AMV products

by Mary Forsythe — last modified Jun 11, 2014 10:50 AM

This page is organised into 5 sub-sections covering aspects of the derivation which would benefit from optimisation for producing high resolution AMVs

and 2 sub-sections highlighting ongoing work and coordinated activities

Background

Most global AMVs are produced using ~24x24 pixel targets. Several studies have looked at the impact of varying this and the image interval (e.g. Sohn & Borde, 2008). Reducing the target box size leads to noisier tracking results as there is a greater likelihood of incorrect matches, but tends to improve the AMV speed bias. To produce local flow we need to reduce the target sizes (probably to somewhere in the range 5x5 - 12x12 depending on pixel resolution - do we have any consensus on this?) so the question becomes how best to deal with the noise.

4 options include

(i) using clustering (as in the GOES-R nested tracking) to filter out outliers (e.g. Daniels & Bresky, 2010)

(ii) using information from the matching surface to filter out AMVs where the tracking is not well resolved (e.g. if multiple maxima or broad maxima - could construct a 2D version of the Met Office model best-fit pressure constraint)

(iii) running the tracking forward from 1->2 and back from 2->1 as a further check of reliability

(iv) use of multiple sequential vectors produced at high temporal resolution using rapid scan (e.g. Shimoji, 2010)

References:

García-Pereda and Borde, 2014: The impact of the tracer size and the temporal gap between images in the extraction of Atmospheric Motion Vectors. J. Atmos. Oceanic Technol. (in press).

Sohn and Borde, 2008. The impact of window size on AMV, Proceedings of IWW9\\

Shimoji, 2010. The Development for MTSAT Rapid Scan High Resolution AMVs at JMA/MSC , Proceedings of IWW10

Daniels and Bresky, 2010. A new nested tracking approach for reducing the slow speed bias associated with Atmospheric Motion Vectors (AMVs), Proceedings of IWW10\\

Suggested work and open questions:

Evaluate techniques to reduce noise from use of smaller targets.

How small do we need to go to represent the local flow?

Background

Most global AMVs are produced using ~15-30 min interval imagery. Several producers operate satellites in rapid scan mode (taking images every 2-10 min) either as a dedicated service (e.g. EUMETSAT operating Meteosat-9 over Europe) or on demand to provide more timely imagery in regions of severe weather. The rapid scan imagery can be used to produce AMVs. Something to bear in mind is that to resolve a useful AMV requires the cloud pattern to move at least ~one pixel between images. This will depend on the channel resolution. For example using 5 min interval imagery the cloud would have to move at a speed of at least 10 m/s for 3 km pixel size and 3.3 m/s for 1 km pixel size. To avoid losing all the slower AMVs it may be best to use 10 min interval for channels with resolution ~ 3 km at sub-pixel point. However, it may be possible to test using shorter intervals for the higher resolution channels (less than or equal to 1 km at sub-pixel point). In Velden et al., (2005) it is noted that the ability to retrieve mesoscale cloud motions is notably enhanced using 3-5 min image interval.

References:

García-Pereda and Borde, 2014: The impact of the tracer size and the temporal gap between images in the extraction of Atmospheric Motion Vectors. J. Atmos. Oceanic Technol. (in press).

Sohn and Borde, 2008. The impact of window size on AMV, Proceedings of IWW9\\

Shimoji, 2010. The Development for MTSAT Rapid Scan High Resolution AMVs at JMA/MSC , Proceedings of IWW10

Daniels and Bresky, 2010. A new nested tracking approach for reducing the slow speed bias associated with Atmospheric Motion Vectors (AMVs), Proceedings of IWW10

Velden et al., 2005. Recent Innovations in Deriving Tropospheric Winds from Meteorological Satellites. BAMS, 86, 205-223.

Suggested work and open questions:
Is there consensus in IWWG community on using 10 min interval for everything except channels with pixel resolution at SSP of less than or equal to 1 km?

Avoid overlapping targets
Probably little point attempting for CSWV features as tend to have broader structures that are unlikely to provide good tracers using small target sizes and image intervals.

Adapt target selection criteria to produce better coverage? What density is best for NWP? Different densities can be produced easily with the NWC SAF software, but interpreting results on the NWP side may be less clear as results of impact experiments can be very noisy. It may be more important in some cases e.g. near tropical cyclones.

Provide more accurate time stamps for each AMV
Borde and Garcia-Pereda (2014) show improved wind fields when the model is not used as a first guess in the tracking (some centres do this to reduce computation). This may be more important for high resolution winds where the model may not capture the detailed flow field. The use of model information in the tracking step is an optional feature of the NWC SAF software.
Hasler et al., (1998) describe an alternative approach to tracking using a Massively Parallel Semi-Fluid Analysis (MPSMA), which allows for cloud deformations and the complex behaviour of multilayered clouds typically seen within the central dense overcast (CDO) of a hurricane or severe storm.

References

Hasler et al., 1998. High resolution wind fields within the inner core and eye of a mature tropical cyclone from GOES 1-min images. BAMS, 79, 2483-2496.

Background
The AMV QIs are constructed from spatial and temporal consistency checks with neighbouring observations. This is a useful approach for removing rogue vectors, but may also penalise information on the local flow which is likely to be more variable than the coarser resolution global AMVs. One topic (also a key aim for the global AMVs) is to produce information on the AMV quality directly from the derivation approach (e.g. estimates of height error from the pixel-based height schemes, ?estimates of error in the tracking from information in the tracking surface etc). These could be used instead of the QIs.

THORPEX / TPARC campaigns for validation.

Use of full resolution sondes, profilers and MISR

NWC SAF maintain and develop AMV derivation software for production of high resolution AMVs. The standard modes are 12x12 or 24x24 pixels, but the software is easily configurable to run with different options. The software has been used to evaluate various settings including:

varying target box size (8x8-40x40) and image interval (5-90 min) - (Garcia-Pereda and Borde, 2014 in press)

use of model first guess to guide the tracking. (Borde and Garcia-Pereda (2014) show improved wind fields when the model is not used as a first guess in the tracking (some centres do this to reduce computation). This is likely to be particularly important for high resolution winds where the model may not capture the detailed flow field. The use of model information in the tracking step is an optional feature of the NWC SAF software.

Met Office (UK) are using the NWC SAF software to produce AMVs for use in the UKV model (1.5 km resolution). Currently using 24x24 pixel resolution and 15 min interval, but plan to investigate use of smaller target box size and image interval. Work will be presented at IWW12.

JMA have been actively working in this area for a few years, both producing AMVs at higher spatial resolution and/or from rapid scan imagery. Key results have been improvement on rainfall forecast skill and tropical cyclone tracks. Results will be presented at IWW12.

CIMSS have been involved in this area for many years. 2 proposals have been provisionally accepted. One is through GOES-R program, and one is through the Hurricane Forecast Improvement Program (HFIP). Focus will be on reprocessing AMVs using the special 1-min scans from GOES-14 in high-res DA and hurricane models. Tropical cyclones (e.g., Sandy) and severe weather events are targeted.

BoM produce their own high resolution winds from MTSAT imagery and have demonstrated improvements to forecasts through assimilation in NWP (Le Marshall et al., 2013)

References

Le Marshall et al., 2013. The Operational Generation of Continuous Winds in the Australian Region and Their Assimilation with 4DVAR, //Wea. Forecasting//, **28**, 504–514.

Dependent on discussion at IWW12