1. HRW High Resolution Winds
15th June 2004
Madrid
José Miguel Fernández-Serdán
ST - STAP / INM

2. Summary HRW product content. Algorithms overview.
Relevant implementation issues in PGE09.
Examples for different conditions or cloudiness.
Some known problems.
Options not in the default configuration.
Further development work.

3. HRW product content In form of BUFR bulletins (observations like):
Basic data:
Latitudes and longitudes ( at t-Δt slot, increment t-Δt to t).
Temperature (K) and pressure level (hPa) for the tracer.
Wind direction (degrees) and speed (mps).
Quality indicator (QI, %).
Additional data:
Time increment (Δt).
Tracer size (= best average resolution) (km).
Recommended quality threshold (set to 60%).
Processing indicators (tracer, tracking, level, guess, q.c. steps) (*).
(*) These are for the most, local descriptors i.e. specific to PGE09/HRW and not found in other AMV BUFR bulletins.
These processing indicators could be taken into account in a detailed validation, or serve to tune the recommended quality threshold.

4. Algorithms overview Steps: Image pre-processing. Tracers.
Guess (NWP)(*).
Level assignment.
Tracking.
Flagging.
And Selection.
(*) Not actually an step, but part of the level assigment or tracking.

5. Algorithms: Image pre-processing
A Region of the HRV channel is processed (*).
(channel most indicated to nowcasting applications).
Raw data are “normalised” ( division by cos-z, z is the sun- zenith angle).
(allow using fixed threshold value).
Range of brightness reduced to values in range (8 bits).
(MOP/MTP heritage, but also because quite low contrasts not much desired at tracking).
(*) In most of the examples: “Spain” (SAFNWC Reference System output).

6. Algorithms: Image pre-processing
HRV brightness
HRV brightness, normalised
Up: sectors before normalisation. Down: same, cos-z normalised (MET-8, 29/03/04 16:00)

7. Algorithms: Tracers 2 methods:
Gradient: search of well defined cloud edges.
(well proven, good, fast).
Tracer characteristics: to fill “holes” in coverage.
Around each “candidate tracer” location, checks for:
A threshold clearly separating “bright” vs. “no bright” pixels.
(the cloudiness or clear part of it, in front of a “background”).
A rough distribution of “bright” pixels showing a “well defined” shape.
(e.g. avoid “too linear” cloudy elements).
And not much scatter in IR-channel temperatures for “bright”.
(avoid multilevel cloudiness) (**).
(**) This test also applied to gradient method tracers. Other tracer characteristics are also computed for these without conditions, for tracking selection purposes.

8. Algorithms: Tracers
Left: Theoretical example showing the importance of the threshold: if just the ‘low’ threshold (left) selected, the tracer would not be accepted (at that exact location): too “linear”.
Right: Example of MOP/MTP VIS coverage (tracer centres) from single-scale procedure; methods gradient -green- and tracer characteristics -red. The red square shows size and shape of a tracer.

9. Algorithms: Guess (NWP)
NWP input is (by now) mandatory .
p-T fields interpolated to centre of each tracer: level assignment.
p-(u,v) fields interpolated to centre and level of each tracer: reduce tracking area.
(full tracking could be too time consuming, nevertheless: guess dependence kept marginal).
Also needed at the flagging step (partial guess QI).
SAFNWC NWCLIB routines are used (*).
(*) Gets as many mandatory levels as possible, for the closest forecast time, using MAX interpolation,etc.

10. Algorithms: Tracers level
The basic input is: mean IR10.8-channel temperature for “bright” pixels (provided by the tracer-characteristics method), converted to pressure.
(these pixels considered as a significant part of the cloudy pixels being tracked).
A “cloud base” is extrapolated for low levels >700hPa (*).
(low level winds known as close to wind at cloud-base level).
A “cloud top” is preferred for high levels <400hPa (the coldest pixels).
Note: IR-channel at 3km vs. HRV at 1km, Ci cloudiness less transparent IR than VIS (**).
(*) Temperature IR corrected by STD_DEV(Temp_IR) x SQRT 2.
(**) The use of CTTH product is a planned option. In this case, the “efective cloudiness” subproduct will also be used. But the CTTH product is still at 3km and representative of IR channels.

11. Algorithms: Tracking
The tracers (selected square segments, HRV region, t-Δt slot) are matched to tracking candidates (same size segments at t-slot, until a “reasonable” search distance) by 2 methods:
Summed squared difference: used for “big” tracers (basic processing, basic scale).
(faster matching).
Cross correlation: for “small” tracers (additional processing for the fine scale).
(more effective matching).
The best match and up to 2 other tracking centres, are kept: up to 3 candidate-winds per tracer.
(to then perform a final selection step)

12. Algorithms: Tracking First image Second image
A tracer (left at 12:00, light blue) and its tracking (next image right at 12:30), in a MOP/MTP case. Tracer centre in both images (light blue). The tracking centre (displaced according to the forecast wind of (u,v) = (8,5) (m/s), and tracking area, in yellow. 2 tracking maxima computed, in green, and red (behind). The first was found to be the good (as one could hope).
First image Second image

13. Algorithms: Flagging (method well known and quite efficient).
Use of Quality Indicators method (QI, Holmund 1998, used at EUMETSAT, adapted in several aspects to HRW) (*).
(method well known and quite efficient).
Normalised tests on consistency are (if possible) computed:
In space (comparison to neighbours t-slot winds, an independent 2 scales spatial test is possible for some fine-scale winds) (**).
In time (comparison to a close wind in the t- Δt slot set of HRW results).
To forecast (NWP).
The final QI is the weighted sum of partial QIs. Given not all test is possible for any wind, an indicator “tests passed” is also added to the QI (***).
(*) Holmlund K., 1998: The utilisation of statistical properties of satellite-derived Atmospheric Motion Vectors to derive Quality Indicators. Wea. Forecasting,
(**) the quality of the wind to be used as reference is previously checked, also in other tests (depending of the option). The 2-scales QI is used in selection, but not summed in the final QI.
(***) One goal with this and other processing indicators could be to get tuned “recommended quality thresholds”, by now a fixed value.

14. Algorithms: ... And Selection
Up to 3 winds per tracer were tracked and flagged, only one wind per tracer is kept for output in the HRW bulletin (BUFR code).
(implies less information but more straightforward to use).
The (default) way of selecting is: to keep the best at most selection criteria (up to 6):
Best matching.
Less change in tracer characteristics.
And the best at each partial QIs (up to 4).
If not yet decisive: the one (significantly) more consistent when compared to the guess. Or the best tracking.

15. Relevant implementation issues in PGE09
A 2-scales procedure (default):
24px coarse scale and 12px fine scale (*).
Most of the procedure is performed twice, similar (**).
Search of fine-scale tracers: where no coarse scale tracer found, or it is “wide” (***).
“Inter-scale” spatial consistency test tried for fine-scale winds in the case of “wide”.
(fine-scale complementary to coarse: where no coarse, or it suggests more detail in wind field can be reached).
The HRW output composed of 2 datasets (BUFR bulletins):
Basic winds: coarse-scale winds (SAFNWC_..._B.buf ).
Detailed winds: fine-scale winds, plus coarse non wide-tracer winds (^) (SAFNWC _..._D.buf.
(“secure” vs. “dense” wind datasets) .
Tracers step is implemented as the final step (t-slot preprocessing for t + Δt HRW winds).
(*) Size in pixels (of the HRV channel) means the tracer side, and the distance between centres of candidate tracers in the initial search.
(**) Few processing parameters or options are changed, e.g. Matching algorithm.
(***) Meaning significant populations of “bright” pixels found in distant locations of the tracer, suggesting more detail could be reached inside the coarse tracer.
(^) Non-wide coarse winds are understood as representative at a best than coarse scale (and slightly relocated for inclusion in the Detailed wind dataset).

16. Examples for different conditions or cloudiness
Lee cloud conditions.
Convection at different stages of development.
Cloudiness related to small-scale circulation.
Low sun elevation.
Slant satellite viewing angle.
Frontal systems (including semi-transparent Ci).
Display:
QI > 59% : >700hPa hPa. <400hPa.
(QI 50-59%)

17. Examples: Lee cloud conditions (23/03/04 at 7:00 and 10:00 UTC)
Detailed winds
Detailed winds (_D, QI>60%) for the slot ending at z. Lower: same for 10z.

18. Examples: Convection at different stages of development (01/06/04 at 15:00)
Basic winds
Detailed winds (_D, QI>60%) for the slot ending at z. Lower: same for 10z.
Detailed winds

19. Examples: Cloudiness related to small-scale circulation (07/06/04 7:45 to 8:30)
VIS0.6
The zone of interest for HRW is the N half of the Iberian Peninsula (conveccion developed during the day in areas of NW and W).
The basic winds are shown and, for a reduced zone of major interest, the detailed winds.
PGE09 first winds were that day at 7:45 slot. Detailed winds at 8:30 panel are not shown, to allow visibility of the cloud patterns.
IR10.8
Detailed winds
Basic winds

20. Examples: Low sun elevation (07/06/04 18:45 and 19:15 UTC)
Despite the HRV very dark over Spain, a lot of winds were still computed at 18:45. At 19:15, only a few, corresponding to bright, thick cloud tops including Cb.
These are basic winds.
The HRV image is shown as background over most of the Peninsula only.

21. Examples: Slant satellite viewing angle (14 to 16/10/03 at 10:00 UTC)
Basic 14 15
The wind pattern (NE edge) seems coherent. The situation changes little, and (as seen in the animation), the wind component is mainly N, and the winds are certainly strong. 16
Detailed

22. Examples: Frontal systems (including Ci) (16/04/04 at 7:15 UTC)
This is the basic dataset. The general flux seems well explained, including the weak frontal band (W).

23. Some known problems
“Ground tracers” (most evident in case of snow). Others: ground-influenced cloudiness.
The “high” normalised threshold requested (same land and sea) misses some low-level cloudiness not bright enough to give tracers (e.g. many Sc, could also present large extensions of little contrast).
The HRW detailed wind pattern still not as “high resolution” as could be desired (e.g. fields of small Cu likely giving few tracers with the described methods). Also, more often than in the case of basic winds, erroneous detailed winds reach a QI of 60%.
A fixed QI reference threshold (60%) (despite the disparity in processing conditions, and applied QI tests, etc.) (*).
The level-assignment based on IR information.
(HRV mis-registration image to image: not yet clear cause of error since MSG data operational).
(*) e.g. the case of close, erroneus but similar, winds (often paralel to ellongated cloud structures: the space consistency test is here to be blamed.

24. Options not in the default configuration
Wind continuity:
Tracers step: at each location, first looks close (in position, level, at same scale and same tracer method) to final location of any wind t-Δt: “persistent tracer”. Otherwise tries for a new tracer.
Time consistency QI tests: only for persistent tracers.
(time consistency closer to EUMETSAT’s symmetry test; provides some tracers’ trajectories and durations).
But: time consuming (more new tracers than expected): a sub-region for the fine-scale is advised in this option.
Single-scale procedure (similar to coarse-only, at 20 pixel resolution or better):
(a compromise, and faster, but gives “less complete” results).
Wind selection on best QI (space&time) (PGE09 v0.0) (*).
(*) Chosen to get winds less dependent on guess, in the future a simple option to just consider the highest final QI should be included.

25. Options not in the default: Persistent tracers
Using Continuity option
Basic winds dataset, 26/03/04
Upper: with option persistent tracers, Basic winds (_B, QI>60%) for the slot ending at :45 07z.
Lower: with default option, 12z.
Using Default

26. Further development work (*)
Addition of a terrain-influence flag (Fernández, 2003)(**), identifying:
Possibly non-cloud tracer (just land feature) (but also: plans to use cloud-top product CTTH)
Tracer possibly blocked by terrain (could perhaps give non realistic winds)
Tracer likely in presence of gravity wave conditions (could give non realistic wind)
New work (2004):
Improvement of lee-wave conditions flagging (avoid fixed vertical lapse- rate and –maybe- fixed distance to obstacle).
Use of terrain-influence flag in wind selection.
Use of CTTH, as option for level-assignment.
(*) Not part of PGE09 version 1.0.
(**) Fernández-Serdán J.M., 2003: Enhancement of algorithms for satellite motion-derived winds: the high-resolution and quality control aspects. Proceedings of The 2003 EUMETSAT Meteorological Satellite Conference, EUM P.39, pp

27. Further development: terrain-influence flag