In late July 2018 wildfires ripped through coastal areas in Greece, causing thousands to flee and a number of fatalities.
On 23 July the devastating fires engulfed villages around the Greek capital of Athens.
It was reported that 15 fires were started at different areas, which primarily affected the villages of Mati and Kineta.
At least 80 people were reported to have died, with many more missing or injured.
On the Meteosat-8 Natural Colour imagery from 23 July (Figures 1 and 2) the plume of smoke (thin blue areas) from the Kineta fires can be seen moving eastwards, off the coast and towards Turkey.
Figure 2: Meteosat-8 Natural Colour RGB animation, 12:00–18:00 UTC
The major fires were also observed by both VIIRS instruments on 23 and 24 July. On 23 July the very large values and area covered by the fire could be clearly seen on the fire radiative power product from VIIRS on Suomi-NPP (Figure 3). This shows the fire near Kineta, which started at on the Monday morning.
The Suomi-NPP 11 µm channel imagery from 24 July (Figure 4), which was not cloud covered over the fires (unlike the Day-Night Band channels in Figures 5 and 6), showed the hot spots, indicating a somewhat intense fire.
Fire Radiative Power (FRP)
11 µm channel
On 24 July imagery was captured at roughly 00:05 UTC, from a near nadir overpass from Suomi-NPP (Figure 5), and again from NOAA-20 (Figure 6) about 50 minutes later. While the NOAA-20 pass was cloud covered, both satellites observed major fires near Mati and near Kienta in the infrared and near-infrared channels.
By Jose Prieto and Vesa Nietosvaara (EUMETSAT)
Information on the fires could also be gathered by using LSA SAF data analysis, plus comparisons with MODIS fire products.
On 23 July, at four particular times, some locations are detected as fires near Athens, but rejected for the FRP calculation (Figures 7 and 8). See Figure 2 for the widespread cloud. The LSA SAF provides 15-minute products for fire detection and monitoring (FDeM, Figure 7) and fire radiative power (FRP, Figure 8).
Figure 7 shows that the detection is only successful at one fire location each time, at different pixels, because the cloud restricts the number of detected fires.
Figure 8 is the fire radiative power (FRP) product at 13:00, 16:00, 19:00 and 22:00 UTC on 23 July 2018, superimposed on Meteosat-11 channel 4 (3.9 µm) image. The blue squares indicate that a calculation has been done for the location as part of the FRP generation. The amount of red inside indicates radiative power. A few fires showed up in the imagery as hot spots (except at 16:00 UTC), but, perhaps due to the cloud in the surroundings, the power values were not retrieved.
For reference, Terra MODIS picked up more fires during its pass over Greece on 23 July (Figure 9).
The MODIS active fire product detects fires in 1 km pixels burning under cloud-free conditions using a contextual algorithm, where thresholds are first applied to the observed middle–infrared and thermal infrared brightness temperature. Then false detections are rejected by examining the brightness temperature relative to neighboring pixels (Giglio, L. et al. 2003). Due to the better resolution of MODIS, smaller fires are picked up, since a higher fraction of pixels are free of cloud, and shows higher brightness temperatures on infrared channels, facilitating the detection.
In areas where there are a lot of fires, like southern Africa in July (see Figure 10), MODIS finds many more fires than Meteosat, due to the former’s horizontal resolution. LSA SAF calculates the radiative power for a good fraction of them, in particular for clear areas and far from the boundaries of the Meteosat field of view, where the difference between brightness temperatures at 3.9µm and 10.8 µm gets distorted by the limb effect, detracting from 3.9 µm brightness.
The LSA SAF Meteosat detection algorithm considers the above mentioned difference (3.9 µm–10.8 µm) for the FRP estimate, but also neighbour and texture analysis. The sub-pixel fractional area occupied by a fire will [often] be too low to significantly elevate the radiance of the target pixel above the background. This results in an error of omission (i.e. a true fire on the ground fails to be detected). This process can be repeated across a landscape, meaning that SEVIRI can fail to detect a proportion of the smaller/less intense fires that higher spatial resolution sensors, such as MODIS are able to quantify.
Sentinel-2, for comparison, on a higher resolution depicts the fire around Kineta on 24 July (Figure 12). Some enhanced versions of this image circulated in the media, giving the wrong impression that Sentinel-2 sees raging flames through its solar channels.
Following a fire the Copernicus Global Land Service Proba-V 300 m Enhanced Burnt Area product can be used to estimate the total burnt area. The product shown here is a decadal map from the period 21-31 July 2018 (Figure 13), and it confirms the large burnt area near Kineta.