This website works best with a newer web browser such as Chrome, Firefox, Safari or Microsoft Edge. Internet Explorer is not supported by this website.

The impact of satellite inputs in ProbSevere v3

On March 19, a potent shortwave trough forced the development of a strong surface cyclone and moisture return into central Illinois, with ample surface theta-e and deep-layer shear to sustain surface-based convection.In the animation below, you can see the quickly developing convection and associated ProbSevere v3 contours evolve. The storm... Read More

On March 19, a potent shortwave trough forced the development of a strong surface cyclone and moisture return into central Illinois, with ample surface theta-e and deep-layer shear to sustain surface-based convection.

NAM 3km estimate of 2 m Theta-E, 10 m wind, and MSLP.

In the animation below, you can see the quickly developing convection and associated ProbSevere v3 contours evolve. The storm that affected Canton, Illinois produced a damaging wind report at 19Z and a tornado report first at 19:30Z.

Early on the storm’s development, the satellite growth rate in ProbSevere was quite impactful. At 18:18Z, the moderate growth rate was the 5th most-important predictor.

  1. Lapse rate 0-3 km (8.9 C/km)
  2. Mean wind 1-3 km AGL (43 kt)
  3. Effective bulk shear (52 kt)
  4. Total lightning flash rate (6 fl/min)
  5. Satellite growth rate (2%/min)

Along with all of the other predictors, this generated a probability of severe of 30%. As an experiment, when we artificially made the satellite growth rate 0, the probability fell to 14%. The satellite growth rate has the most impact on storms during the development phase of convection. Once convection is more mature, the deep-learning-based IntenseStormNet predictor (which uses images of ABI and GLM fields as inputs) is the more important satellite feature in the ProbSevere v3 models.

Using the Python SHAP library, we created a “decision” plot to visually see how different predictors in the model affected the final prediction. The bar on the top and bottom is in “log-odds-space”, which allows us to see meaningful deflections of the predictors. That is, the deflections are all proportional to their importance on the final prediction. Rightward deflections increase the probability of severe whereas leftward deflections decrease the final probability. The satellite growth rate was about as important as the flash rate for this storm at this time.

A SHAP “decision plot” for the ProbSevere v3 model for the storm at 18:18Z

Shortly after 18:18Z, a strong lightning jump also boosted the probability of severe to about 40%. About 35 minutes later, damaging wind was report. Interestingly, the moderate satellite growth rate did not boot the ProbSevere v2 probability, probability because the meager radar and lightning predictors were squashing the probability overall. However, the ProbSevere v3 models are gradient-boosted decision trees, and appear to utilize the satellite data more skillfully (as well as other inputs) than the previous version.

Time series of ProbSevere v2, ProbSevere v3, and total lightning for the development stage of this storm.

The probability of tornado began ramping up in this storm around 19:10Z, hitting 56% at the time of the first tornado report (19:30Z), which is a very high value for ProbTor v3. In the AWIPS display below, a user’s mouse “hovering” over the ProbSevere object produces the pop-up text. This text provide forecasters the exact probabilities and select predictor values. A time series of the probability values opens up when a user double-clicks on the storm object. These visualization features help forecasters quickly interrogate severe or tornadic threats, which is particularly important during busy severe-weather situations.

ProbSevere contours, MRMS Merged Reflectivity, and NWS severe weather warnings.

View only this post Read Less

von Kármán vortex street in the Bering Sea

10-minute Full Disk scan GOES-18 (GOES-West) True Color RGB images created using Geo2Grid (above) showed a von Kármán vortex street — created by north-northwesterly winds interacting with the western tip of Nelson Island, Alaska — propagating south across the Bering Sea on 20 March 2025. In addition, through breaks in cloud cover the tidal... Read More

GOES-18 True Color RGB images, from 1730 UTC on 20 March to 0440 UTC on 21 March [click to play animated GIF | MP4]

10-minute Full Disk scan GOES-18 (GOES-West) True Color RGB images created using Geo2Grid (above) showed a von Kármán vortex street — created by north-northwesterly winds interacting with the western tip of Nelson Island, Alaska — propagating south across the Bering Sea on 20 March 2025. In addition, through breaks in cloud cover the tidal ebb and flow of sea ice was apparent.

The vortices were also evident in Suomi-NPP Visible images valid at 2202 UTC and 2342 UTC (below).

Suomi-NPP VIIRS Visible (0.64 µm) images, valid at 2202 UTC and 2342 UTC on 20 March [click to enlarge]

In a toggle between the 2202 UTC Suomi-NPP VIIRS Visible image and Topography (below), an arc of slightly elevated terrain along the far western tip of Nelson Island (abruptly rising to around 900 ft, darker shades of tan) likely perturbed the northerly flow enough to initiate formation of the von Kármán vortices.

Suomi-NPP VIIRS Visible (0.64 µm) image valid at 2202 UTC + Topography [click to enlarge]

Special thanks to Jason Ahsenmacher (NWS Fairbanks) for bringing this interesting feature to my attention.

View only this post Read Less

Day Cloud Phase Distinction and Snow Bands

The Slider above compares the Day Cloud Phase Distinction RGB over the central United States with the derived Land Surface Temperature (LST), an hourly Level-2 product. Two snow bands over the central United States are a characteristic bright green in the RGB because there are minimal contributions there in the red and blue.... Read More

The Slider above compares the Day Cloud Phase Distinction RGB over the central United States with the derived Land Surface Temperature (LST), an hourly Level-2 product. Two snow bands over the central United States are a characteristic bright green in the RGB because there are minimal contributions there in the red and blue. The frozen lakes over northern Minnesota share the bright green color in the RGB. Land Surface Temperature over the snow (around 30oF in Iowa) is very different from the bare ground on either side (low/mid-60s F). LST values over the snow in Kansas are above freezing, so you might expect the snow there to melt more quickly. (This toggle compares 1701 and 2016 UTC imagery, and the erosion of snow over Kansas, Nebraska and Illinois is obvious).

Surface observations plotted over the Land Surface Temperature plot, below, show that temperatures at thermometer level (typically about 1.5 m above the ground) are far cooler than the surface skin temperature under the strong March Sun on the Equinox.

GOES-16 Day Cloud Phase Distinction and Land Surface Temperature in clear air, 1801 UTC on 20 March 2025 (Click to enlarge)

View only this post Read Less

Another dusty storm in the midwest

An active weather pattern has produced another strong storm over the Great Plains of the United States on 18-19 March, and another dust storm. Dust RGB Imagery, above, shows the evolution of the dust plume (in magenta/pink in the RGB) from Mexico into the western Great Lakes. The pink occurs because of a strong... Read More

GOES-East Dust RGB imagery, 1801 UTC 18 March 2025 – 1346 UTC 19 March 2025 (Click to enlarge)

An active weather pattern has produced another strong storm over the Great Plains of the United States on 18-19 March, and another dust storm. Dust RGB Imagery, above, shows the evolution of the dust plume (in magenta/pink in the RGB) from Mexico into the western Great Lakes. The pink occurs because of a strong contribution from the Split Window Difference (10.3 µm – 12.3 µm) imagery (Quick Guide); an animation of that is here, the dust is brown in the animation. The Split Window Difference is also the Red component of the Night Microphysics RGB (Quick Guide), so that RGB can also give a view of Dust Plumes at night (as shown in this blog post).

VIIRS has a Day Night Band that can be used to detect dust plumes (or smoke plumes) at night, given sufficient lunar illumination. In the 0745 UTC Day Night Band image below, toggled with the GOES-based Dust RGB (one could also compute a Dust RGB image from VIIRS data!), the dust cloud is a apparent from central Texas northward through eastern Oklahoma and eastern Kansas.

VIIRS Day Night Band visible (0.7 µm) imagery and GOES-16 Dust RGB, 0745 UTC on 19 March 2025 (click to enlarge)

Geosphere imagery, below, from 18 March, shows the beginnings of the dust plume near El Paso TX. GOES-16 Aerosol Optical Depth is also plotted in clear skies. El Paso had 1/2-mile visibility due to dust and wind gusts exceeding 50 mph during the afternoon on 18 March.

True Color Imagery and GOES-16 Aerosol Optical Depth, 18 March 2025, 1501 – 1906 UTC.

________________

1-minute GOES-18 Visible images with 30-minute plots of Peak Wind Gusts (yellow/red) and hourly plots of Ceiling/Visibility (cyan), from 1501 UTC on 18 March to 0030 UTC on 19 March (courtesy Scott Bachmeier, CIMSS) [click to play MP4 animation]

1-minute Mesoscale Domain Sector GOES-18 (GOES-West) Visible images (above) included plots of 30-minute Peak Wind Gusts and hourly Ceiling/Visibility — which showed Peak Wind Gusts as high as 65 kts (75 mph) at METAR sites Roswell, New Mexico (KROW) and El Paso, Texas (KELP), although there were wind gusts to 101 mph at higher elevations (these strong winds in tandem with severe to exceptional drought conditions were conducive to create such a pronounced blowing dust event). The surface visibility was reduced to 1/4 mile (or even near zero) at several sites from El Paso to southern New Mexico.

10-minute Full Disk scan True Color RGB and nighttime Dust RGB images from GOES-18 (GOES-West), GOES-19 (Preliminary/Non-operational) and GOES-16 (GOES-East) created using Geo2Grid (below) displayed distinct signatures of the dense blowing dust that originated over parts of northern Mexico and southern New Mexico — and was subsequently transported northeastward across parts of Texas, Oklahoma and Kansas during the 15-hour period shown. There was also a notable plume of blowing dust/sand whose source was White Sands National Park in southern New Mexico. In addition, of interest was a small circular pocket of dust that became entrained into the center of low pressure that was located near the Colorado/Kansas/Oklahoma border at 0600 UTC.

True Color RGB and nighttime Dust RGB images from GOES-18 (left), GOES-19 (Preliminary/Non-operational, center) and GOES-16 (right), from 1500 UTC on 18 March to 0600 UTC on 19 March (courtesy Scott Bachmeier, CIMSS) [click to play animated GIF | MP4]

View only this post Read Less