Canadian forest fires of August 2017
by Jos de Laat (KNMIKoninklijk Nederlands Meteorologisch Instituut), 12 September 2017
In the first half of August 2017, Canada was suffering from some of the worst forest fires in living memory, and possibly the worst since 1950. These fires were accompanied by very large plumes of smoke. On 14 August 2017, the EUMETSAT GOME-2 satellite instrument on board of Metop-B captured a very large smoke plume over northern Canada that was at least 1000 km across and 2000 km long (Fig. 1).
Figure 1. GOME-2 (Metop-B) absorbing aerosol index measurements on 14 August 2017 (left) and 21 August 2017 (right). Maximum GOME-2 AAI values exceeded 25 and were among the largest values ever recorded.
In the ensuing days this plume started to move east, as parts of the plume were captured by the jet streams. Remnants of the plume reached Europe in the following two weeks, some of it visible in the lower stratosphere (Fig. 2). Remnants of the smoke plumes were still detected by satellites weeks later, in particular in the lower stratosphere up to 20 km altitude.
Not only is this a scientifically interesting episode (size, severity, advection), it is also relevant from the perspective of the development of EUNADICS-AV services.
Figure 2. Vertical profiles of clouds and particles over Payerne, Switzerland, measured by a Meteo Swiss Raman Lidar, on 20 August 2017. The smoke layers can be seen as slim blueish bands and are in this located at the tropopause. Courtesy Giovanni Martucci.
Detecting smoke by satellites
Although figure 1 suggests that detecting smoke and smoke properties by satellites seems to be quite easy, it actually turns out to be a rather complex issue.
First of all, smoke particles are generally quite small and smaller than InfraRed (IR) wavelengths emitted by Earth and its atmosphere. This means that smoke cannot easily be detected by IR satellite instruments. This strongly reduces the number of satellite instruments and satellites that can measure smoke. In essence, there are only the geostationary broadband UV/VIS instruments like SEVIRI, the polar orbiting broadband UV/VIS instruments like MODIS, and the UV/VIS polar orbiting spectrometers like GOME-2, OMI, and OMPS.
This also means that night detection of smoke is not possible. Thus, typically during half of the day there is no satellite information on the smoke available.
The second complication is that smoke parameters can be difficult to retrieve. For example, because there is no IR information (temperatures), identification of smoke heights is more difficult. IR measurements can be converted into brightness temperatures which can be used to estimated smoke heights.
In principle it is possible to use UV/VIS spectrometer data to retrieve aerosol layer heights by using oxygen absorption in the O2A (GOME-2) and O2-O2 (OMI) wavelength bands. However, these algorithms are still in an development phase and no operational products were available.
Active instruments like CALIPSO or CATS currently only provide information about aerosol layers and layering with several days of delay, and thus are not useful for estimating aerosol layer heights from an operational perspective (within a few hours).
Both polar and geostationary satellites do provide estimates of the aerosol optical depth (AOD), but only during daytime. However, an instrument like MODIS turns out to flag optically very dense smoke layers as clouds rather than as an aerosol layer (Figure 3 , middle panel; Figure 4 ).
The cloud top height data from MODIS then does provide height information, but this comes from InfraRed measurements, and like stated earlier, is not sensitive to smoke at all. So, IR cloud heights are associated with clouds within the scene, which will only be representative of the smoke height if the cloud resides in the same layer as the smoke. However, information on whether the cloud is below, within, or above the smoke layer is not easily to derive (see Figure 3, lower panel).
Figure 3. MODIS (TERRA) pseudo-RGB image on 14 August 2017 over central Canada (top panel). In the middle panel the MODIS AOD is overlaid. In the bottom panel the MODIS CTH is overlaid. Courtesy NASA EOSDIS Worldview webpage (https://worldview.earthdata.nasa.gov).
Discriminating aerosols types
To complicate matters even more, the AOD does not discriminate the type of aerosols. It can be smoke, dust, volcanic, or industrial. The same is true for the AAI, which can also point to smoke, dust, or volcanic aerosols. Note that this would also apply for aerosols and aerosol layer heights from the O2A (GOME-2) and O2-O2 (OMI) wavelength bands.
In this particular event, smoke is identified as it is associated with fires by looking at day-to-day variations in visible imagery in combination with detection of fires (Fire Radiative Power by MODIS or SEVIRI), lack of IR signal identifying volcanic ash, and the general knowledge or understanding that it is improbable that the aerosols at this location would be desert dust.
However, none of this is easy to mold into an numerical algorithm that just looks at the retrieved values of the various parameters with operational data. Identification of aerosol types is possible using measurements of the polarization of light, which can be done by for example the CALIPSO instrument, but which is not operational. In the past, the POLDER instrument on the PARASOL mission could be used to discriminate between aerosol types, but that mission ended in 2013.
Although smoke layers have clear signatures that can be identified by satellites - in particular the property of smoke to be dark and absorb sunlight, other aerosol types have similar physical properties. Furthermore, smoke can only be detected using visible radiation, which in case of passive sensors excludes typically half of the day (night). This strongly limits the number of satellites and derived products that can be used to identify smoke, and also complicates directly measuring the height of the smoke layers.
The August 2016 Canadian forest fire event highlights the complexities of detection and attribution of atmospheric smoke, and for EUNADICS-AV it provides valuable lessons on detection and attribution using multiple sources of information.
Figure 4. Upper panel as figure 3, middle panel, but for 20 August 2017. The insert shows the corresponding OMI AAI, which highlights a smoke plume with index values >> 5, yet MODIS is unable to characterize this plume as an aerosol layer and instead identifies it as a cloud. The lower panel is as in figure 3, lower panel. Clearly not only does MODIS characterize thick aerosol layers as cloud, MODIS also attributes a CTH to the smoke for regions that are clearly cloud free (circle). Furthermore, the MODIS cloud height is low (< 1 km) while it is obvious that the smoke is located high in the troposphere around 5-10 km and even higher).