Mapping Fires in Ethiopia
Getting a FIRMS grasp on the Tigray conflict
Recently, there has been quite a bit of news about how the conflict in Ethiopia has sparked fires in the country’s Tigray region. For example, Christiaan Triebert and his team at the New York Times put out an excellent story about the fire that destroyed the Hitsats refugee camp.
These conflict-induced fires got me thinking about how to quantify them. Specific stories about fires are extremely compelling, but their true extent can be glimpsed by applying data to map the fires at scale.
To that end, I used historical data from NASA’s FIRMS system to map the fires in the Tigray region over the last five years. FIRMS (Fire Information for Resource Management System) uses a combination of four satellites to take pictures of the earth, which are then processed and run through a NASA algorithm to detect ongoing fires seen in the images. What that large-scale mapping revealed in Ethiopia was stunning.
With the exception of 2017, when there was a severe drought in Ethiopia, 2020 had the most fires detected by NASA since 2014. Moreover, NASA detected these fires in places that previously hadn’t been ravaged by crop burning or yearly wildfires - for instance, near the city of Mekelle or on the country’s northeast border with Eritrea.
Today, I want to show how I downloaded, cleaned, and displayed the data from NASA.
Downloading and Displaying FIRMS Data
NASA allows users to download archived data sets showing summaries of all the fires detected by NASA filtered by country and year. I downloaded the 2019 archive for Ethiopia from the VIIRS S-NPP satellite but hit a wall because Microsoft Excel couldn’t open the file.
To troubleshoot the issue, I fired up Jupyter notebook, imported the 2019 archive file, and pulled some stats for it.
Well that’ll do it - the file has 214,558 lines. I knew I needed to filter it down to make it more manageable, so I filtered the file for fires detected between November 1 and November 30 of that year. (November 2020 was the first month of the Tigray conflict, so I figured it would make sense to compare November 2020 with November 2019, November 2018, November 2017, and so on.)
Better, but the file still has well over 8,800 lines. I needed to get creative if I wanted to make this file workable. What if I filtered the file by coordinates, so that it would only show fires that were between November 1 and November 30 and also within the Tigray region? Luckily the Tigray region is roughly rectangular, so if I could draw a rectangle over the region on a map and identify the coordinates at the corners of the rectangle, I could filter the file for fires detected in coordinates above, below, left, and right of the sides of the rectangle covering the region.
Luckily, Doogal.co.uk makes this process a breeze. I simply drew a rectangle over the Tigray region, clicked over to the CSV tab, and copied and pasted the relevant coordinates there into my code:
Success! The file filtered by dates and coordinates has only 726 lines. To double check that I had drawn the box correctly and hadn’t messed up my code in any way, I exported the newly filtered file to a CSV and opened it in Google Maps.
There are a few fire detections outside the borders of the Tigray region (in the bottom left, for instance), but on the whole it’s fine for our purposes - after all, this isn’t a scientific paper.
I repeated the same steps as above for each annual Ethiopia archive file since 2014 and displayed them, with color coding, in Google Maps.
There it is: every November fire detected by NASA from 2014 to 2020. The color coding is as follows:
November 2014 fires: Green
November 2015 fires: Brown
November 2016 fires: Yellow
November 2017 fires: White
November 2018 fires: Blue
November 2019 fires: Purple
November 2020 (the conflict month) fires: Red
By itself, that image is too crowded and busy to tell us much, so let’s dive into some case studies.
By far the clearest example of where fires raged through the Tigray region is in the northwest corner of the country, near the Eritrean border.
While a few fires were spotted in that area in 2016 and 2018, the region was blanketed by fire in November 2020. Clicking into each individual point shows that NASA’s FIRMS system detected fires in that region throughout the month as well, indicating that the fires didn’t just pop up one day and go out the next. They were seen on November 5, 23, 24, 25, and 30 (and likely other days as well).
While we can’t conclusively prove that the wildfires in the region were caused by the conflict, we do know from news reports that fighting bracketed the northwest border region since the conflict started in early November. The BBC, for instance, reported that fighting has been ongoing in Humera (just west of the image above) since early November 2020. They go on to confirm that the fighting reached Adebay (just south of the image above) later that month, before their sources were forced to flee west to Sudan.
Some of the fires have a clear military connection. These two fires were detected on the same day (November 20, 2020) on a brush-covered hillside north of the town on Kerserber. The hillside slopes up to what appears to be a military base, judging by the large antenna, high walls, ditches, and strategic position overlooking a road connecting the Eritrean capital, Asmara, with the Tigrayan capital, Mekelle.
My guess is that shelling targeting the base also hit the hillside on (or before) November 20, causing the fires to break out. It is less clear whether other fires targeted military infrastructure.
FIRMS detected this fire on November 21 in Quiha, just north of the airport servicing Mekelle, the Tigray capital. While the airport also contains an air force base, the fire is actually at a building belonging to the Meles Zenawi Institute of Science and Technology Campus. Did belligerents target the campus on purpose? Were they trying to hit the military section of the nearby airport? Or did a short circuit cause a totally unrelated fire at the building?
Finally, some fires were certainly started by the fighting and also destroyed purely civilian infrastructure. This one, for example, was detected on November 26, and was concentrated directly over the Adwa Almeda Textiles factory.
News reports and social media activists confirmed that the factory was ransacked, looted, and destroyed during the fighting, allegedly by Eritrean troops. In these images, the charred wreckage of the sewing machines and factory building is clearly visible.
Lastly, using a code very similar to the snippet I used to pull the yearly FIRMS data for Ethiopia, I ran some statistics for the fires detected by NASA in the Tigray region from year to year.
As the program output shows, NASA detected 838 fires in the Tigray region in November 2020. Running the same code for each yearly file produced the following results:
656 fires in November 2014
418 fires in November 2015
815 fires in November 2016
1,461 fires in November 2017
690 fires in November 2018
726 fires in November 2019
Granted, 2017 is an outlier. The significantly higher number of fire detections in 2017 could be due to the severe drought in Ethiopia that year, which may have made crops and land more susceptible to burning. Aside from 2017, 2020 had more fires than any other year since 2014.
Based on this analysis alone, it’s impossible to tell what exactly caused every single one of the 838 fires detected in the Tigray region in November 2020. However, the FIRMS data clearly shows that not only were more fires than normal detected in the region that month, they were also detected in unique areas - occasionally areas connected to the conflict, but often civilian areas containing people simply trying to survive.
Regardless of the cause of the fires or their location, I hope mapping aggregate fire detections using NASA data and Google Earth can at least provide a new sense of the scale and impact of fires in conflict areas like the Tigray region.