated because target feature mismatch results from the lower number of pixels.

To capture atmospheric motion in detail, it is essential to reduce noise of correlation values. Averaging operation to correlation surfaces makes it possible to distinguish an actual peak from false ones generated by noise.

Figure 6 shows a conceptual diagram of this idea. Actual peaks on the matching surface computed from a small target box are pushed to the upside by summation with the surface produced with a larger target box. It should be noted that the positions of peaks on the correlation surface are not largely affected by this operation because those positions are determined from first and second derivative functions of the correlation surface. Smaller correlation surfaces generally show higher wind speeds than larger ones.

Table 2 shows the results of sonde statistical analysis for two IR AMV datasets derived from a single correlation surface using a large target box (16 x 16 pixels: CNTL) and an averaged correlation surface derived using large and small target boxes (16 x 16 and 5 x 5 pixels: TEST). As can be seen, mean wind speeds are accelerated through the use of a new algorithm. As a result, the IR negative wind speed bias against sonde values is improved. These results are consistent with those of recent research showing that AMV derivation using a small target box increases wind speed values (Sohn and Borde 2008; Daniels and Bresky 2010).

Figure 6: Conceptual diagram showing the results of the new algorithm. The two images on the left show correlation surfaces generated from large and small target boxes. The one on the right is generated from probabilistic averages for the two surfaces. The number of false peaks on the surface from the small target box is mitigated by this operation, as shown in the figure on the right.

Table 2: Sonde statistical verification of IR AMVs derived using the regular operational method (left: 16 x 16 target box) and the new algorithm (right: 16 x 16 and 5 x 5 target boxes) for QI > 0.85. The statistical method is based on CGMS statistical reporting. Blue highlighting represents improvement, and red represents deterioration. A significant difference is seen in the increased wind speed data.

The parameters used here are as follows:

MVD: mean vector difference

RMSVD: RMS vector difference

BIAS: speed bias

SPD: wind speed

NCMV: no. of computed wind data

NC: no. of collocations

JMA is also developing a new AMV algorithm for the Himawari-8 and Himawari-9 follow-on satellites that can handle newly added IR/WV channels for height assignment. Implementation of the H2O intercept method and other approaches involving the use of three or more channels are planned. For the AMV software, cloud mask, type and height products will be used to provide input data for cloud detection and height assignment. The implementation of the new tracking algorithm described above and multi-channel tracking are also planned.

6JMA’s current activity and future plan to the stand-alone AMV software package for development to Himwari-8/9

Recommendation 39.26: Satellite AMV providers are invited to examine the stand-alone AMV software package from the NWCSAF and to report back to CGMS 40.

In response to above CGMS recommendation 39.26,JMA/MSC is currently utilizing portable stand-alone software developed by the NowcastingSAF (NWCSAF) for the Japanese follow-on satellite Himawari-8.Current activity of JMA/MSC is to remodel the software for the purpose to handle both of JMA NWP data and MTSAT imagery.

As for the handling JMA NWP data, the agency generated cloud product and AMV using the stand-alone software with MSG imagery and the JMA NWP data.

Study to compare between two results those are generated from MSG imagery and with ECMWF and JMA NWP data was introduced in poster session of the EUMETSAT user's conference 2012 in Sopot.

From result of the study, it is considered that difference of NWP vertical profiles has no small effect on height assignment especially to semi-transparent clouds at upper level. In the next stage of this study, JMA is developing module to handle MTSAT imagery for detailed analysis to the height difference.


Daniels, J. and W. Bresky, 2010: A New Nested Tracking Approach for Reducing the Slow Speed Bias Associated with Atmospheric Motion Vectors (AMVs), Proc. of the Tenth International Winds Workshop, Tokyo, Japan.

Dunion, J. P. and C. S. Velden, 2002: Application of Surface-Adjusted GOES Low-Level Cloud-Drift Winds in the Environment of Atlantic Tropical Cyclones. Part I: Methodology and Validation, Mon. Wea. Rev., 130, 1333 – 1346.

Holmlund, K., 1998: The Utilization of Statistical Properties of Satellite-derived Atmospheric Motion Vectors to Derive Quality Indicators. Wea. Forecasting, 13, 1093 – 1104.

Imai, T. and R. Oyama, 2008: Developments for Quality Improvement of an Atmospheric Motion Vector Product, Meteorological Satellite Center Technical Notes, 51, 41 – 55 (in Japanese).

Ohshima, T., H. Uchida, T. Hamada and S. Osano, 1991: A Comparison of GMS Cloud Motion Winds with Ship-Observed Winds in Typhoon Vicinity, Geophys. Mag., 44, 27 – 36.

Oyama, R., 2010: Upgrade of atmospheric motion vector derivation algorithms at JMA/MSC, Meteorological Satellite Center Technical Notes, 54, 1 – 32.

Sohn, E. H. and R. Borde 2008: The Impact of Window Size on AMV, Proc. of the Ninth International Winds Workshop, Annapolis, Maryland, USA.

Yamashita, K. 2012: An Observing System Experiment of MTSAT Rapid Scan AMV Using JMA Meso-Scale Operational NWP System, Proc. of the Eleventh International Winds Workshop, New Zealand.

1 / 14