RAPIDLY  DEVELOPING  THUNDERSTORM  - Convection Warning (PGE19 V4.0) v2016

  1. Improvement from previous version
  2. Goal of the RDT product
  3. RDT Algorithm & characteristics
    1. The detection of cloud systems
    2. The tracking of cloud systems
    3. The discrimination of convective objects
    4. Forecasting cloud systems
    5. Additional attributes
  4. The output products
    1. Example of visualization
    2. Visualization of additional attributes
    3. A case study
    4. An extended application
  5. The configuration file
  6. The Documentation

Improvement from previous version (v2013 to v2016)

Goal of the RDT-CW product

The RDT-CW (Rapid Development Thunderstorm) product has been developed by Meteo-France in the framework of the EUMETSAT SAF in support to Nowcasting. Using mainly geostationnary satellite data, it provides information on clouds related to significant convective systems, from meso-alpha scale (200 to 2000 km) down to smaller scales (few pixels). It is provided to users in the form of list of numerical data stored in an output file.

The objectives of RDT-CW are :
The object-oriented approach underlying the RDT product allows to add value to the satellite image by characterizing convective, spatially consistent, entities through various parameters of interest to the forecaster : motion vector, cooling and expansion rate, cloud top height … and their time series. It supports easy and meaningful downstream data fusion (surface observations, NWP fields, radar data...).
Thereby, RDT is a tool for forecaster but can be used by research teams, and end-users like aeronautical users.

RDT-CW Algorithm

The RDT process includes four steps:

The detection of cloud systems

The detection algorithm allows to define “cells” which represent the cloud systems. In the RDT-CW algorithm, “cells” are defined on infrared images (channel IR10.8) by applying a threshold which is specific to each cloud system, depending on local brightness temperature pattern. But a pre-processing is included prior to this step:

Processing mask :

CT product

Non cloudy pixels identification

mask for stretched pixels 
(backup file)
stretched pixels + non cloudy pixels

IR10.8 image

Working image for detection step

Cells identification :

The basic idea is to get the threshold adapted to the topography of the cloud top :

Hence, the threshold used for a given cloud tower depends on the temperature pattern in the vicinity, and may evolve just because nearby towers do evolve (click on figure for zooming: base of cloud towers defined by a red contour, as soon as vertical extension reaches 6degC ).

Thus, the RDT-CW cells linked in time to form a trajectory do not necessarily really depict along time the phenomena at the same threshold temperature. But the advantage of the method used (adaptative threshold) is to focus on  convective  parts  of the cloud systems, in order to perform the discrimination process (see below).

In order to increase the relevancy of the cloud contour corresponding to a given threshold, the vertical gradient of contour surface is taken into account for a definitive choice of representative level (base of cloud tower level). It is applied when the vertical morphology of the cloud presents particular shapes, with cloud systems defined by only one tower, and when the chosen representative threshold, automatically set to the warmest value, leads to a bottom contour much larger than the horizontal extension of the top tower itself in its coldest part. Moreover, as detailled later below, a secondary contour, more representative of the top of tower, may complement the cloud system representation.

The tracking of cloud systems

The choice of an adaptative threshold algorithm makes complex the cell comparability due to various phenomena depicted. This method also induces numerous merge and split.
The tracking algorithm is mainly built on the overlapping between cells in two successive images. Before the cells overlap processing, the previous cells are moved according to their (formerly analyzed) move and speed. Nevertheless, correlation or neighboorhood methods are applied when overlapping method doesn't succeed. Moreover, a guess movement field is taken into account through NWP and HRW data for triggering cells without any link with previous cells.
Then, the temporal links are processed as follow:
The temporal links allow to compute move, speed and trend of  all cloud objects. The time series of cloud's characteristics (peripheral gradient, volume, cooling rate...) are key input for the discrimination algorithm.

The discrimination of convective objects

As was mentioned previously, the RDT-CW detection algorithm is able to detect cloud structure from meso-alpha scale down to pixel scale. The goal of the discrimination method is to identify the convective RDT-CW objects among all  cloud cells, adding a strong constraint which is that the discrimination should be effective as soon as possible  after the first detection by RDT-CW software.

The left picture below displays  all  RDT-CW detected cells. It points out  the detection and tracking efficiency of RDT-CW.  We can notice the phenomena and scale diversity of RDT-CW objects.
The right image displays convective objects only.
The ratio between no convective and convective objects is about 100.

 all cells    conv cells  

The discrimination method makes use of discrimination parameters calculated from five MSG channels: IR 10.8µm, IR 8.7µm, IR 12.0µm, WV 6.2µm and WV 7.3µm.
Two kinds of  discrimination parameters are computed:
The discrimination scheme is a mix between statistical models  and empirical rules :
Statistical models for each configuration (processing modes, vertical categories, temporal depth) have been previously tuned on a learning database over widened France. The ground truth used for this data base is cloud to ground lightning occurrence.
The discrimination scheme  has then been validated over a large Europe domain using the ground truth from the extended EUCLID lightning network.

Since v2011 , NWP data is used for a guidance of discrimination scheme when available

NWP mask : 0 = non convective areas = union of 3 non convective indexes

The discrimination scheme is described into ATBD document.

Forecasting convective cloud systems

Convective cells have their own dynamic and can have a trajectory that does not always follow the environmental displacement fields. The object mode of RDT-CW analyses the motion of each cell, and computes this speed. Speed estimate has previously been improved for orphan and triggering cells through movement guess field. This guess field is derived from current HRW winds when available, or in a lower extend from NWP winds. A temporal coherence checking has been added to reduce variability of speed estimate.

The forecast scheme uses this speed estimate to forecast the successive position of each cell.  The Lagrangian method is proved to be quite efficient up to 1 hour range. Thus the forecast is proposed up to this limit.  This leads, for MSG FDSS (15min update rate), to an independant product for each lead range (+15', +30, +45 and +60'). Corresponding products constitute the Convection-Warning part of RDT-CW.

Analyzed (yellow contours) and 60min-forecast (magenta contours) RDT-CW cells

External attributes of convective objects, and 3D description

The output products

The final products concerning RDT-CW are
Example of RDT-CW analyzed map. For a correspondance between encoded values and Cloud type and phase, see DOF documentation

Optimal Visualization
the RDT-CW should need development of a dedicated post-processing, depending on given specifications : access to current cell's localizations and attributes, or access to the recent current trajectories (with historical past series) from a single RDT-CW product, or to the whole trajectory through a serie of RDT-CW products, the latter case implying to manage the temporal links between cloud cells

The Current Object-Netcdf and Trajectory-ASCII formats are described in the Document Output Format (DOF) of SAFNWC. Trajectory file is produced only on user's request (see configuration file). 
For non-regression purposes, RDT-CW may still offer the possibility to encode RDT in BUFR format (preferentially the latest
fourth version including OTD characteristics). But Netcdf output offers more flexible possibilities to display the product.
The case study  below gives an example of a simple RDT-CW operating tool. Because RDT-CW output is a convective object approach allowing downstream data fusion,  RDT-CW objects could also be integrated into more complex systems.

Example of visualization

=>RDT-CW may be visualized at Meteo-France on several platforms:
All visualization tools process current and previous RDT-CW products to display whole cloud trajectories. But Netcdf format already content a recent history of the cloud systems, allowing a direct access to the close trajectory.

Those tools usually combine RDT-CW product and MSG satellite data (either infra-red, visible or RGB), and rely three levels of visualization :
yellow => triggering = warmest categories, i.e. minimum temperature > -25degC
red => growing = minimum temperature > -40degC
magenta => mature categories, i.e. minimum temperature <= -40deg
orange => split cases (inherited convective diagnostic)

=>In NWCSAF real-time production chain, other choices for the vizualisation have been made : the cells are distinguished by their cooling rates.

Example of RDT-CW visualization through SYNERGIE forecaster's tool (left) , through SYNOPSIS forecaster's tool (right)

Other SAFNWC PGEs attributes  , 3D description

CT, CRR and second level

The Figure below gives a full description of a convective cloud cell over Italy. This cell is described at two levels.
Characteristics from external input data are computed over the horizontal cell extension (except for lightning data,  associated when possible to the whole cloud system). For that reason, large cells should offer very different parameter's value  between main and second level.  But this can be the case even for small cells : in the figure below main cell is associated with maximum CRR value of 7mm/h, where second level limits to 3mm/h.

main BT level     2nd level
Main and second level attributes of RDT-CW cloud cell over Italy. Concerned contour highlighted in black

Overshooting top detection (OTD)

Example below illustrate the visualisation of OTDs for a given cloud cell, around 13h UTC on the 25th of May , 2009. This topical case show a several hours lasting convective cell, regenerating on the south-eastern edge and moving eastwards. This cell is exceptionnally associated with 2 OTDs. Their BTD signature is quite high and relevant, as the minimum temperatures. They rise 3-4 deg above NWP tropopause.
The visualization is allowed through punctual localization on the IR image and limited characteritics in the jauge
  • The main points (minimum temperature, maximum of BTD ...) of overshooting top are pointed with a green diamond. 
  • Some of their characteristics may be displayed in the jauge :  temperature, BTD=WV6.2-IR10.8, temperature gap over NWP tropopause
One can note below the U-shape like of the seconf level contour
OTD OTD jauge


A case study (25/05/2009)

The following case study deals with diurnal convection triggering over Adriatic edges, and heavy convection developing in a convergence line in the afternoon over France.
This case study is also detailled for previous releases in SAFNWC Topical Image Gallery  or in RDT-CW web pages
On this visualization tool, yellow lines represent convective cloud
trajectories, and black arrows a 30min move of gravity centers. The contour's colors are the same as mentionned above.
Second level contour appears sometimes depending on cloud morphology, and allow to identify Cloud Cap
Ovsershooting tops appear as green diamond

v2013 20090525

An extended application

The lack of ground truth over Africa didn't allow to evaluate discrimination and precocity skill of RDT-CW on such an area. Nevertheless, the final result seems to be fine for the users involved (take note of better ratio between convective and no convective systems and a stronger cooling rates compared to European cloud systems).

The RDT-CW is now processed over the whole African continent, and displayed through forecaster tool SYNERGIE for French Army, but also for Aeronautical purposes. Below is presented a 3h-loop of "observed" RDT-CW, extended with 1h nowcasts, based on dilatation and advection of diagnosed motion.
On this example, RDT-CW manages small rapid developing convective cells (with deep cooling up to 20-30degC in 15min) , but also large MCSs lasting several hours (displayed here without additionnal level, full trajectory does not appear in this example because limited to 3 hours).
The loop shows good tracking of MCS and good depiction of small cells.

The configuration file

Input Data

        SEV_BANDS         channel bands to be used by PGE VIS06 WV62 WV73 IR87 IR108 IR120 IR134

        CT                        0(default) or 1 if using CT product for additional attributes Cloud Type and Cloud Phase
        CMa                     0(default) or 1 if using CT product for masking non cloud pixels
        CMIC                    0(default) or 1 if using CMIC product for microphysics characteristics
        CTTH                    0(default) or 1 if adding top cloud attribute via CTTH product
        CRR                     0(default) or 1 if using CRR product for additional attribute
        HRW                    0(default) or 1 if using HRW product as guess data for movement estimation

NWP data parameters                           

        NWP_PARAM NWP_SP             Surface pressure : resolution(1) and interpolation method(BLI)
        NWP_PARAM NWP_2T             2m temperature
        NWP_PARAM NWP_2D             2m dew point temperature
        NWP_PARAM NWP_2RH          2m relative humidity (when available)
        NWP_PARAM NWP_SGEOP      Surface geopotential height
        NWP_PARAM NWP_ALTM        altitude of ground (when available)
        NWP_PARAM NWP_T               temperature at pressure levels specified in nwp_conf_file
        NWP_PARAM NWP_RH            relative humidity at pressure levels specified in nwp_conf_file
        NWP_PARAM NWP_GEOP        geopotential height at pressure levels specified in nwp_conf_file
        NWP_PARAM NWP_UW          
U component of wind at pressure levels specified in nwp_conf_file
        NWP_PARAM NWP_VW           V component of wind at pressure levels specified in nwp_conf_file
        NWP_PARAM NWP_TT            tropopause temperature  (when available)
        NWP_PARAM NWP_TP            tropopause pressure (when available)
        NWP_PARAM NWP_LI             Lifted index (when available)
        NWP_PARAM NWP_KI            K index (when available)
        NWP_PARAM NWP_SHW       Showalter index (when available)

        NWPMVTLVL        Pressure level of U/V NWP data as guess data for movement estimation (default 700hPa)

Lightning data parameters

        LGH                     0(default) or N (-1 or N>0) if using lightning data for data blending and additional attribute
                                                            0 (default) : no association
                                                            -1 : association without forcing discrimination result
                                                            N : association with impact forcing convective diagnostic from Nth impact
        LGHDTANT          time limit (sec) before current slot for considering lightning data
        LGHDTPOST        time limit (sec) after current slot for considering lightning data
        LGHTLR               spatial tolerance (nb of pixels - default 3) for pairing lightning stroke with cloud cell
        LGHPROXI           maximum distance (nb of pixels - default 0) for processing distance between lightning stroke and cloud cell

Detection parameters

        TCOLD           cold temperature threshold when multiple thresholding, deg Celsius. Recommended temperature is -75 degC
        TWARM         warm temperature threshold when multiple thresholding, deg Celsius. Recommended temperature is 5degC
        DELTATEMPE temperature step between Tcold and Twarm, deg, default 1
        SMIN             min detection area , km2, default 1. Recommended value is 60
        SMAX            max detection area , km2, default no limit. Recommended value is 200 000

Discrimination options

        CRRDISCRI                 0(default) or >=1 if using high values of CRR for forcing convective diagnostic
        TROPICALDISCRI        0(default) or value(degC) if using colder values of BT for forcing convective diagnostic

Additional options

        DBG                    More precise Informations on program running, default 0 (no information)

Output Data

        FCST                   0(default) or maximum range value (min) for forecast products - nb of products = FCST/update rate
                                                         recommanded value = 60min for FDSS, 20min for RSS  (4 additional forecast products)
        PARALLAX          0(default) or 1 for activation of parallax correction
        SMOOTHPTS      1(default) or >1 for nb of pts taken into account for smoothing contours 
        NCMAPINCLD    0(default) or 1 if map of convective cells included in Netcdf product

        INT_PRODUCT     Production of the ended trajectories file (default NO )
        TRAJ                    Format specification of the ended trajectories file when requested (ex: TISXOLH)
        TRAJPROD           production frequency of ended trajectories file : 0=> monthly   1 => daily   2 => each slot(default)


The Documentation

SAFNWC user documentation for PGE19: