The mp4 animation above shows the development of Tropical Storm Sanvu out of the Monsoon Trough in the western Pacific ocean on 19-20 April 2023. The ‘Pacific Tropical Airmass RGB’ used in the animation is described in this blog post, and it gives more information about cloud-top features compared to the Airmass RGB (a similar mp4 animation for the airmass RGB is here). An interesting (and suggestive) aspect of the animation is the development of strong outflow to the north of the Monsoon Trough after 18 April that continues through the end of the animation.

MIMIC Total Precipitable Water fields — that incorporate wind information from the GFS — also show the development of a circulation as the system develops. Note that the storm is near the northern edge of the moist trough; the abundant dry air to its north might affect the future strength of the system.

Gridded NUCAPS fields (source) can also give views of the environment that contains the developing tropical cyclone. The three toggles below show the evolution from ca. 0300 UTC on 19 April through ca. 0230 UTC on 21 April 2023. Mid-level Total (850-700mb) Precipitable Water, Relative Humidity at 700 mb, and the 850-700 mb Lapse Rate all testify to the hostile environment (dry and stable) to the north of this developing system.

---

Diagnostics from the SSEC/CIMSS Tropical Weather website for the storm, below, show the system forecast to move towards a region where higher shear now exists.

SATCON diagnostics from CIMSS, above, show that an initial period of strengthening has leveled off.The forecast from the Joint Typhoon Warning Center takes Sanvu towards the southern Marianas islands as a system with weak winds (but likely abundant moisture), as shown below.

.

---

More information on this system is available at the webpages of the National Weather Service in Guam (link).

View only this post Read Less

There was a conditional outlook for severe storms in Oklahoma on April 19th, owing to uncertainty in the forcing for ascent and a stout warm-sector cap. Well, the cap broke, and mayhem ensued in central Oklahoma.

A break in the cirrus deck in southwest Oklahoma likely contributed to surface heating and enhanced mixing of the boundary layer, allowing the cap to break. ProbSevere LightningCast, an AI model that uses solely images of ABI data to predict next-hour lightning at any given point, was able to capture lightning initiation (Figure 1). Glaciating cloud tops are a prime signal of imminent lightning, and the one-minute scans from the GOES-16 mesoscale sector helped the model maximize lead time (Figure 2).

ProbSevere version 3, a set of machine-learning models that predicts probabilities of severe weather hazards in the near future, was able to track these storms as they became severe (Figure 3). These storms produced huge hail (up to 3″ in diameter), straight-line winds exceeding 80 mph, and deadly tornadoes (Figure 4). One aspect of ProbSevere’s automated guidance that forecasters have frequently reported is that during busy situations, ProbSevere helps them quickly triage which storms or threats to prioritize and investigate further. In this outbreak, most storms had at least severe thunderstorm warnings, while several were tornado-warned. Both ProbSevere v3 and LightningCast will be evaluated by forecasters at NOAA’s Hazardous Weather Testbed this spring.

Just a little further north in Kansas, LightningCast was able to correctly predict lightning initiation (with about 16 minutes of lead time) for the storm despite moderate overlapping cirrus clouds (Figure 5). Despite the obscuring ice clouds, LightningCast was able to discern elevated lightning potential with the aid of the visible red band (0.64-µm reflectance), snow-ice band (1.6-µm reflectance), and long-wave infrared bands (10.3-µm and 12.3-µm brightness temperatures) and the spatial patterns evident in the growing cumuliform clouds.

In fact, for this storm, the reflective bands were quite important. When they were removed from the model (a “data-denial” experiment), the probability of lightning was < 10% for this convection at 21:25 UTC, whereas the full LightningCast model produced a maximum probability of about 50% at this time (Figure 6). While we have found that LightningCast performs quite well at night (when the reflective bands are uniformly zero), moderate cirrus cover at night might be an instance where users should expect less lead time to lightning initiation.

Notice how difficult it is to visually pick out growing convection in this scene, with only the long-wave infrared bands plotted. LightningCast can only discern what humans are able to pick out in the satellite imagery, but LightningCast does it quickly, automatically, objectively, and without ceasing, aiding forecasters in their decision making.

View only this post Read Less

There are more than 7300 satellites in orbit above the Earth, and the mp4 animation above (from Rick Kohrs, SSEC; click here for an animated gif) shows their locations during about 3 hours on 6 April 2023. The points were computed using TLE data from https://celestrak.org. Some satellites are in geostationary orbit along the equator and show little movement; various polar orbiters and starlink orbits are apparent as well. That’s a lot of satellites!

View only this post Read Less

Severe weather on 15 April and 19 April affords another opportunity to compare Significant Tornado Parameter predictions from a 4-km WRF simulation that includes information from Polar Hyperspectral Soundings and a 3-km HRRR simulation that does not. Consider the example, above, at 2200 and 2300 UTC on 15 April and 0000 and 0100 UTC on 16 April. Each model shows the averages of 9 different forecasts valid at the given time. The HRRR output shows a maximum in STP over southeastern Missouri. The WRF output (driven by an initial field that includes information from Polar Hyperspectral Soundings) has a maximum in STP farther to the west where severe weather was occurring.

The example below is from the severe weather event on 19 April, again showing STP from the PHS-enhanced WRF simluation on the left and from the HRRR on the right. As on 15 April, the alignment of the severe weather events more closely matches the model predictions of STP when the model has PHS data are part of its input cycle.

---

Imagery for this blog post is courtesy Qi Zhang, CIMSS; PHS model runs are available here in real time. This model output will be evaluated at the Hazardous Weather Testbed in May and June.

---

The animation above (created at the CSPP Geosphere site) shows abundant — but thin — clouds over the southern USA between 1700 and 1900 UTC on 19 April 2023, i.e., before the PHS imagery shown above from 0000 to 0300 UTC on 20 April. The satellite retrievals at 1700 UTC on the 19th, shown below, included both infrared and microwave components over Oklahoma, as shown below (image courtesy Bill Smith Sr). That infrared retrievals could occur is testimony to the broken cloud field. How do the satellite retrievals alter the relative humidities? That is also shown below for 1700 and 2300 UTC (top and bottom) for 850 and 700 mb (left and right). Satellite data has altered the RH such that deeper moisture is farther to the west.

The image below, created by Bill Smith Sr, shows how the changed low-level moisture field from the Polar Hyperspectral data in this case leads to a more accurate simulation (the SAT/WRF model on the right, compared to the HRR model) of where severe weather occurs.

View only this post Read Less