Author Archives: Scott Lindstrom

Methods of Fog Detection in the GOES-16 Era

GOES-16 ‘Fog Detection’ Channel Difference (10.3 µm – 3.9 µm), 0912 – 1132 UTC, 29 June 2017 (Click to enlarge)

GOES-16 data posted on this page are (still!) preliminary, non-operational data and are undergoing testing

The 16 channels on the GOES-R Series Advanced Baseline Imager (ABI) allow for many different channel combinations that can be used to detect atmospheric phenomena. The animation above shows the traditional method for detecting low stratus: the brightness temperature difference between the shortwave infrared (3.9 µm) and the cleanest longwave infrared (10.3 µm) windows. Cloudtops composed of water droplets are highlighted in the animation because they do not emit 3.9 µm radiation as a blackbody, but do emit 10.3 µm radiation as a blackbody; thus, the brightness temperature difference at night (when no reflected solar radiation at 3.9 µm is present) is positive. The range of the colorbar in the above animation is from -50 to +50 C; stratus appears as green over much of northern Wisconsin, Minnesota and North Dakota.  Higher cirrus clouds are cyan, and they interfere with the satellite detection of low clouds, especially over eastern North Dakota where IFR conditions were widespread (source), and where a Dense Fog Advisory existed.  Note the apparent disappearance of the fog signal — in green — as the sun rises.  Increasing amounts of reflected solar radiation are causing the brightness temperature difference value to switch sign from (10.3 µm – 3.9 µm) > 0 at night (because of emissivity differences) to (10.3 µm – 3.9 µm) < 0 during the day (because of solar reflectance).

The ‘Nighttime Microphysics Advanced RGB’ is also used as a fog detection device. In the animation below, low stratus (and by inference, fog) is highlighted in cyan, a signal that comes mostly from the ‘green’ part of the RGB, namely the Brightness Temperature Difference as shown above. Because the two products are linked by the 10.3 – 3.9 brightness temperature difference, shortcomings in that product as far as fog detection affect the RGB. Note how the fog signal erodes over Minnesota/Wisconsin as the sun rises, and how it is obscured by high clouds (dark purple/magenta) over North Dakota.

 

Note: GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information; GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017

GOES-R IFR Probability fields, shown for this event below, were designed to mitigate detection issues noted above.  Where high clouds are present, meaning the satellite cannot detect low clouds, information about low-level saturation from the Rapid Refresh is used to assess whether or not fog is likely.    That low-level information from the model also can be used to distinguish between fog and elevated stratus that can look very similar from the top, as a satellite views it.  The fusing of model and satellite data makes for a product that has better statistics in detecting low ceilings and reduced visibilities.

At the end of the two animations above, for example, how confident will a satellite analyst or forecaster be that there is dense fog over eastern North Dakota?  How about the analyst/forecaster using IFR Probability fields? IFR Probabilities maintain a signal for fog over the entire region from North Dakota to Wisconsin, even through sunrise and under high clouds.

GOES-R IFR Probabilities, 0915 to 1115 UTC on 29 June 2017 (Click to enlarge)

Fog Detection Methods when Cirrus Clouds are present

GOES-R IFR Probability fields, 0545, 0800, 1000 and 1300 UTC on 13 June 2017 (Click to enlarge)

GOES-16 data posted on this page are (still!) preliminary, non-operational data and are undergoing testing

Note: GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information; GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017

Reduced visibilities and lowered ceilings — IFR and Low IFR conditions — occurred over a wide stretch of the southeastern United States on the morning of 13 June 2017.  The animation above shows enhanced IFR Probabilities aligned northwest to southeast from central Alabama to northwest Florida, the region where IFR Conditions develop.  The flat nature of the field suggests that satellite data are not widely available as a predictor for low clouds on this morning, and that is because of widespread cirrus clouds over the southeastern United States (Click here to see the 0545 and 1300 UTC GOES-13 Water Vapor Imagery with cold brightness temperatures over the southeast).  When high clouds are present, satellite-only detection of low clouds is not feasible.

For example, the brightness temperature difference field from GOES-13 (3.9 µm – 10.7 µm), below, has little predictive value for the northwest-to-southeast-oriented feature of low clouds/reduced visibilities because cirrus clouds are blocking the view.

GOES-13 Brightness Temperature Difference (3.9 µm – 10.7 µm) at 0545, 0800, 1000 and 1315 UTC on 13 June 2017 (Click to enlarge)

Various other detection techniques that rely on only satellite data similarly will be challenged by the presence of high clouds. For example, the Nighttime Microphysics RGB (From this site) shows little signal of fog (cyan/white in the RGB composite) over Georgia and Florida. Intermittent signals will appear occaionally as high clouds thin to allow the low-cloud signal through. The GOES-16 Brightness Temperature Difference field (10.33 µm – 3.9 µm) similarly is challenged by the presence of high clouds. The example shown at 1300 UTC shows a negative value because of reflectance off of high clouds.

Because the IFR Probability fields include surface information in the form of output from the Rapid Refresh model (saturation in the lowest part of the model is used as a predictor for the presence of fog), IFR Probability fields can fill in regions where satellite data cannot be used to detect low clouds because of the presence of high clouds. In regions where high clouds are present, satellite-only detection of fog and low stratus will always be a challenge.

Dense Fog under high clouds in the Deep South

GOES-13 Brightness Temperature Difference (3.9 µm – 10.7 µm) Values at 0815 UTC on 23 May 2017 (Click to enlarge)

Note: GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information; GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017

The legacy method of detecting fog/low clouds from satellite is the Brightness Temperature Difference product that compares computed brightness temperatures at 3.9 µm and at 10.7 µm. At night, because clouds composed of water droplets do not emit 3.9 µm radiation as a blackbody, the inferred 3.9 µm brightness temperature is colder than the brightness temperature computed using 10.7 µm radiation. In the image above, the brightness temperature difference has been color-enhanced so that clouds composed of water droplets are orange — this region is mostly confined to southeast Texas. Widespread cirrus and mid-level clouds are blocking the satellite view of low clouds. IFR and near-IFR conditions are widespread over east Texas, Louisiana and Mississippi. The GOES-R IFR Probability field, below, from the same time, suggests IFR conditions are likely over the region where IFR conditions are observed.

This is a case where the model information that is included in this fused product (that includes both satellite observations where possible and model predictions) fills in regions where cirrus and mid-level clouds obstruct the satellite’s view of low clouds.. As a situational awareness tool, GOES-R IFR Probability can give a more informed representation of where restricted visibilities and ceilings might be occurring.

IFR Conditions continued into early morning as noted in this screenshot from the Aviation Weather Center.

GOES-R IFR Probability fields, 0815 UTC on 23 May 2017 (Click to enlarge)

IFR Conditions under multiple cloud decks in the Upper Midwest

GOES-R IFR Probability Field, along with observations of surface visibility and ceiling heights, 1100 UTC on 17 May 2017 (Click to enlarge)

Note: GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information; GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017

The GOES-16 data posted on this page are preliminary, non-operational data and are undergoing testing

A morning screenshot from the Aviation Weather Center website shows a wide region of IFR and Low IFR Conditions reported over the Upper Midwest, from eastern North Dakota eastward to Lake Superior. The GOES-R IFR Probability field, above, from 1100 UTC on 17 May 2017, shows high probabilities over that region.

Over much of Minnesota and Michigan, the character of the IFR Probability field is flat.  This is typical of IFR Probability when high clouds prevent satellite data from being used as a statistical predictor for IFR Conditions.  If high clouds are present, the satellite cannot detect the presence of low clouds, and the chief predictor of IFR conditions will therefore be model data that typically does not vary strongly from gridpoint to gridpoint when IFR conditions are present.  The pronounced boundary apparent in the IFR Probability field that extends nortwestward from Green Bay in Wisconsin is the boundary between night-time predictors (to the west) and daytime predictors (to the east).

GOES-R IFR Probability values are largest in the region of IFR conditions over North Dakota.  In this region, high clouds are not present and the satellite is able to detect low clouds, and that information is part of the computation of IFR Probabilities.  Note also the region in east-central Minnesota where satellite data are also being used in the computation of IFR Probabilities;  the resultant field there is pixelated.

The toggle below shows the brightness temperature difference field between the shortwave and longwave infrared window channels from GOES-13 (3.9 µm and 10.7 µm) and GOES-16 (3.9 µm and 10.33 µm).  This brightness temperature difference field is used to detect stratus, and by inference fog, because stratus cloud tops composed of water droplets emit radiation around 10.3-10.7 µm as a blackbody, but do not emit 3.9 µm radiation as a blackbody.  Satellite detection of radiation, and computation of the inferred temperature of the emitting surface, assumes blackbody emissions.  Consequently, the brightness temperature computed using detected 3.9 µm radiation is colder than that computed using 10.7 µm (or 10.33 µm) radiation.

Both satellites capture the region of low stratus/fog over North Dakota, and the superior spatial resolution of GOES-16 is apparent. Note, however, that neither satellite can detect low clouds associated with dense fog in regions of higher clouds — over Michigan’s Keewenaw Peninsula, for example, or along the shore of Lake Superior in Minnesota. This is an unavoidable shortcoming of satellite-only-based detection of low clouds/fog.

GOES-13 (3.9 µm – 10.7 µm) and GOES-16 (10.33 µm – 3.9 µm) Brightness Temperature Difference fields, 1100 UTC on 17 May 2017 (Click to enlarge)

Dense Fog from the Ohio Valley to North Carolina

Note: GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information; GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017.

The weather.gov website on Wednesday morning 10 May 2017 showed two dense fog advisories, one near Cincinnati, OH and one near Greensboro, NC. The aviation weather website showed an IFR Sigmet in between the two regions of dense fog. The fog formed along a stationary front that sat over the region.

How well did GOES-R IFR Probabilities and GOES-13 Brightness Temperature Difference fields capture this event? The animation of GOES-R IFR Probability, below, computed using data from GOES-13 and the Rapid Refresh Model, shows enhanced probabilities early in the evening that increased with time. The orientation of the field — from west-northwest to east-southeast — aligns well with the regions of developing fog.

GOES-R IFR Probability fields, 0100, 0400 and 0700 – 1100 UTC on 10 May 2017 (Click to enlarge)

The brightness temperature difference field, below, did not perform as well in outlining the region of low ceilings/reduced visibilities because of the presence of high clouds that interfered with the ability to detect low clouds. Consequently, the highest brightness temperature differences (3.9 µm – 10.7 µm) do not align so well with the regions of developing fog. Note also that at the end of the animation — 1100 UTC — increasing amounts of reflected solar 3.9 µm radiation is changing the character of the field from negative to positive. In contrast, the IFR Probability fields (above) maintain a consistent signal through sunrise.

GOES-13 Brightness Temperature Difference fields (3.9 µm – 10.7 µm), 0700-1100 UTC on 10 May 2017 (Click to enlarge)

IFR Probability and Low IFR Probability in the Pacific Northwest

GOES-R IFR Probability fields, hourly from 0300 through 1500 UTC on 4 May 2017 (Click to enlarge)

GOES-R Low IFR Probability fields, hourly from 0300 through 1500 UTC on 4 May 2017 (Click to enlarge)

Note: GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information; GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017.

Dense fog with IFR and Low IFR Conditions occurred along the Oregon and Washington Coasts early on 4 May 2017. The animations above show the evolution of IFR Probability and Low IFR Probability. Note that IFR Conditions/Low IFR Conditions mostly occurred where Probabilities were high, with a few exceptions (KSMP, Stampede Pass, WA; KKLS, Kelso WA at 1400 UTC). Both IFR and Low IFR Probabilities show a general areal increased between 0800 and 0900; this can be traced to a big increase in the brightness temperature difference that occurred between 0845 and 0900 UTC (shown here) that is likely due to stray light intruding into the satellite detectors. (Brightness Temperature Difference values decreased after 0900 UTC — note that the Brightness Temperature Difference enhancement has color starting when ‘counts’ in the image reach -6).

Low IFR Probabilities do a particularly good job above of outlining the regions of visibility and ceiling restrictions along the coasts of Oregon and of Puget Sound.  Note also that a strip of missing satellite data exists at 1100 UTC over northern Washington.  When satellite data are missing completely, IFR Probabilities are not computed.

A difficulty in using Brightness Temperature Difference fields is shown below. The 1300 and 1400 UTC Brightness Temperature Difference fields show an apparent decrease in low clouds detected as the sun rises (in reality, the amount of reflected 3.9 radiation is increasing as the Sun rises). Fog persists through sunrise as shown in the observations; IFR Probabilities (and Low IFR Probabilities) maintain a signal throughout sunrise.

GOES-15 Brightness Temperature Difference (3.9 µm – 10.7 µm) at 1300 and 1400 UTC on 4 May 2017 followed by GOES-R IFR Probability fields at 1300 and 1400 UTC on 4 May 2017 (Click to enlarge)

Dense Fog over the Tennessee River Valley

GOES-R IFR Probabilities, 0100-1315 UTC on 25 April 2017 (Click to enlarge)

Note: GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information; GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017.

Dense Fog Advisories were issued over the Tennessee River Valley on Tuesday 25 April.   The National Weather Service Aviation Weather website highlighted the regions of IFR conditions shortly after sunrise.  IFR Probability fields, above, showed a slow increase in probabilities as ceilings and visibilities lowered during the night. The field outlined the region where IFR conditions were developing/occurring, meaning that it was a good situational awareness tool as the fog developed.

Brightness Temperature Difference fields, below, have historically been used to detect low clouds and by implication, fog. Clouds composed of water do not emit 3.9 µm radiation as a blackbody in contrast to their emissions of 10.7 µm radiation that are more like that of a blackbody. Thus, computed brightness temperature values are colder using 3.9 µm radiation than 10.7 µm radiation over water clouds. In the animation below, Brightness Temperature Difference values cooler than -1 C are highlighted in yellow.

Note that High Clouds in the animation below over the Smoky Mountains prevent an accurate depiction of low clouds formation there.  IFR Probability fields, at top, include a signal in that region because model data from the Rapid Refresh model suggests saturation is occurring.

As the sun rises, at the end of the animation below, increasing amounts of reflected 3.9 µm radiation cause the brightness temperature difference field to flip in sign.  In contrast, the IFR Probability fields, at top, maintain a coherent signal through sunrise.

Alert readers may note that Brightness Temperature Difference fields and IFR Probabilities are not shown from 0400 UTC.  At that time, Stray Light signals were present in the Brightness Temperature Difference field and they contaminated both the Brightness Temperature Difference and the IFR Probability fields.

GOES-13 Brightness Temperature Difference (3.9 µm – 10.7 µm) from 0100 through 1315 UTC on 25 April 2017 (Click to enlarge)

GOES-16 is transmitting non-operational data that are undergoing testing and refinement.  The toggle below shows the brightness temperature difference field from GOES-16 and GOES-13 for one time on 25 April.  Note the superior resolution in the GOES-16 data:  2 km at the subsatellite point vs. 4 km for GOES-13.  As noted at the start of this blog post, GOES-R IFR Probabilities are being computed with GOES-13 and GOES-15 data, not with GOES-16 data.  Incorporation of GOES-16 data into the algorithm will occur near the end of 2017.

The GOES-16 data posted on this page are preliminary, non-operational data and are undergoing testing.

GOES-16 and GOES-13 Brightness Temperature Difference fields, 1300 UTC on 25 April 2017 (Click to enlarge)

IFR Conditions over North Dakota

GOES-13 Brightness Temperature Difference and GOES-R IFR Probability at 0615 UTC on 24 April 2017, along with surface observations (Click to enlarge)

Note: GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information; GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017.

A late-season snow storm was affecting North Dakota on Monday 24 April. Snow amounts were modest and IFR Conditions were widespread.   The multiple clouds layers associated with the storm meant that satellite detection of low clouds/fog was difficult.  The toggle above shows Brightness Temperature Difference (3.9 µm – 10.7 µm) and GOES-R IFR Probabilities.  The IFR Probability field distinctly outlines the region where visibilities and ceilings are restricted by the storm.  It is very difficult to discern from the Brightness Temperature Difference product where low clouds/fog exist.  Because the IFR Probability field incorporates surface information (that is, low-level saturation as predicted by the Rapid Refresh Model), it is better able to alert a forecaster to the presence of IFR or near-IFR conditions.

Dense Fog along the Florida Gulf Coast

GOES-R IFR Probability Fields, hourly from 0415-1107 UTC on 17 April 2017 (Click to enlarge)

Note: GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information; GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017.

GOES-R IFR Probability fields, above, show increases in IFR Probability starting around midnight local time. The IFR Probability fields generally outline the regions where fog occurred (and which led to the issuance of dense fog advisories to the east of Mobile). How did the brightness temperature difference field capture this event? Brightness Temperature Difference fields have been used historically to identify regions of stratus.

GOES-13 Brightness Temperature Difference fields (3.9 µm – 10.7 µm), hourly from 0415-1107 UTC on 17 April 2017 (Click to enlarge)

The brightness temperature difference field, above, shows a slow increase in negative values (negative because the brightness temperature computed from emitted 3.9 µm radiation is cooler than that computed from emitted 10.7 µm radiation over clouds composed of water because such clouds do not emit 3.9 µm radiation as a blackbody), but careful inspection of the field shows IFR conditions in regions outside the largest signal (highlighted in this enhancement in yellow).  This occurs mostly were cirrus clouds are indicated (represented as dark regions in the enhancement, over central Mississippi, for example).  In such regions, the IFR Probability fields can maintain a signal of IFR conditions because low-level saturation is predicted by the Rapid Refresh Model.

This fused product combines the strengths of both inputs:  Satellite detection of low clouds, and model prediction of low-level saturation.

As the sun rises, the amount of reflected solar radiation at 3.9 µm increases, and the sign of the brightness temperature difference changes from negative at night over water clouds to positive.  This toggle below shows the brightness temperature difference at 1107 and 1145 UTC.  The decrease in signal does not necessarily mean fog is decreasing.

GOES-13 Brightness Temperature Difference fields (3.9 µm – 10.7 µm) at 1107 UTC and 1145 UTC on 17 April 2017 (Click to enlarge)

A toggle of the IFR Probability and the Brightness Temperature Difference at 1145 UTC, below, shows that the IFR Probability fields can maintain a useful signal during the time of rapid changes in reflected solar radiation with a wavelength of 3.9 µm.

GOES-R IFR Probability and GOES-13 Brightness Temperature Difference fields (3.9 µm – 10.7 µm) at 1145 UTC on 17 April 2017 (Click to enlarge)

IFR Probabilities with a strong storm in Maine

GOES-R IFR Probabilities, 0100-1000 UTC on 7 April 2017, along with surface reports of ceilings and visibilities (Click to enlarge)

Note:  GOES-R IFR Probabilities are computed using Legacy GOES (GOES-13 and GOES-15) and Rapid Refresh model information;  GOES-16 data will be incorporated into the IFR Probability algorithm in late 2017

A strong storm over the northeastern United States produced widespread IFR conditions over that region.  The storm was also accompanied by multiple cloud layers, however, and that made diagnosis of regions low clouds/fog difficult.  For these cases, a fused data approach is vital — using model information (in the case of IFR Probability, above, the model is the Rapid Refresh) to provide information at low levels allows for a better tool to alert a forecaster to the presence of reduced visibilities.

In the animation above, Maine intially shows IFR Probabilities around 50% — but the flat nature of the field should alert a user to the fact that satellite predictors cannot be included in the computation of IFR Probabilities because high clouds are preventing a satellite view of low clouds.  Accordingly, the computed Probability is lower.  In contrast, high clouds are not present over southern New England at the start of the animation, and IFR Probabilities are much larger there:  both satellite and model predictors are used. As the high clouds lift north from northern New England the region of higher IFR Probabilities expands from the south.

Note the influence of topographic features on the IFR Probability field.  The Adirondack Mountains and St. Lawrence Seaway have higher and lower Probabilities, respectively, because of the higher terrain in the Mountains, and the lower terrain along the St. Lawrence.

An example of why fused data are important is shown below.  Look at the conditions in Charlottetown, on Prince Edward Island, in the far northeast part of the domain.  Between 0315 and 0400, ceilings and visibilities deteriorate as IFR Probabilities increase.  The brightness temperature difference field, at bottom, shows no distinct difference between those two times because the clouds being viewed are high clouds.