1 of 16

1

NOAA’s ProbSevere LightningCast: Progress and Updates toward Transition-to-Operations

John Cintineo

University of Wisconsin – Madison (UW)

Cooperative Institute for Meteorological Satellite Studies (CIMSS)

Mike Pavolonis (NOAA/NESDIS)

Scott Lindstrom (UW-CIMSS)

Justin Sieglaff (UW-CIMSS)

​

104th AMS Annual Meeting, Baltimore MD

14th Conference on Research to Operations

1 February 2024

2 of 16

What is LightningCast?

2

ProbSevere

​

  • A portfolio of machine-learning models for short-term prediction of convective hazards
    • ProbSevere v2
      • operational at NCO since 2020
    • ProbSevere v3
      • to replace PSv2 in 2024/25?
    • IntenseStormNet
      • satellite-only “intense” convection model
      • used in PSv3
    • LightningCast

​

  • LightningCast
    • “Deep-learning” (AI) model
    • Trained to predict next-hour lightning using only GOES-R ABI bands
      • GOES-R ABI provides a rich, image-based dataset to leverage AI methods
      • Coverage where you don’t have radar
      • Information before radar!

3 of 16

What is LightningCast?

3

  • LightningCast is a deep-learning model that uses images of ABI data to predict next-hour lightning occurrence

​

  • Target/truth: 60-min accumulations of GLM flash-extent density

​

  • Inputs:
    • 0.64-µm reflectance (CH02)
    • 1.6-µm reflectance (CH05)
    • 10.3-µm BT (CH13)
    • 12.3-µm BT (CH15)

​

  • Output: probability of ≥ 1 GLM flash in the next 60 min at every pixel.

​

  • Trained on GOES-16 CONUS sector

​

  • Often provides ≥ 20 min of actionable lead time to lightning initiation

day cloud phase distinction RGB

Published in WAF (July 2022): https://journals.ametsoc.org/view/journals/wefo/37/7/WAF-D-22-0019.1.xml

4 of 16

What is LightningCast?

4

5 of 16

What is LightningCast?

5

40 min

45 min

35 min

20 min

15 min

20 min

25 min

25 min

10 min

10 min

40 min

35 min

25 min

Lead time to first GLM flash, measured from 25% contour (cyan)

​

  • Varies from 10 min to 45 min

​

  • Few false alarms

6 of 16

Why LightningCast?

6

  • NWS sees value in observations-based short-term lightning prediction
    • Objective
    • Automatic
    • Probabilistic
    • General convective initiation guidance
      • situational awareness
    • Advanced guidance for airports, mariners, stadiums, festivals, etc. (IDSS)
  • Directly addresses public safety

​

7 of 16

Why LightningCast?

7

  • Survey results:

​

    • 50-60% of forecaster responses (1 per forecaster per day) said LC provides ≥ 20 min of lead time to first flash

​

    • 60-70% of forecaster responses said LC provides ≥ 20 min of lead time to thunderstorms advecting into an area

​

    • 80-90% of forecaster responses said LC guidance is generally well calibrated

​

    • 100% said they would use LC regularly or all the time

​

> 75% said IDSS support, CI/maintenance SA, severe weather SA, and TAF amendments

LightningCast evaluated at 2022 and 2023 Hazardous Weather Testbed (HWT)

8 of 16

LightningCast T2O details

8

  • Where/Who
    • NOAA NESDIS OSPO
      • Office of Satellite and Product Operations
    • ASSISTT - Algorithm Scientific Software Integration and System Transition Team
      • Integration of code to NCCF (NESDIS Common Cloud Framework)
      • Ensure security of code
      • Documentation
      • Maintenance structures
  • LightningCast sectors
    • GOES-East
      • CONUS (5 min)
      • MESO1 and MESO2 (1 min)
      • OPC/TAFB offshore zones (10 min)
    • GOES-West
      • PACUS (5 min)
      • MESO1 and MESO2 (1 min)
      • Alaska (10 min)
      • American Samoa (10 min)
    • Himawari
      • Guam

​

  • Dashboards (via internet)
    • TAF airports
    • D1 college football stadiums
    • Phase 2…
    • Latency
      • 3-30 seconds
      • After satellite completion time
    • File sizes
      • up to 2 MB compressed
  • What (Phase 1)
    • NetCDF4 (for AWIPS)
      • parallax-corrected and

uncorrected

    • GeoJSON
      • parallax-corrected and

uncorrected

    • GRLevel2 placefiles
      • parallax-corrected only
      • CONUS/PACUS
      • Mesos
  • When
    • 30 September 2024

9 of 16

LightningCast T2O details (Phase 2)

9

https://cimss.ssec.wisc.edu/severe_conv/pltg.html

10 of 16

LightningCast T2O details (Phase 2)

10

Toggle on airport layer

https://cimss.ssec.wisc.edu/severe_conv/pltg.html

11 of 16

LightningCast T2O details (Phase 2)

11

Click airport and click link

https://cimss.ssec.wisc.edu/severe_conv/pltg.html

12 of 16

LightningCast T2O details (Phase 2)

12

  • Trends in probability of lightning
  • Trends in GLM-observed lightning
  • Interactive

13 of 16

LightningCast T2O – CSPP Geo

13

CSPP Geo software to create LightningCast

  • Community Satellite Processing Package for Geostationary Data, a standalone package (it creates the environments needed to work)
  • Creates imagery with (or without) GLM overlays

​

  • Requires access to GOES-R ABI Level 1b netcdf files (or Himawari HSD files) and, optionally, to GLMF files (for the FED overlays)

​

​

  • Beta release for the software in 2024 Q2.

​

14 of 16

LightningCast T2O

14

Summary

​

  • LightningCast provides objective, quantitative guidance on lightning potential using satellite observations

​

  • A “force” multiplier

​

  • Lightning dashboards provide “quick-look” at lightning potential at a given location.

​

  • Phase 1 operational by 30 September 2024
    • Phase 2 operational by 30 September 2025?

​

  • LightningCast in CSPP-Geo
    • Beta release ~April 2024

​

15 of 16

ProbSevere and last week’s tornado

15

Max ProbTor ~40%

16 of 16

ProbSevere and last week’s tornado

16

Leading contributing predictors:

  • low-level azshear
  • 1-3 km wind speed (51 kt)
  • mid-level azshear
  • lightning flash rate

Leading detracting predictors:

  • satellite-based “intense convection probability”