NWCSAF Convection products: v2018 improvements, validation and - - PowerPoint PPT Presentation

▶
nwcsaf convection products v2018 improvements validation
SMART_READER_LITE
LIVE PREVIEW

NWCSAF Convection products: v2018 improvements, validation and - - PowerPoint PPT Presentation

Sep 13, 2022 125 likes •395 views

NWCSAF Convection products: v2018 improvements, validation and adaptability to end-users Eumetsat 2019 October, 3 rd , 2019 Claudon, M., Moisselin, J.-M., Autons, F. NWCSAF GEO Products: storms monitoring at different development stages. A

slide-1
SLIDE 1

NWCSAF Convection products: v2018 improvements, validation and adaptability to end-users

Eumetsat 2019 October, 3rd, 2019 Claudon, M., Moisselin, J.-M., Autonès, F.

slide-2
SLIDE 2

Eumetsat Conference, Boston 2019

NWCSAF GEO Products: storms monitoring at different development stages. A portfolio for convection

Time

Any time Cloud products (CMA, CT, CTTH, CMIC), High Resolution Winds (HRW), ASII

Pre-convective environment iSHAI (imaging Satellite Humidity and Instability) Convection Initiation CI (convection initiation) Developing convective storm RDT-CW (Rapidly Developing Thunderstorm) Precipitation products

Courtesy NWCSAF LE

slide-3
SLIDE 3

Eumetsat Conference, Boston 2019

Overview

1.CI : Convection Initiation

  • 2. RDT: Rapidly Developing Thunderstorm
  • 3. Conclusion
slide-4
SLIDE 4

Eumetsat Conference, Boston 2019

Convection Initiation at a glance

Probability for a pixel to become a thunderstorm

First version: v2016

adaptation of SATCAST methodology (Best Practice Document, 2013, for EUMETSAT Convection Working Group, Eds

J.Mecikalski, K. Bedka and M. König)

Visiting Scientist Activity (Karagiannidis, A., 2016, Final Report on Visiting Scientist Activity for the validation and

improvement of the Convection Initiation (CI) product of NWC SAF v2016 and v2018, Visiting Scientist Activity followed in Nowcasting Department of Météo France, Toulouse, France Period June-December 2016)

Now : v2018 (PRE-OPERATIONAL status)

Input:

  • Satellite data (multiple channels), Numerical Weather Prediction data, NWC SAF products: Cloud

Products (CT, CMIC), HRW Output:

  • NetCDF Pixel-based product, with 4 classes of probability (very low, low, medium, high) and 3 forecast

periods (30, 60 and 90 minutes)

V2018 improvements:

  • Use of CMIC and Cloud Type for the identification of areas of interest
  • New tuning, including a mode for daytime and nighttime
  • Tracking improvement, Forecast horizon extension, Parallelization capabilities
  • Validation improvement (AS work by TROPOS from Leibnitz Institute)

Validation improvement (AS work by TROPOS from Leibnitz Institute)

slide-5
SLIDE 5

Eumetsat Conference, Boston 2019

CI product validation process

MSG Qualitative Validation TROPOS Quantitative Validation

2018 NWC SAF Validation Report for Convection products*

Radar data as ground truth Pixels with reflectivity over threshold (30 or 35 dBZ) Radar data as ground truth Newly developed

  • bjects with

reflectivity over threshold (35 dBZ) GOES-16 Qualitative Validation MSG Quantitative Validation Radar-derived convective objects as ground truth Tracked radar convective cells Use of parameters (age, lightning pairing, cloud top pressure) Radar-derived convective

  • bjects as

ground truth

V2018.1 patch FUTURE

* http://www.nwcsaf.org/Downloads/GEO/2018/Documents/Scientific_Docs/NWC-CDOP3-GEO-MF-PI-SCI-VR-Convection_v1.0.pdf

slide-6
SLIDE 6

Eumetsat Conference, Boston 2019

MSG-IR10.8 + CI probability [0’-30’] product 13h00Z + Ground Truth = radar > 30dBZ [13Z-13h30Z] Generally relevant, even if all cases encountered:

  • Good Detection (GD)
  • False Alarms (FA)
  • Misses (MI)
  • Double penalties

The relevancy is clearly to analyse regarding the situation

(isolated, embedded, edge of cloud systems, etc.)

CI v2018 - Validation on MSG case studies Radar Ground Truth

CI classes GT

slide-7
SLIDE 7

Eumetsat Conference, Boston 2019

CI v2018 - Validation on case studies - Summary

FAR problem seems the main one

  • Sometimes explained by spatial double penalty as CI not so far away from new

convective clouds

  • Sometimes explained by delayed convection (CI [0-30’] should have been CI [0-

60’]) Less relevant in cold-air mass. Explanations:

  • Threshold to be tuned
  • Fractioned cloud type excluded of CI calculation
  • Movement field more difficult to assess in that case

Useful signal for forecasters or other experienced users. As additional information (rather than replacing other ones)

slide-8
SLIDE 8

Eumetsat Conference, Boston 2019

CI v2018 - Quantitative validation methodology

Cloud Object Radar Object

✗ HRV field

filtered using a Gaussian filter

✗ Minimum life time

= 30 minutes

✗ Minimum object

size = 10 pixels

✗ Connectivity-type

= 8 pixels

✗ Parallax-

corrected tracks

✗ Motion of cloud

fields using the TV-L1 optical flow algorithm

✗ Newly

developing convective cells

✗ Reflectivity

threshold at 35 dBZ

✗ Minimum life

time = 30 minutes

from TROPOS AS report

http://www.nwcsaf.org/aemetRest/downloadAttachment/5278

CI level of probability POD FAR Level 2 (low prob) 0.52 0.76 Level 4 (high prob) 0.28 0.50

slide-9
SLIDE 9

Eumetsat Conference, Boston 2019

CI v2018 – GOES-16 qualitative validation Definition of radar GT (ground truth)

Ground Truth NEW RADAR CONVECTIVE OBJECTS PAIRED WITH LIGHTNING, ‘WARM’ CLOUDS Spatial Tolerance 0.1°

ALL RADAR CONVECTIVE OBJECTS

NEW RADAR CONVECTIVE OBJECTS

2018-10-02T19:30:00Z

IR satellite imagery superimposed with radar reflectivity (threshold 32dBZ, green / yellow) and radar convective objects (red contours)

French Guiana

slide-10
SLIDE 10

Eumetsat Conference, Boston 2019

CI v2018 – GOES-16 validation - Results

Infrared satellite imagery superimposed with:

  • Already existing radar convective objects

(threshold at 32 dBZ, cyan contours)

  • New radar convective object = Ground truth

(threshold at 32 dBZ, green contours). Birth at 19:30:00Z

  • CI forecast [0;30min]. Yellow to magenta

Miss

CI at 19:05:00Z, GT birth = 19:30:00Z CI at 19:35:00Z, GT birth = 19:30:00Z

Late good detection

2018-10-02 Case study summary

➢ CI Radar signature appears often before satellite

  • ne.

➢ When taking into account 0.1° spatial tolerance

around GT, there are lots of late good detections (80 %).

➢ In CI production, use of IR10.3 is similar to IR11.2,

but produces significantly less false alarms (not shown).

slide-11
SLIDE 11

Eumetsat Conference, Boston 2019

Overview

  • 1. CI: Convection Initiation

2.RDT: Rapidly Developing Thunderstorm

  • 3. Conclusion
slide-12
SLIDE 12

Eumetsat Conference, Boston 2019

v2018 RDT - A well-known product

Detection, tracking, characterisation and nowcast of convective cloud systems └ (stage, severity, overshooting tops detection

  • vershooting tops detection …)

✔ Labelled as Operational (Eumetsat sense)

✔ V2018 improvements:

  • Improved and configurable detection (thanks to users’ feedback)

limit for upper cloud cells’ contours with tropopause and/or fixed T°

  • Lightning jump detection

Lightning jump detection

rapid increase of lightning trend correlated to - and sometimes precursor of - severe weather (hail ...) analysis of trends at lightning data time scale linked to lightning network performances (Cloud-to-Ground and IntraCloud) input to severity attribute

  • Additional Discrimination scheme adapted to and tuned with current satellite configurations

v2018.1 adapted to GOES16 adapted to GOES16

  • Other (high altitude Ice Crystal calculation, new lightning pairing rules ...)
slide-13
SLIDE 13

Eumetsat Conference, Boston 2019

Validation of Overshooting Tops (OT) Detection within RDT (1/3)

Pairing method between CHMI-OT and RDT-OT

  • - Time synchronisation : area scanned approximatively +11’ for MSG/FDSS, 3’ for

MSG/RSS, ~ 1’ for MSG-2.5’

  • - Time tolerance: maximum 5’ (RSS) or 15’ (FDSS) between RDT-OT and CHMI-OT
  • - Spatial tolerance: 20 km maximum distance (~ mean OT size)
  • - Score calculation:

✔ HIT: at least one RDT-OT associated to a CHMI-OT ✔ MISS: CHMI-OT without associated RDT-OT ✔ FA: RDT-OT without associated CHMI-OT Expertised CHMI1 OT database (published by CWG2)

  • 2.5’ experimental MSG1 scan

20130620 [09h-19h30] and 20130729 [13h-18h30]

  • 1800 OT identified
  • limited area over central Europe

(1)Czech HydroMeteorological Institute (2)Convection Working Group

Automatic OTD within Reprocessed RDT 4 configurations :

  • FDSS-15’ and RSS-5’
  • v2018 and devt version with use of HRV
slide-14
SLIDE 14

Eumetsat Conference, Boston 2019

RDT-OT vs CHMI-OT (2/3) – 20130620 case study

11h45Z(+11’) 12h00Z (+11’) 11h57Z 12h10Z 12h12Z HI MI ? MI ? Expert CHMI-OTs:

  • much more numerous
  • high space and time variability of OTs from one slot to the other

RDT-OTD:

  • seems subjectively more or less OK,
  • lot of misses (or FA?) due to scan rate differences
  • correspondance not so obvious

MI FA ?

RDT-OTD CHMI-OT

FA ? 12h07Z

slide-15
SLIDE 15

Eumetsat Conference, Boston 2019

  • Large number of misses, due to large scan rate differences (~1800 expert OTs vs ~700 RDT-OT

RSS and ~200 RDT-OT FDSS)

Expected low POD = HI/(HI+MI) or TS=HI/(HI+FA+MI) Focus on FAR=FA/(HI+FA)

RDT-OT vs CHMI-OT (3/3) – Quantitative Results

RDT-OTD vs CHMI-OT

➔ Globally better results with RSS mode for RDT-OT

High frequency scan rate allow easier detection of short-lived OT (  for MTG)

➔ Use of HRV (reflectance+gradient) slightly improves scores

seems to lower FA on some cases, to increase HI on others, but limited impact further studies needed (HRV texture)

➔ Results dependent on mode and day

better scores on 20130729, especially with RSS mode Signatures of RDT-OTs’ parameters (BTD, BT, reflectance) differ from one day to the other

slide-16
SLIDE 16

Eumetsat Conference, Boston 2019

Validation of Lightning Jump (LJ) Detection within RDT (1/3)

Pairing method between severe events and RDT-LJ

  • Visualisation of RDT cells with LJ diagnosis prior or close to hail events from both ESWD

database and HYDRE data

  • Europe: ESSL European Severe Weather

Database (reports hail, wind gusts, tornadoes,

lightning damages)

  • France: HYDRE MF data fusion product (5’)

for hydrometeor diagnosis (focus on

medium/large hail ...)

electric activity at fine time- scale for a RDT cell

OR

RDT LJ FDSS v2018 with Meteorage and partners network (CG+IC) Reprocessed on case studies

Two references (Ground Truths)

slide-17
SLIDE 17

Eumetsat Conference, Boston 2019

20180529 case study:

  • [15h30-16h00] RDT (contours)
  • [16h-16h15] HYDRE medium and large hail detection (accumulated pixels)

All convectives RDT cells RDT cells with Lightning Jump

Lighning Jump Validation (2/3) - RDT-LJ vs HYDRE Hail detection

+ Good colocation RDT/rain + Subjectively good co-location Hail/RDT-LJ + RDT-LJ sometimes precursor of Hail event

HI HI HI FA? MI

  • anticipation difficult to assess

(still need to be quantified/confirmed)

  • need detailed analysis of each hail event
  • isolated hail pixels to consider ?
slide-18
SLIDE 18

Eumetsat Conference, Boston 2019

14h-15h 15h-16h 16h-17h 19h-20h 14h-15h 15h-16h 18h-19h 13h-14h

  • Step by step analysis of RDT-LJ sequences vs following SW allow subjective good pairing
  • Most severe weather events match with previous RDT with LJ
  • Numerous non paired RDT-LJ : false alarms or lack of observation ?
  • Objective quantification needed for “paired” and “missed” SW events

RDT-LJ ESWD Lighning Jump Validation (3/3) - RDT-LJ vs ESWD data

slide-19
SLIDE 19

Eumetsat Conference, Boston 2019

RDT-CW applicable to GOES16 (v2018.1 patch)

✔ Higher spatial resolution => more numerous cloud systems, more detailed ✔ Main channel => IR10.3 as window channel for detection/tracking/discrimination ✔ Higher time resolution => adapted calibrated discrimination for GOES16 / 10min & 15min ✔ Subjective validation on case study RDT vs current and following lightning activity

20190419

LGH [15h50-17h00Z] RDT 16h00Z + LGH [15h50-16h00Z] RDT confirmed by further GT

slide-20
SLIDE 20

Eumetsat Conference, Boston 2019

Overview

  • 1. CI: Convection initiation
  • 2. RDT: Rapidly Developing Thunderstorm

3.Conclusion

slide-21
SLIDE 21

Eumetsat Conference, Boston 2019

Conclusion – Way forward

CI

  • Validation

Good skill to detect new convective cells. Radar signature often appears prior to satellite CI signature. For GOES-16 validation, use of IR10.3 is recommended

  • Way forward

Ongoing work to define Ground Truth with radar-derived convective objects to validate the product (qualitative and quantitative validation) - Tuning improvement RDT

  • OTD

Achieve tuning of OTD and extended use of HRV and other channels (BTDs with O3 & CO2)

  • Lightning Jump

Quantitative evaluations (RDT-LJ vs HYDRE - vs ESWD 2017 database) - Assess LJ with other lightning data (GLM) Anticipation of hail events to investigate CDOP4

  • CI: Use of rapid scan and visible channels - Probabilities calibration improvement
  • RDT-CW: Top Cloud features, Ongoing calibration ...
  • CI as input to RDT
slide-22
SLIDE 22

Eumetsat Conference, Boston 2019

Thanks for your attention

slide-23
SLIDE 23

Eumetsat Conference, Boston 2019

Annex - CI v2018 – 20181002 case study Further work on GT depending on radar convective

  • bjects’ attributes – POD results

Warm radar convective cells Satellite CI has poor added value compared to radar convective

  • bjects

Lightning-paired radar convective cells Satellite CI has skill to forecast convection before radar convective objects appear or at the same time

POD (%)

Radar convective objects’ attributes to consider to build GT:

✔ Lifetime (5, 10, 30 minutes) ✔ Lightning pairing (Yes / No) ✔ Warm clouds (>= 500hPa or not)

slide-24
SLIDE 24

Eumetsat Conference, Boston 2019

Annex - CI v2018 – 20181002 case study Further work on GT depending on radar convective

  • bjects’ attributes – FAR results

FAR (%)

Whatever GT considered, very high FAR Lightning-paired radar convective cells: FAR slightly lower Radar convective objects’ attributes to consider to build GT:

✔

Lifetime (5, 10, 30 minutes)

✔

Lightning pairing (Yes / No)

✔

Warm clouds (>= 500hPa or not)

slide-25
SLIDE 25

Eumetsat Conference, Boston 2019

Annex - RDT-RSS-OT vs CHMI-OT – 20130729 case study

16h00Z(+3’) 16h05Z (+3’) 16h07Z 16h10Z 16h12Z HI MI ? MI ? MI FA ?

RDT-OTD CHMI-OT

HI

RDT-OTD

16h10Z (+3’) HI HI MI ? MI ? RDT-OTD:

  • Better performance

with RSS

  • Still some misses,

less FA Expert CHMI-OTs: less space & time variability of OTs for this example

slide-26
SLIDE 26

Eumetsat Conference, Boston 2019

20180809 case study:

  • [16h-18h] RDT (accumulated contours)
  • [16h-19h] HYDRE medium and large hail detection (accumulated pixels)

Accumulated RDT cells with Lightning Jump diagnosis

Annex - Lighning Jump Validation RDT-LJ vs HYDRE Hail detection

High correlation between HYDRE Hail detection and RDT track path with Lightning Jump occurrence

FA? MI MI

Accumulated pixels with medium/large hail detection