Multispectral Imaging Information

This blog post will explore an experimental RGB “sandwich” product that has shown application in monitoring convection, particularly during the early stages of the convective lifecycle (aka convective initiation). The intended purpose of this revolves around identifying, interrogating, and monitoring these features of interest:

1. Moisture gradients as a source of focused ascent.
2. Early stages of convective lifecycle (i.e. convective initiation).
3. Cloud top structure and movement to glean additional info like stability, shear, and overall convective behavior.

An experimental RGB dubbed “Moisture Gradient Convective RGB” was developed using bands within the Advanced Baseline Imagery (ABI) onboard the current operational GOES satellites.

The Split Window Difference product (10.35 – 12.3 um) has shown utility in identifying low level moisture gradients, and has been shared numerous times on this blog (Just a few examples: here, here, and here). The 1.37 um Cirrus Band as well as the 1.61 um Snow/Ice Band can be useful in monitoring the early stages of the convective lifecycle [1]. And when looking to interrogate cloud top structure and movement, good ole trusty 0.64 um Visible Band is a clear winner.

The Moisture Gradient Convective RGB recipe* is as follows:

• R: 1.37 um Cirrus Band
  • Min – Max: 0 – 10
  • Gamma: 1.90
• G: Split Window Difference product (10.3-12.3 um)
  • Min – Max: 0 – (-6)
  • Gamma: 1.00
• B: 1.61 um Snow/Ice Band
  • Min – Max: 0 – 70
  • Gamma: 0.50

*Note on the RGB recipe: this recipe has not been tested for optimization across different regions and environments. Ranges and gamma values will likely need to be adjusted depending on the moisture and thermal environment.

Here is an example of this RGB:

But wait, where’s the 0.64 um Visible Band to help in interrogating cloud top structure and movement??

This is where the “Sandwich” approach is handy. We simply overlay a semi-transparent 0.64 um Visible Band on top of the RGB to add such desired information:

The RGB Sandwich requires daylight for this application, and is most useful with limited to no higher clouds masking information on moisture gradients and growing, infant convection.

In areas absent of masking cloud cover, greener colors represent relatively greater low level moisture content compared to areas that are bluer. Within clouds, red coloring represents clouds that have grown vertically high enough to extend into a dry air mass aloft while exhibiting glaciation (could either be optically thick cirrus and/or glaciated tops of robust convection), whereas purple-blue clouds represent shallower liquid phased clouds. Clouds that are more yellow-orange represent optically thin cirrus clouds.

Application Example: Southern Plains Severe Convection March 2, 2023

Robust convection was anticipated within central Texas March 2, 2023, initiated along a merging dryline/Pacific front. Identifying these initiating boundaries and how/when convection develops along these boundaries would prove useful in operational forecasting and decision support services.

Near the southern High Plains and Trans Pecos region of Texas, congestus cumulus started to develop between 1700 – 1730 UTC along one of these boundaries: a dryline, particularly near the Midland and Big Springs area. The useful Day Cloud Phase Distinction RGB has a proven track record in monitoring for convective initiation (as seen above), and can show the infant cumulus just starting to sprout.

Supplementing this imagery with automated surface observations and objective analysis can be useful in identifying boundaries suspect of initiating convection. However, there still is some form of interpolation needed when these boundaries don’t have attached clouds to help identify them. This is where adding information on moisture gradients can come in handy.

Comparing the Day Cloud Phase Distinction RGB to the Moisture Gradient Convective RGB Sandwich, can the exact location of the initiating boundary/ies location be identified in between areas of growing cumulus?

The location of the dryline is apparent at this time (orange annotated line), including other boundaries where low level moisture is converging (black annotated lines).

Moving forward with time, we can see cumulus agitate and grow right along this boundaries, particularly near Big Springs where the dryline “bulged” forward and where moisture convergent boundaries intersected. Additionally, one can notice a collection of mesoscale, convergent moisture bands just ahead of the dryline, acting as additional local sources of initiation and moisture pooling.

Other Considerations

While the Moisture Gradient Convective RGB Sandwich can be very applicable in dryline situations, it can be useful in any other convective scenarios where moisture gradients are suspected to be a source of initiation so long as there aren’t any masking clouds to cover moisture information given by the Split Window Difference product. Of course all limitations of any of the RGB components must be considered when utilizing this multispectral imaging.

Important caveats to keep in mind when using the Split Window Difference (SWD) to analyze low-level moisture features include:1) must have clear sky conditions, 2) the SWD values depend on low level moisture content AND low-level lapse rates (see basic graphic below). For example, in the presence of moisture, a steeper positive ll lapse rate (temperature decreasing with height) will result in greater SWD values compared to weaker positive ll lapse rates with the same amount of moisture. A neutral or negative lapse rate (temperature increasing with height) in the presence of low-level moisture will result in SWD values lower or opposite of those with a positive lapse rate. In short, interpretation of this RGB as outlined in this blog post requires a positive low-level lapse rate.

Keep in mind this is an experimental multispectral imaging technique. At the time of this post’s creation, it hasn’t undergone extensive research and application to prove its usefulness and find common pitfalls. Additionally, it may not be optimized for different regions, nor at different times of the year which may have implications on overall moisture content. For example, the SWD range may need to be adjusted during the peak Spring/Summer convective season to account for greater (negative) SWD values.

If you happen to find any interesting or notable uses of the Moisture Gradient Convective RGB Sandwich, as well as any pitfalls, please comment or reach out to either of the authors.

Carl Jones (NWS Grand Forks), with input from Bill Line (NESDIS/STAR)

Starting the day of June 16, 2021, a plume of elevated smoke originating from the Robertson Draw fire in Montana began moving over the Dakotas. Day shift forecasters at the Grand Forks, ND, NWS office were briefed of this smoke moving towards their CWA, possibly impacting daytime temperatures and in turn, potential convection later in the afternoon along a cold fropa. With the expectation of smoke hindering the amount of insolation during peak heating hours, forecasters set out to track this smoke using GOES imagery, namely visible imagery from ABI’s channels 1 and 2, CIRA’s Geocolor RGB, as well as the GOES Aerosol Optical Depth product.

While plume guidance was available from the HRRR-smoke model, close observation of the elevated smoke using satellite imagery was performed to truly define the spatial bounds in an effort to set an observed trajectory. Given the expected absence of large changes within the mid level environment, a persistence forecast using the smoke’s observed trajectory was performed using the Distance Speed and Time of Arrival tools within AWIPS. Using both persistence trajectory forecast and model guidance, an assessment could be made to highlight the geographical areas that would most likely be under veil of elevated smoke during peak heating hours (roughly thought to be between 15 – 00 UTC during this time of year), thus indicating areas that have best chance of seeing some loss of insolation.

Armed with the knowledge of Channel 1’s superior aerosol detection, forecasters looked at this imagery to start the process of the smoke’s aerial extent and tracking. However, default colormap ranges within AWIPS made boundary definition of the smoke somewhat difficult during the low light hours of the morning, with little contrast between smoke and clear sky. Luckily, some minor colormap manipulation helped draw out the feature of interest, i.e. the area of smoke. The default minimum/maximum colormap ranges of 0-130 were edited to 10-30. Figure 1 shows the comparison between the default AWIPS range versus the constrained colormap range. While the edited range overexposed clouds within the area of smoke as well as underexposed land features, this was tolerable as it better revealed the spatial bounds of the smoke, from which trajectory analysis could made.

Figure 2 compares Channel 1 (0.47 um ‘blue’ visible band) to Channel 2 (0.64 um ‘red’ visible band), both of which have their colormap ranges edited to 10-30 and 5-30, respectively. Notice Channel 1’s superior ability to reveal the smoke compared to Channel 2. This is due to Channel 1’s sensitivity near the 0.47 um wavelength which can detect scattered solar radiation by atmospheric constituents better than Channel 2’s near 0.64 um wavelength, especially by larger aerosol particulates like dust, smoke, and pollutants. And perhaps the muted detail from land surface features (again, from Channel 1’s increased sensitivity to scattered radiation) as well as lower spatial resolution within Channel 1 compared to Channel 2 helps draw out the overall feature of interest, i.e. smoke.

The area of smoke was just within viewing range of GOES-East’s Aerosol Optical Depth product (AOD), too. This product helped increase confidence of the smoke’s extent near the leading and trailing edges, especially further into the day as the visual enhancement of the smoke due to additional scattering lessened with a decreasing solar zenith angle. Figure 3 shows the Aerosol Optical Depth product overlaid on top of range edited Channel 1. Confidence was placed in the presence of smoke for values of AOD generally larger than ~0.2-0.4 (blue-cyan colored). Note the masked data within AOD as flagged by things like clouds and GOES-East’s local zenith angle greater than 60 degrees, as well as others. Even portions of the smoke itself was masked by the product algorithm’s quality flags, however it was assumed smoke was still present between the edges of higher AOD as confirmed by Channel 1 imagery.

While colormap range manipulation during times of low solar light has shown operational relevance, it may be worth mentioning how changing the colormap’s grayscale may affect appearance of the smoke within visible imagery. The “ease” of feature identification can depend upon the observer, and is where subjective analysis of imagery can vary from person to person. Figure 4 displays a four panel animation of Channel 1 with four different colormap configurations: left panels = square root grayscale, right panels = linear grayscale, top panels = AWIPS default 0-130 ranges, bottom panels = range edited 10-30. Which colormap configuration can you most easily track the smoke? Does time of day influence your ability to observe the spatial bounds of smoke between the different colormaps? How about the ranges?

The CIRA Geocolor RGB also helped reveal the area of smoke in addition to differentiating smoke from clouds, as shown in Figure 5.

Ultimately, smoke and clouds did effect temperatures within portions of the Red River Valley into northwest Minnesota by limiting the amount of heating during this day, although it is hard to tell which may have had more of an impact, further compounded upon forecast accuracy of a changing lower level thermodynamic environment associated with thermal ridging and moisture advection ahead of a cold front. Nevertheless, here is an example showing Fargo only topping out at 88 F, whereas the interquartile and deterministic NBM forecasts were all higher. This also likely played a role in lessening convective potential along a cold fropa that afternoon making for a more effective capping inversion. The lack of insolation had a role in the removal of a marginal risk within an update to the Day 1 convective outlook collaborated between FGF and SPC.

Carl Jones (NWS Grand Forks) and Bill Line (NESDIS and CIRA)

While there are many, many, many examples of RGB’s excelling at diagnosing the early stages of convection during the daytime, there seems to be a lack of examples showcasing similar RGB use during the nighttime. This is likely due to the absence of solar reflectance, particularly in the near-infrared. Reflectance within this wavelength spectrum aids in monitoring the stages of early convective growth through easy detection of cloud top glaciation. Such reflectance gives RGB’s like the Day Cloud Phase Distinction its superiority in monitoring the convective lifecycle. But is there an RGB that can perform a similar application without solar reflectance, i.e. at night? This post will attempt to shed light [get it?] on how the Nighttime Microphysics RGB can be utilized in anticipating convective initiation.

On the night of August 18, 2020, forecasters at the National Weather Service in Grand Forks monitored the potential for overnight thunderstorm development, but were unsure if forcing would be sufficient enough to overcome strong capping over the area. There was anticipation of a low level jet to develop somewhere over eastern North Dakota into northwest Minnesota serving as a potential spark to ignite convection through the capping inversion. While questions remained on where exactly this would happen, focus was given to the mesoanalyst role to monitor for this potential.

At around 12:30 am CDT, the Nighttime Microphysics RGB easily picked up on low level stratus developing northeastward over the northern Red River Valley in far northwest Minnesota. This stratus stood out from other nearby clouds with its telltale pale cyan color compared to higher level dark blue and red cousins to the east and south.

The pale cyan color is a result of increased values within the green gun, the Night Fog difference product (10.3 – 3.9 um), as well as a slight dimming in the blue gun, the longwave infrared band (LIR; 10.3 um), while no information was added from the red gun, the Split Window Difference (SWD; 10.3-12.3 um). These were all signs of low level stratus, mainly through the increased values within the Night Fog product indicative of clouds made of up water droplets. While it drew the attention of the forecasters, questions still remained: Does this low stratus represent the seeds to convection? Or is this simply a benign cloud feature?

Over the next hour, characteristics of this cloud feature changed. The low stratus changed from its uniform pale cyan color and ameba-like structure, growing dark red specks that slowly veered and expanded east-southeast.

The change in color is a result of increased values in the red gun (SWD), sharply decreasing values in the green gun (Night Fog product), and further decreasing values in the blue gun (LIR). This indicated parts of the stratus cloud were starting to glaciate as suggested by the sharp decrease in values from the Night Fog product, continued cooling in the LIR, and increasing difference between the “clean” and “dirty” LIR channels (SWD).

The awareness and knowledge of the subtle change in cloud characteristics as illustrated by the Nighttime Microphysics RGB was crucial in realizing the stratus cloud was continuing to grow one or more updrafts that were beginning to glaciate. This is analogous to the Day Cloud Phase Distinction RGB revealing glaciation of water comprised cumulus, a threshold designating convective initiation.

So we proved that the Nighttime Microphysics RGB can be used to assess convective initiation, but we already have other satellite tools to do this for us, particularly the 10.3 um LIR channel. This single LIR band has a long standing legacy as a useful tool in monitoring convective activity at night. But can this application of the Nighttime Microphysics RGB provide additional lead time towards convective initiation compared to monitoring cloud top temperatures on the 10.3 um LIR channel?

The animation above is a time matched side by side comparison of the Nighttime Microphysics RGB and LIR band. In this case, the easily definable signal of stratus becoming glaciated within the RGB gave around 30 minutes to 1 hour of additional lead time in raising awareness toward potential convective initiation compared to typical LIR if using -24 C as a threshold (standard color curve for LIR in AWIPS turns blue at -24 C). This additional lead time allowed forecasters to feel better prepared in messaging and internal warning operations (better preparation =  less surprises and more confidence in warning/no-warning designation).

While the Nighttime Microphysics RGB can provide crucial information of pre-CI development, it lacks valuable cloud top information after CI, an area where LIR still reigns supreme. So why not have both?

The animation above takes the best of both products by overlaying an adjusted LIR color table on top of the Nighttime Microphysics RGB. Simply make values lower than -24 C transparent within the LIR color table and keep it above RGB in the hierarchy of display within AWIPS. The LIR’s bright colors of ongoing convection probably stands out the most displaying details like a sprawling anvil, overshooting top, and warm trench indicative of an above anvil cirrus plume. But the RGB’s input in this same image can help focus attention west and north of ongoing convection. Notice the tight packs of small , discrete but glaciating cells as shown by a reddening color? This should raise awareness towards the potential of additional convection soon to initiate.

This Nighttime Microphysics RGB – LIR “sandwich” yields information that is helpful in both the pre-CI and post-CI environment. The remainder of the loop showed that many cells matured into robust convection. And while not all of these glaciating cells went on to become mature storms (notice the orphan anvils?), it still signaled the potential of additional development outside of ongoing convection. This knowledge directly led to refined messaging of severe threats for targeted locations, bridging the gap between outlook and warning phases.

For those wondering what hazards this event brought: hail. Numerous reports of large hail up to the size of golf balls fell during the early morning hours of August 19, 2020, within the central Red River Valley into northwest Minnesota. More environmental information can be found via SPC’s Event Archive.

Carl Jones
Meteorologist
NWS Grand Forks