Enhancing Smoke Detection with Red Visible imagery

August 18th, 2021 |
GOES-16 Visible Imagery (0.64  µm), 1501-1857 UTC, 4 August 2021 (Click to enlarge) 2-km resolution shown, using mean pixel values

Detection of smoke is a challenge in regions with clouds. The bright reflectance from clouds can make the subtler (and usually not so bright) smoke features difficult to view. This blog post briefly shows how visible imagery can be manipulated to facilitate smoke detection. The animation above shows visible imagery (GOES-16 Band 2 at 0.64  µm) at 2-km resolution. The sixteen half-km pixel visible reflectances within the larger pixel have been averaged to create the 2-km resolution image, and a square-root enhancement (from 0-100%) has been applied.

August 4 was a day with a smoke pall over much of the upper midwest, as suggested below by HRRR Smoke Model output below. How distinct is that pall in the imagery above, especially in regions of cumulus and cirrus clouds, such as over northeastern Minnesota? Can the visible imagery above be manipulated so that smoke features are enhanced?

HRRR Smoke forecast of Vertically Integrated Smoke at 1500 UTC on 4 August 2021, an 8-hour forecast from the 0700 UTC Model Run (Click to enlarge)

The animation below shows 2-km resolution data again (with a square-root enhancement applied), but instead of the mean reflectance value from the 16 1/2-km pixels used, the minimum reflectance of those 16 pixels is shown. The smoke in the air is more noticeable (and the clouds are less noticeable).

GOES-16 Visible Imagery (0.64  µm), 1501-1857 UTC, 4 August 2021 (Click to enlarge) 2-km resolution shown using minimum pixel values

You can also display the maximum pixel value of the 16 1/2-km pixels within the 2-km pixel footprint. That’s shown below (also using a square-root enhancement). This brings out the clouds at the expense of smoke detection.

GOES-16 Visible Imagery (0.64  µm), 1501-1857 UTC, 4 August 2021 (Click to enlarge) 2-km resolution shown using minimum pixel values

A PowerPoint presentation that contains the imagery above is here. Thanks to Andy Heidinger, NOAA, for these images.

The Day Night Band and Smoke

July 29th, 2021 |
Suomi NPP VIIRS Day Night Band Visible (0.70 µm) image, 0740 UTC on 29 July 2021 (Click to enlarge)

Smoke from fires (unless it is extraordinarily thick) is very challenging to detect at night in infrared imagery. When there is sufficent lunar illumination, however, as occurred on 29 July 2021 over Canada (the image above shows northwestern Ontario and southern Manitoba — Lakes Winnepeg, Winnepegosis and Manitoba are apparent, and the city of Winnipeg’s lights are apparent along the southern border of the image), the pall of smoke can be detected just as during the daytime using Day Night band visible imagery. In the image above, bright signals are showing light emitted by active fires, and streams of smoked from the fires are apparent. The toggle below of the Day Night band image and the VIIRS I04 3.74 µm imagery underscores that the bright spots are also very warm spots; that is: fires. Note that a smoke signal is not present at all in the infrared.

VIIRS Visible (0.7 µm) Day Night Band imagery and I04 Infrared (3.74 µm) imagery, 0740 UTC on 29 July 2021 (Click to enlarge)

Suomi NPP VIIRS Day Night Band Visible (0.70 µm) image, 0740 UTC on 29 July 2021 (Click to enlarge)

A zoomed-out Day Night band image shows the horizontal extent of the smoke pall that extends southward into Minnesota. This kind of night-time imagery can be useful to compare observations to model estimates of smoke coverage as created by the HRRR Smoke model. Compare the image above, for example, to the two-hour forecast of vertically integrated smoke valid at 1300 UTC on 29 July, shown below. There is good agreement in the coverage over northern Minnesota. What does that kind of smoke look like from the ground? Here’s a image from a webcam at Lake Bemidji (from this source). GOES Imagery, at bottom, from just before Noon Central time (and from the CSPP Geosphere site) also shows the very thick smoke over Minnesota.

HRRR Smoke forecast valid at 1300 UTC, a 2-hour forecast from an 1100 UTC/29 July 2021 initialization (Click to enlarge)
CSPP Geosphere True Color imagery, 1650 UTC on 29 July 2021 (Click to enlarge)

Use the Day Night band image (as available here, for example) to identify smoke plumes when lunar illumination is present.

Aerosol Optical Depth and surface visibility

July 22nd, 2021 |
GOES-16 Aerosol Optical Depth and GOES-16 Band 2 Visible (0.64 µm) imagery, 1401 UTC on 22 July 2021

The image above shows the Level 2 GOES-R product, Aerosol Optical Depth (AOD), a product created in clear skies, overlain with the GOES-16 Visible imagery from the same time. AOD measures the extinction of light via scattering and absorption by small particles in the atmosphere, and it can be used as a proxy for particles smaller than 2.5 µm in diameter (PM25). The red regions show the highest values. The plot below shows surface observations of ceilings (plotted to the left of the circles) and visibility (plotted below the circles) at the same time as the AOD image above. Is there a relationship?

Look at the string of lower visibilities stretching along the North Carolina/South Carolina border, extending westward to Tennessee and then northward into Illinois. This is the region where AOD exceeds about 0.4 — cyan in the enhancement used above. In this instance, AOD can be used to highlight regions where surface visibilities are most restricted by aerosols. (Some of these aerosols are likely from smoke. However, this product does not tell you what kind of aerosol is there, only that it is causing extinction).

Surface observations of ceilings and visibilities, 1401 UTC on 22 July 2021

The toggle below steps through the observations, AOD, and Visible imagery at 1401 UTC. Kudos to Frank Alsheimer, the Science and Operations Office (SOO) in Columbia SC, for alerting us to this case.

Surface observations of ceilings and visibilities, GOES-16 Aerosol Optical Depth and GOES-16 Band 2 Visible (0.64 µm) imagery, 1401 UTC on 22 July 2021

True-color imagery, below, (saved in this case from the CSPP Geosphere site, using this link) also shows the extent of the aerosol-rich air.

GOES-16 ‘True-Color’ imagery at 1401 UTC on 22 July 2021

The relationship between AOD values and surface visibility persisted on 23 July 2021, below.

GOES-16 Aerosol Optical Depth and GOES-16 Band 2 Visible (0.64 µm) imagery, 1201 UTC on 23 July 2021

Blowing dust in the Upper Midwest

May 25th, 2021 |

GOES-16 Split Window Difference images, with plots of wind barbs and gusts [click to play animation | MP4]

GOES-16 Split Window Difference images, with plots of wind barbs and gusts [click to play animation | MP4]

GOES-16 (GOES-East) Split Window Difference images (above) showed widespread strong winds across the Dakotas and northern Minnesota which were responsible for producing plumes of blowing dust (darker shades of gray) — most notably from eastern North Dakota into northwestern Minnesota — on 24 May 2021.

The corresponding GOES-16 Split Window Difference images with plots of surface visibility are shown below — at 23 UTC the visibility dropped to 4 miles at Grand Forks, North Dakota as a dense dust plume moved through that location (where southwesterly winds were gusting to 31 knots at that time).

GOES-16 Split Window Difference images, with plots of surface visibility [click to play animation | MP4]

GOES-16 Split Window Difference images, with plots of surface visibility [click to play animation | MP4]

GOES-16 True Color RGB images [click to play animation | MP4]

GOES-16 True Color RGB images [click to play animation | MP4]

GOES-16 True Color RGB images (above) and Dust RGB images (below) created using Geo2Grid highlighted the more dense plumes of blowing dust — the source region for the more prominent dust plumes appeared to be dry agricultural fields in southeastern North Dakota that had received very little rainfall during the preceding week.

GOES-16 Dust RGB images [click to play animation | MP4]

Ground-based lidar data from Grand Forks indicated that the dust was lofted to altitudes of around 10,000 feet.


===== 25 May Update =====

GOES-16 Dust RGB images, with and without plots of surface reports [click to play animation | MP4]

GOES-16 Dust RGB images, with and without plots of surface reports [click to play animation | MP4]

Strong winds persisted across that same region on 25 May — and GOES-16 Dust RGB images (above) again displayed the subtle signature of blowing dust (light shades of pink/magenta) along the leading edge of cloudiness that was moving eastward into northwestern Minnesota.

GOES-16 True Color RGB images (below) once again showed the hazy signature of blowing dust.

GOES-16 True Color RGB images [click to play animation | MP4]

GOES-16 True Color RGB images [click to play animation | MP4]