Automatic Satellite Image Interpretation  PGE10  NWCSAF     Table of contents 1.- Goal of ASII product  2.- ASII algorithm summary description  3.- List of inputs for ASII  4.- Coverage and resolution 5.- Description of ASII outputs 6.- Example of ASII visualisation Access to "Algorithm Theoretical Basis Document for the "Automatic Satellite Image Interpretation" product (ASII-PGE10, v2.2)" for a more detailed description.

1.- Goal of ASII product

The aim of this SAFNWC product is an Automatic Satellite Image Interpretation in terms of conceptual models (CM). CMs are a synthesis of physical processes and the typical features they cause in satellite images, as well as in other synoptic material like for instance in numerical model output parameters. A CM diagnosis can be used for a deepening of synoptic diagnosis and for the early detection of ongoing meteorological processes, as well as for the determination of the stage of development within those processes.

PGE10 carries out the recognition of CMs in two ways:

• ASII:     Automatic Satellite Image Interpretation from MSG SEVIRI satellite data alone
• ASIINWP:     Automatic Satellite Image Interpretation from MSG SEVIRI satellite data supplemented by typical key parameters from the numerical model output (Numerical model used: ECMWF)

2.- ASII algorithm summary description

Basic concept

The basic method is the use of pattern recognition methods, which are applied to extract fundamental information concerning the shape and location of typical cloud features. These pattern recognition methods are:

• Detection of frontal areas
• Detection of S-shaped lines
• Detection of circular-shaped cells
• Detection of black stripes in WV images
• Detection of narrow cloud fibres
• Detection of spiral cloud structures

For each CM a combination of necessary and appropriate pattern recognition methods is applied. For a detailed description of the pattern recognition methods used for each CM see the relevant chapters in the Product User Manual (PUM).

 

Conceptual model	Chapter PUM
Cold front	2.3.1.1
Cold front in warm air advection	2.3.1.1
Warm front	2.3.1.2
Occlusion	2.3.1.3
Wave	2.3.1.4
Developing wave	2.3.1.4
Upper wave	2.3.1.5
Front intensification by jet	2.3.1.6
Dry intrusion	2.3.1.7
Upper level low	2.3.1.8
Comma cloud	2.3.1.9
Enhanced cumulus	2.3.1.10
Cumulonimbus	2.3.1.11
Mesoscale convective system	2.3.1.12
Cold air cloudiness	2.3.1.13
Fibres	2.3.1.14
Lee cloudiness	2.3.1.15

Table 1: List of detected conceptual models in ASII

Introductory steps

After a preprocessing of the IR image, Atmospheric Motion Vectors (AMV’s) using the standard cross correlation technique are derived and motion-corrected difference images are computed. Then a basic pattern recognition method namely image segmentation and classification, is used. Segmentation divides the IR satellite image into sub-regions being coherent in terms of brightness and texture. The classification process assignes qualities of CM’s to these coherent areas.
 

Pattern recognition methods

Besides the basic pattern recognition methods mentioned above, more complex pattern recognition methods are summarised below.
 

• Detection of frontal areas:

  
  The identification of coherent areas of sufficiently large size plays an important role e.g. in the recognition of frontal systems and of S-shaped bulges (waves). The number of contiguous pixels above a certain brightness (temperature) threshold is determined. Contour lines encircles extended frontal areas. For this process, images compressed by the Gauss pyramid method are used.

   

• Detection of S-shaped lines:

  
  The recognition of S shapes is applied to contour lines. The algorithm for the identification of S-lines is comprised of the following steps:

  • Derivation of the contour lines.
  • Through each pixel of a contour line, a batch of straight lines with given radius and angle increment is constructed.
  • If a straight line exists which cuts the contour line at 3 disjoint locations the presence of an S-line is assumed

   

• Detection of circular shaped cells:

  Convective cells are recognisable by their local temperature (brightness) maximum and their compact circular or elliptical shape. The steps of their detection are:

  • Extraction of the local temperature minimum (brightness maximum) being colder (brighter) than a given threshold.
  • Around each of these points, a number of concentric circles with prescribed radii is constructed.
  • For each circle, one investigates whether a certain percentage of all pixels lying on the circle exhibit a minimum difference to the brightness of the centre. This criterion helps to eliminate homogeneous and fibrous structures. Each maximum passing the test is marked as the centre of a convective cell.

   

• Detection of dark stripes in WV images:

  The detection of dark stripes has the following principal steps:

  • For each pixel, it is checked whether the pixel is dark enough, i.e. whether it is darker than an empirical threshold.
  • It is verified for each candidate pixel that it is darker than most of its neighbours.

   

• Detection of fibres:

  The detection of fibres is based on an algorithm which was originally developed for the detection of black stripes in WV imagery. The fibre detection is based on a temperature threshold, a temperature gradient between the fibre and its surroundings, a radius in which the criterion for fibre is checked and a mesure for the spatial extension of the fibre.

   

• Detection of spiral cloud structures:

  
  The detection of cloud spirals is done via the three following steps:

  • Cloud texture:

    The method is based on the direction of texture elements, i.e. the direction in which the local features of the satellite image appear to be oriented. This parameter can be assessed quantitatively by applying a Sobel filter to the image, with subsequent filtering of noise by applying a series of median filters.

  • Curvature analysis:

    Streamlines of these texture elements are determined and, finally, the curvature of those streamlines is computed. A large curvature radius indicates straight streamlines, whereas a strong curvature is indicative for spiraled cloud features.

  • Hough knots:

    The final step is obtained via the so-called Hough knot method. In case of spirals, lines perpendicular to the texture elements will cut in a point called Hough knot.

     

3.- List of inputs for ASII

MSG radiance values:
The following SEVIRI channels are required at full MSG resolution, for the current and the previous slot:

• R12.0m Mandatory
• R10.8m Mandatory
• R6.2m Mandatory


ECMWF NWP data:

• Mandatory for the ASIINWP product. Required levels: 1000, 850, 700, 500, 400, 300 hPa. Required basic parameters: wind, temperature, geopotential height, humidity.

4.- Coverage and resolution

The ASII products deal with pattern recognition on a synoptic scale. For products of this kind, it is beneficial to consider an area as extended as possible. Therefore, the coverage is fixed to MSG-N.

The products will be derived every 15 minutes, and the designations of analysed conceptual models are given on a grid with a mesh size of ~70 km.
 

5.-Description of ASII outputs

The ASII and ASIINWP products are output in separate BUFR files. These files contain a certain number (Version 2.2: 24) of telegrams which describe the longitudes and latitudes of the grid points where conceptual models are diagnosed. Details about the template of the BUFR telegrams can be found in the Algorithm Theoretical Basis Document (ATBD chapter 3.2.2).

6.- Example of ASII visualisation

As example, the 14^th October 2003, 06 UTC has been chosen.

Visualisation of the ASII product (satellite information only):
A north-south oriented cold front cloud band has been analysed by ASII (blue c) over the Atlantic. At the rear side of the frontal cloud band, a frontal wave (~) and CBs (cb) have been detected. The northern cloud spiral has been analysed as occlusion (geen o). The area of cellular cloudiness behind the cold front is analysed as enhanced cumuli (ec). The onset of a warm front has been detected by ASII near Greenland (red w).
Over Algeria and the Mediterranean Sea, another frontal system has been analysed as cold front (blue c) and as a warm front (red w). A small occlusion is located over Spain (green o) and embedded mesoscale convective systems over Sardaigna and the Mediterranean Sea (dots)

Visualisation of the ASIINWP product (additional NWP fields):
The ASIINWP analysis of the same cold front over the Atlantic give a more differentiate image. Additional to the cold front analysis, some red c tags can be seen indicating warm air advection. The frontal wave is confirmed by the model fields, moreover an intensification of the cold front in the left exit region of a jet streak has been detected (ji) and an upper wave at the same postion on the rear side of the cloud band. The occlusion of the northern part of the system has been confirmed, the inner part of the cloud spiral is reanalysed as comma cloud (co). The enhanced cumuli are confirmed by NWP data.
The cold front over Africa is shown to be almost completely under warm air advection. The mesoscale convective systems over the Mediterranean Sea are located in an unstable air mass.

For a complete list of the used symbols see table 2.

 
 Table 2a and 2b: Symbols used for the visualisation of the ASII products.
2a: SAT: IR and WV satellite image as sole input
2b: SAT+NWP: NWP fields are added to the satellite image data