High Resolution Winds from HRV channel PGE09 NWCSAF (updated description including developments for version 1.0, with examples) Table of contents 1.- Goal of HRW product 2.- HRW algorithm summary description  3.- List of inputs for HRW  4.- Coverage and resolution 5.- Description of HRW outputs  6.- Examples of HRW product visualisation Access to "User Manual for the PGE09 of the SAFNWC/MSG Scientific Part" for a more detailed description.

1.- Goal of HRW product

The HRW product aims to provide detailed and frequently updated sets of winds, daytime from HRV SEVIRI-channel data (1km nominal horizontal resolution), as one dynamical information in the SAFNWC/MSG. The product includes wind pressure-level information, and wind quality-control flagging giving some indication of its error in probabilistic terms. This product should be useful in near-real time applications as nowcasting (used in synergy with the diverse other data available to the forecaster) and maybe in mesoscale analysis. 

The targeted –best- scale/resolution is around 12 pixels (i.e. around 12-18km): a 2-scale procedure (24 pixel "coarse" + 12 pixel "fine scale") is suggested as baseline. The single-scale procedure is suggested as some basic option at around 20 pixel (20-30km) resolution. The renewal frequency of the product is 15min (daytime). HRW output is similar to other products of atmospheric motion winds or vectors -AMV- in GTS (BUFR bulletins). Some processing parameters could be configured, in particular the regions to be processed.

 

2.- HRW algorithm description

The development and pre-validation of  HRW algorithms was performed with satellite data available during SAFNWC development phase (MET-7 and GOES-8 imager data). The PGE09 version 1.0 algorithm has been improved and, where needed, adapted to MSG (MET-8) data. Detailed up-to-date description of the algorithm is in the User Manual (Science part).

• Pre-processing: counts for the HRV region are normalised (division by co sinus of sun zenith angle <= 87º and reduction to 8 bits).

• Tracer determination: 

 

• A tracer is a square (12, 24 or 20 pixel side) segment of the first of 2 images, expected to include cloudiness adequate to be tracked. 2 methods are applied in sequence: gradient or sharp edge (throughout the HRV region, uniform grid); and tracer characteristics (only applied to fill as possible 'holes' in coverage of the former).  Average distance between potential tracers is by default set to same to tracer size.

• Tracer location is optimised (around initial search centres); the tracer selection method looks if possible to avoid  the tracer to be contiguous to an already selected one (in what case a later check will  reject it because the `closeness to another target' rejection condition).

• A "continuity option" is included that in a first part computes “persistent tracers” (close in location and level to the final tracer position of a wind included in last wind-set and found of quality enough), then new tracers only where no persistent tracer.  This option is not the default, because  too many  tracers being computed  that could lead to a perhaps too long computing time (if other default parameters are kept, as entire region also for fine-scale tracers processing).  

Gradient (or sharp edge) tracer method : 

• For each location in the uniform grid (unless fine scale part in the 2-scales procedure, see below), the tracer centre will be where the sharpest brightness difference is found around at a distance less than half tracer side, provided brightness and contrast thresholds reached.
  

• It provides most of the tracers at coarse scale. Some ‘tracer characteristics’ (see below) are also computed with less conditions (only to get a brightness 'frontier', and IR10.8 temperature dispersion checked).

Tracer characteristics method : 

• Tests on rough tracer shape, and IR10.8 temperature dispersion, for any potential tracer; evaluated at increasing distance from the centre of the 'hole', until success (got tracer) or maximum allowed distance is reached: 
• In the brightness histogram of the potential tracer, it is searched for significant thresholds or 'frontiers' discriminating 'bright' from 'dark' pixels.
• It is then requested that (for at least one frontier), IR standard deviation for bright tracer pixels not be too large (10K);
• and that 'rough tracer' (simplified to a 4x4 array, each value being 0, 1 or 2 according to bright pixels population) not be too ‘undefined’ (4x4 array checked along linear arrays in 4 directions).

• It gets most of the tracers at fine scale.

In the 2–scales procedure, the fine-scale tracers are searched for, not in a basic uniform grid, but at points given the coarse scale scale search, if one of 2 conditions are met: 

• no coarse tracer (but bright pixels present), 
• or tracer being ‘wide’ (i.e. rough tracer, see above tracer characteristics method)  has values of 2 left and right, or up and down, in the 4x4 array). 

Methods are otherwise applied similar for both scales (as is in general for the next steps described below except that some few coefficients are changed). Up to 4 fine-scale tracers could in principle be computed where only one coarse tracer was possible.

The tracers determination step, is in fact implemented at the end of the chain, as preparation for next step.

• Wind pressure-level assignment:

• IR10.8 representative temperature, and interpolation to a forecast profile (or simple climatological profile).

• Mean temperature for the tracer bright pixels is used (levels 700 to 400hPa), or is corrected to a “cloud base” using the temperature standard deviation (1000 to 700hPa), or the temperature (and pressure) of the non-isolated significant coldest class of the tracer is used (400 to 200hPa).

• Tracer tracking :

• Summed squared difference (LP, used at coarse scale), or cross correlation (CC, fine scale),  tracer to final-image segment of same size, is repeatedly performed inside a large (increased 35x35) segment in the final image centred according to forecast wind (to reduce tracking computations, which are slow in particular CC, nevertheless HRW to forecast wind difference in  rectangular components, can be as high as 150km/h).

• “Gradual approach”: LP or CC is in fact not computed for any centre: each 4 lines and elements in a first iteration;  each 2 and at any centre in the 2^nd and 3^rd iteration, but only around 15 local LP minima or CC maxima at the prior iteration.

• Finally several LP/CC tracking centres retained (absolute and secondary centres).

• Absolute CC (>.44) and difference to absolute (LP/CC) also checked.

• Wind flagging and selection:

• Before computing winds (up to 3 for each tracer) for tracking centres, it is checked if a secondary one changes less in terms of tracer characteristics than the absolute.

• The Quality Indicator method (QI,  EUMETSAT, updated description ) is then applied (with some few adaptations) to any computed wind. As possible, spatial, time (only persistent tracers in the continuity option; otherwise the closest wind in the former slot, if close enough), 2-scales (fine scale wind compared to wind related to a ‘wide’ coarse scale tracer) and to-forecast consistency, tests provide QI flags for all computed winds. Only winds with verified QI threshold useful in spatial, time and 2-scales comparison.

• Only one wind per tracer remains in the HRW product output: best at most available criteria (tracking, tracer characteristics change, all  QI flags), or the significantly best at QI, or best tracking. Optionally (old option): best at spatial plus time QI etc.. As available, spatial, time and forecast QI are summed (weighted sum) to provide the wind overall QI

• In the 2-scales procedure, there are 2 datasets in the HRW product: “basic”, secure winds (coarse scale) and “detailed” winds: fine scale winds plus `non wide’ coarse winds, relocated to the centre of gravity of bright pixels (array 4x4), which are considered as ‘close to fine scale’ winds.

3.- List of inputs for HRW

HRW model configuration file (options and parameters) and region(s) configuration file.

Real time data (each 15min -preferred):

• Data file of HRV counts and IR10.8 temperatures, full resolution, for the region selected, at time t.
• Latitudes and longitudes for both, and sun zenith angles for HRV.
• List of tracers, locations and characteristics at time t-15min (or t-30min) if available, in that case:
• Normalised HRV, and IR10.8, and corresponding latitudes and longitudes, at   t-15min (or t-30min).

NWP data “frequently updated”:

• Fields of temperatures (K degrees x 10) and (u,v) (m/s) rectangular components of the wind. Covering at least the region selected. For as many as possible of the following mandatory p-levels: 1000, 925, 850, 700, 500, 400, 300, 250, 200hPa.

4.- Coverage and resolution

• The objective for PGE09 software is to extract the HRW product on regions included in both HRV and N formats  of MSG (Europe, N. Africa, surrounding seas), regions are selected as rectangles of numlin x numele SEVIRI HRV pixels in satellite projection, and a sub-region (lines and elements) of the coarse scale area can be selected for the fine scale. The validation will be inside that area, A region of 1536x1536  HRV pixels (corresponding to 512x512 in other channels), is retained in particular in  the SAFNWC reference system. Processing and results should not be much dependent of the actual location. 

• HRW should be derived every 15 minutes (daytime), but loss of one time slot is allowed (at the price of degraded quality), the product average resolution (where complete or optimal coverage) will be, as mentioned, around 15km for the detailed dataset (20-30km in other cases).

5.-Description of HRW outputs

 

Two files:  basic(coarse) and detailed winds (or just one in the case of single scale), in form of  BUFR bulletins, are produced for each processed region, each 15min (daytime), with the following content (see SAF/NWC/INM/SW/ICD/3 for description of BUFR structure, see also BUFR standards ):

Basic parameters of the wind:

• Initial latitude and  longitude of the tracer.
• latitude and  longitude increment in the time step.
• Temperature and pressure-level at initial tracer location.
• Direction and speed of the wind .
• %Confidence (overall QI) of the wind.

Processing indicators for the wind:

• Type of tracer (only for the complete HRW dataset: fine scale related to a wide coarse scale tracer or other fine-scale or coarse scale non-wide tracer)
• Pressure-level assignment method (IR/forecast or IR/NWP, plus cloud top or cloud base or cloud medium level).
• Tracking centre selection criteria (whether/not was the best in adjustment and tracer characteristics).
• TEST indicators (4 QI consistency tests applied or not, in the old selection method or if no selection needed; best or slightly or rather worse than other wind or not applied, in the new selection method version 0.1 with 1 more position possible combining tracking and change in characteristics).
• Wind number, and wind number of the predecessor wind (in the continuity option) in the corresponding previous BUFR bulletin.
• Nominal confidence threshold (currently set to 60%).
• Number of winds initially computed for the tracer (1 to 3).
• Number of levels available for the T and (u,v) guess.
• Actual tracer size /side (km).

General processing indicators:

• Satellite and generating centre.
• Year, month, day, hour, minute final image.
• Time increment initial to final image.
• Search-zone increment at nadir (px).
• Instrument and basic method/channel.
• Channel used , central frequency and width.
• Correlation method (plus gap applied yes or not). 
  
• Quality control source (automatic).

6.- Examples of HRW visualisation

The examples were selected from the MSG/MET-8 real time output of PGE09 version 1.0 included in the SAFNWC reference system. The winds are for displacements ending at the displayed image. All the examples for the same belong to the same processing slot. The figure (presented as introduction, top of this page,  beside the index) shows the overall coverage of the generated product (and basic winds same colors as in figures below, but reduced resolution). 

Figure 1: Basic winds (24px tracer side), for the SW part of the region. With QI>59%: high- (magenta), medium- (light blue), and low levels (yellow). Winds between 40 and 59% (all levels) are in dark blue. 

Figure 2: Detailed winds (12px plus some relocated non wide tracers of 24px side). Area and display as in fig. 1

Figure 3: Basic winds, for the NE part of the region. Display as in figs. 1 and 2.

Figure 4: Detailed winds. Area as in fig. 3. Display as in figs 1, 2 and 3.

Back to TOC