The probability of “intense convection” using geostationary satellite data

September 27th, 2019 |

Researchers from NOAA and UW-CIMSS have developed an experimental model that predicts the “probability of intense convection” inferred from GOES ABI and GLM fields. The model is a convolutional neural network, which carries the assumption that the inputs are images and have spatial context. It is a great tool for image classification.

GOES-16 ABI CH02 reflectance (a visible channel), ABI CH13 brightness temperature (an infrared window channel [IR]), and GLM flash extent density (FED; generated using glmtools),  were used as inputs to the model. The model learned important features that have been traditionally difficult or expensive to code into an algorithm, such as pronounced overshooting tops (OTs), enhanced-V features, thermal couplets, above-anvil cirrus plumes (AACPs), strong brightness temperature gradients, and texture from visible reflectance.

It is hoped that such a model may be able to one day:

  • provide earlier notice of developing or decaying intense convection
  • provide guidance in regions with no weather radars
  • provide a quantitative way to leverage 1-min mesoscale scans
  • ultimately improve the accuracy and lead time of severe weather warnings

The model is very experimental and is not yet running in real-time. The remainder of this post catalogues some examples of the deployed model on select scenes.

The movies below use as a background the CH02-CH13 “sandwich” product, whereby cloud-top 11-µm brightness temperature and 0.64-µm reflectance can be seen in tandem. This generally helps observers see how changes in storm-top structure correlate with changes in 11-µm brightness temperature. A grid of “probability of intense convection” was generated for each scene with a moving 32×32 pixel window (each pixel = ~2 km), with the model generating one probability for each window. These probabilities were then contoured with the 25%, 50%, and 90% contours as blue, cyan, and magenta. Preliminary severe local storm reports from the SPC rough log are also plotted as circles.

The example below shows that the model handled two separate severe wind threats in Missouri, identifying cold cloud top regions in the IR that also looked “bubbly” from the visible channel. As the sun was setting, a cold front lit up with very intense convection from Oklahoma through Missouri. Again, the model did a decent job highlighting the strongest areas of convection which correlated well with severe local storm reports. It should also be noted that the model does not seem to have significantly degraded output when the visible channel is missing (after sunset).

 

The next example is at a higher satellite viewing angle in western Nebraska, western South Dakota, and eastern Wyoming. The model again does a good job highlighting the strongest areas of storms. It should be noted that not every identified region has severe reports and not every severe report has a probability of intense convection ? 25%, but that there is generally good correspondence between reports and the model probabilities nonetheless.

 

This example is from the Southeast U.S. in more of a low-shear “microburst” environment instead of a high-shear “supercell” environment. You can see that instead of predicting high probabilities for all of the convective storms, the model exhibits the highest probabilities for the storm clusters that at least subjectively look the strongest.

 

This next example from the Central Plains demonstrates the ability to discern decaying convection, as the first storm moves into Missouri and then quickly diminishes in appearance and in probability. It also demonstrates the model’s ability to pick out multiple threat areas within a large cloud mass at night.

 

This is an example using mesoscale scans. Despite not being trained with 1-min data, the model predictions still look very fluid and reasonable. This could be an excellent way for scientists and forecasters to leverage 1-min observations in a quantitative manner.

 

Another example using GOES-East 1-min mesoscale scans. The model generally picks out the strongest portions of a MCS in Illinois and Indiana.

 

At a very high viewing angle, the model predicts probabilities of ?90% for a storm in Arizona. The storm did not generate severe reports, but was warned on by the NWS multiple times.

 

The model is deployed during an early autumn severe weather outbreak.

Severe thunderstorms in the Upper Midwest

September 24th, 2019 |

GOES-16 “Clean” Infrared Window (10.35 µm) images, with SPC Storm Reports plotted in cyan [click to play animation | MP4]

GOES-16 “Clean” Infrared Window (10.35 µm) images, with SPC Storm Reports plotted in cyan [click to play animation | MP4]

1-minute Mesoscale Domain Sector GOES-16 (GOES-East) “Clean” Infrared Window (10.35 µm) images (above) showed the development of severe thunderstorms across parts of the Upper Midwest on 24 September 2019 — these storms produced hail as large as 2.5 inches in diameter in Nebraska, a wind gust to 80 mph in Minnesota and an EF3-rated tornado in Wisconsin (SPC Storm Reports | NWS Twin Cities | NWS La Crosse).

The corresponding 1-minute GOES-16 “Red” Visible (0.64 µm) images leading up to sunset are shown below.

GOES-16 “Red” Visible (0.64 µm) images, with SPC Storm Reports plotted in red [click to play animation | MP4]

GOES-16 “Red” Visible (0.64 µm) images, with SPC Storm Reports plotted in red [click to play animation | MP4]

A “probability of intense convection” model was run for this particular event (below).

“Probability of intense convection” model [click to play MP4 animation]

Severe thunderstorms in Arizona

September 23rd, 2019 |

GOES-17 “Red” Visible (0.64 µm) and “Clean” Infrared Window (10.35 µm) images, with surface reports plotted in cyan [click to play animation | MP4]

GOES-17 “Red” Visible (0.64 µm) and “Clean” Infrared Window (10.35 µm) images, with surface reports plotted in cyan [click to play animation | MP4]

1-minute Mesoscale Domain Sector GOES-17 (GOES-West) “Red” Visible (0.64 µm) and “Clean” Infrared Window (10.35 µm) images (above) showed the development of severe thunderstorms over southern/central Arizona from 1600-1900 UTC on 23 September 2019. The far western storm exhibited a well-defined Above-Anvil Cirrus Plume (AACP) that extended northeastward from the cold overshooting top (whose coldest infrared brightness temperature was -74ºC); note that the AACP feature appeared colder (shades of yellow to orange) on the Infrared images (for example, at 1817 UTC).

As the western storm began to weaken somewhat, a new storm just to the east (located about 20-30 miles north-northeast of the Phoenix metro area) began to intensify, prompting the issuance of a Tornado Warning at 1914 UTC (the last tornado warning issued by NWS Phoenix was 21 January 2010) — a brief EF0 tornado was documented (NWS Phoenix summary).

GOES-17 “Clean” Infrared Window (10.35 µm) images, with surface reports plotted in cyan [click to play animation | MP4]

GOES-17 “Clean” Infrared Window (10.35 µm) images, with surface reports plotted in cyan [click to play animation | MP4]

Much of the moisture helping to fuel the development of this severe convection was from the remnants of Tropical Storm Lorena in the East Pacific Ocean — the northward transport of this moisture could be seen using the hourly MIMIC Total Precipitable Water product (below).

MIMIC Total Precipitable Water product [click to play animation | MP4]

MIMIC Total Precipitable Water product [click to play animation | MP4]


 

GOES-17 ABI Band 13 (10.35 µm) Clean Window Imagery and Derived Convective Available Potential Energy, 1501 – 1856 UTC on 23 September 2019 (Click to animate)

 

Stability parameters from GOES-16 showed that the reigon of thunderstorm development was just east of a strong gradient in Convective Available Potential Energy.  The animation above shows the GOES-17 Clean Window;  in regions of clear sky, the baseline Derived Stability Index CAPE is shown.  CAPE values are zero over much of California (except for the southeasternmost corner) but they increase rapidly over Arizona to values approaching 1000 J/kg.

On 23 September, skies were clear enough that an instability signal was obvious in the clear-sky baseline CAPE. An ‘All-Sky’ product has been developed that can be used on days with more widespread cloudiness; it is available at this link. Values of All-Sky CAPE at 1156 and 1556 UTC on 23 September are shown below, and they also show a sharp gradient in the instability, and the link down to moisture from Lorena’s remants.

‘All-Sky’ values of Convective Available Potential Energy (CAPE) at 1156 and 1556 UTC on 23 September 2019 (Click to enlarge)

NOAA/CIMSS ProbSevere is a product designed to indicate the likelihood that a given object will produce severe weather within the next 60 minutes. An animation of the product at 5-minute intervals, below, shows that the right-moving radar cell (also associated, as noted above, with an AACP) that developed over far southwestern Arizona (becoming a warned storm at 1647 UTC) was very likely to produce severe weather.

NOAA/CIMSS ProbSevere from 16:30 UTC to 18:00 UTC. Contours surrounding radar objects are color-coded such that pink/magenta contours are the highest probability.  Warning polygons (yellow for severe thunderstorm) are also shown (Click to enlarge)

Parameters that are used to determine the probability can be revealed at the ProbSevere site by mousing over the colored object contours.  The values for the warned storm over SW Arizona are shown below at 1650 UTC, 3 minutes after the warning was issued.  This image shows the 1710 UTC readout with the highest ProbWind value (76%); this image shows the 1725 UTC readout with the highest ‘ProbHail’ value (99%); ProbTor values on this day were not exceptionally large — for the later tornado-warned storm farther east, they were 28% at 1915 UTC and 30% at 1920 UTC.

NOAA/CIMSS ProbSevere display from 1650 UTC on 23 September 2019; parameters used in the probability computation, and Severe Thunderstorm Warning polygon parameters are also shown (Click to enlarge)

CIMSS is developing a machine-learning tool that combines ABI and GLM imagery (that is, only satellite data) to identify regions where supercellular thunderstorms capable of producing severe weather might be occurring. An mp4 animation for this event (courtesy John Cintineo, CIMSS) is shown below.  (This experimental product was also shown in this blog post)

Severe thunderstorms in Wyoming, Nebraska and South Dakota

September 10th, 2019 |

GOES-16

GOES-16 “Red” Visible (0.64 µm) images, with SPC Storm Reports plotted in red [click to play animation | MP4]

1-minute Mesoscale Domain Sector GOES-16 “Red” Visible (0.64 µm) images (above) showed  the development of thunderstorms that produced large hail, tornadoes and damaging winds (SPC Storm Reports) across eastern Wyoming, northern Nebraska and southern South Dakota on 10 September 2019. Note that many of the storms exhibited Above-Anvil Cirrus Plumes. Pulsing overshooting tops reached -80ºC and colder (violet pixels) just east of Valentine, Nebraska (KVTN) from 0001-0004 UTC (0002 UTC image) — and a few minutes following their collapse, a wind gust of 60 mph was reported in that general vicinity.

The corresponding GOES-16 “Clean” Infrared Window (10.35 µm) images are shown below.

GOES-16 "Clean" Infrared Window (10.35 µm) images, with SPC Storm Reports plotted in cyan [click to play animation | MP4]

GOES-16 “Clean” Infrared Window (10.35 µm) images, with SPC Storm Reports plotted in cyan [click to play animation | MP4]

The animation shown below is from an experimental product at CIMSS/SSEC, whereby the contours were produced using a ‘deep learning’ artificial intelligence model that was trained on ABI imagery and GLM gridded products to generate the ‘probability of supercell-like features inferred from satellites’, or more concisely, the ‘probability of supercell’. Note that the model does a decent job of identifying active portions of the storms (e.g., persistent Overshooting Tops), which correspond well with severe weather reports.

GOES-16 Visible/Infrared Sandwich RGB and

GOES-16 Visible/Infrared Sandwich RGB and “Clean” Infrared Window (10.35 µm) images, with “probability of supercell” contours and SPC Storm Reports (courtesy of John Cintineo, CIMSS) [click to play MP4 animation]

During the subsequent nighttime hours, GOES-16 Infrared images (below) showed a convective cluster which produced 3 EF-2 tornadoes and damaging winds in and around Sioux Falls, South Dakota (NWS summary). Note that pulsing overshooting tops west of Sioux Falls (KFSD) exhibited infrared brightness temperatures of -80ºC and colder (violet pixels) from 0402-0406 UTC (0404 UTC image), which was about 20 minutes prior to the first tornado reports.

GOES-16 "Clean" Infrared Window (10.35 µm) images, with SPC Storm Reports plotted in cyan [click to play animation | MP4]

GOES-16 “Clean” Infrared Window (10.35 µm) images, with SPC Storm Reports plotted in cyan [click to play animation | MP4]