The war in Ukraine not only caused significant human tragedies, but also caused serious damage to the surrounding natural environment. A recently developed tool provides an opportunity to assess the level of destruction of forests in Ukraine as a result of military operations, which opens new perspectives for understanding the environmental impact of the conflict.
Among the many victims of Russia’s full-scale invasion of Ukraine are some of Eastern Europe’s most important ecosystems: Ukrainian forests and protected areas. However, the full amount of damages is unknown. That’s why we’re launching a new tool to help open-source researchers track destruction from afar.
In September 2022, Ukrainian environmental researchers visited national parks, which are more resilient to climate change than artificial plantations and support important biodiversity, to assess damage to forests and wildlife. Initial findings revealed broken trees, damaged root systems from trenching, and unexploded ordnance strewn across protected lands.
“Forests have suffered a lot on the front lines…huge areas of forests have been mined,” Yehor Hrynyk, an environmental activist with the Ukrainian Environmental Protection Group, told Bellingcat. But a significant part of huge Ukrainian national parks, mountainous areas and forests are inaccessible for environmental monitoring on the ground. This is where open source methods come in handy.
We launched the “OSINT Forest Area Tracker” hosted on Google Earth Engine. Our tool compares data collected by Sentinel-2, a satellite that detects changes in infrared wavelengths and can be used to study the condition of forests.
The tool reveals the scale and intensity of anomalous changes on land. This narrows the range of searches for researchers dealing with the problems of environmental damage in Ukraine.
Importantly, the map does not indicate the cause of these changes, meaning that it is imperative to find corroborating evidence from other sources before concluding that they were the result of military activity.
The tool uses the Normalized Burn Ratio (NBR) index to assess burn severity. Researchers can also use this tool to select custom date ranges for geographic locations of interest.
Since there are more than 7,500 objects in the official database of protected areas of Ukraine, we decided not to study them all – among them there are botanical gardens, city parks and archaeological sites. This list also includes many areas in the far west of the country where there has been no intense conflict.
Therefore, we selected 16 areas where the highest number of fires were detected during the first year of the war, based on the Moderate Resolution Imaging Spectroradiometer (MODIS) data. MODIS is a sensor that allows satellites to detect thermal anomalies, including fires in active combat zones (along with VIIRS, MODIS data can be accessed through FIRMS.
The tool includes a drop-down list of preset areas from around the country, including those near military operations. These given territories are denoted by their abbreviations, for example ShNP for the “Svyati Gory” NPP. A full list of these abbreviations can be found on the tool’s GitHub page.
If researchers are interested in areas of the country that are not included in the drop-down menu, the coordinates can be entered manually.
Although the new tool is focused on Ukraine by default, the methods it uses can be used to analyze areas in other parts of the world.
To show you how the tool works, let’s evaluate the Holy Mountains National Park, a protected area in eastern Ukraine.
This forest area, also known as the Holy Mountains National Park, is located in the hills in the north of Donetsk Oblast, near the border with Kharkiv and Luhansk Oblasts. It is famous for the Sviatohirska Lavra monastery, which lies along the Siverskyi Donets River — which separated Russian and Ukrainian forces for several months before the Ukrainian counteroffensive.
In May 2022, the Ukrainian Conservation Group (UNCG) expressed concern about the consequences of hostilities in the region. So what can the OSINT Forest Area Tracker tell us about this?
We can look for possible forest damage by comparing data before the 2021 invasion and during the 2022 invasion. It is usually best to compare the same time periods to account for seasonal changes. I would not like to compare the summer in Ukraine in 2021 with the winter of next year – vegetation and tree crowns could not be compared even in the absence of an armed conflict.
The image below shows the difference in NBR (dNBR) from June 1, 2021 to September 20, 2021 compared to the same period in 2022.
The OSINT Forest Area Tracker is powered by Google Earth Engine, a geospatial platform that allows researchers to analyze remote sensing data by importing data sets from a variety of satellite sources. These include full-color images, as well as colors that represent infrared waves that can be applied to the Earth’s surface. These different data sets are suitable for studying a wide range of characteristics, such as temperature or humidity, that you won’t see as easily in standard photographs.
The data used by Forest Tracker comes from the Sentinel-2 satellite, which collects near-infrared (NIR) and short-wave infrared (SWIR) bands as it orbits the globe. When an area is burned, the NIR reflectance decreases due to vegetation loss, while the SWIR band increases.
The calculation that gives us the normalized burn rate is (NIR – SWIR) / (NIR + SWIR). The number ranges from 1 to -1, with negative values indicating burned areas or damage. Therefore, a high NBR value (indicated by green pixels) may indicate healthy vegetation, while a low value (red pixels) may indicate bare ground or burned areas.
We say “may” because fires are not the only effect on NBR results. Drought, logging and climatic conditions can also contribute to changes in NBR. This is why the term “unburned” should be understood in context and not always taken literally.
The tool is designed to present data only in areas where you would expect to find forests, and excludes areas that the NBR does not effectively study, such as buildings and roads. This is achieved by displaying readings only in areas that the Dynamic World dataset classifies as forested, even within the aforementioned protected areas.
Corroborating findings with other methods, such as social media reports, real-color satellite imagery, fire data, or local eyewitness accounts, can confirm the cause of a negative NBR score.
This is why the tool itself may not tell you why the area might have changed, but it can tell you where to look. It is better to use it in combination with other research methods.
The tool calculates the NBR for a collection of before and after images. Then the delta normalized burn rate (dNBR) is displayed; delta refers to the difference between two data points. With dNBR, we can establish the extent of forest change between two time frames.
The desired time period for comparison can be selected in the right column of the tool below. It is important to choose time periods for comparison – this allows the tool to draw from a wider range of satellite remote sensing data to improve accuracy.
More images are likely to provide better evidence of anomalous changes rather than regular changes, such as those associated with plowing agricultural land or loss of tree cover in deciduous forests during winter.
It is possible, especially during the winter months, that large areas for analysis may be obscured by clouds. A technique called masking was used during the development of the tool to minimize the effect of the cloud on the results.
Google Earth Engine also allows you to use this method with data from several satellites, including Landsat. However, our tool uses the Sentinel-2 satellite because it has relatively fresh data available for import, allowing us to analyze events over the past few months. These Sentinel-2 parameters allowed us to create a map that shows dNBR, displaying colors that indicate different levels of deforestation. You can read more about the NBR tool and index in the GitHub repository.
As you can see by zooming in, the results in the Holy Mountains National Park seen above include large areas that appear to show severe forest disturbance. Note the red pixels along the southern edge of the study area.
Red and purple pixels show particularly affected areas with low dNBR values, especially near the Siverskyi Donets River. However, the cause is not confirmed; this may indicate burning and deforestation. In early August 2022, open source researchers geolocated a large-scale shelling near the same river.
We can further test our results by performing the same analysis, but instead of comparing 2022 to 2021, we can compare 2022 to 2020 and 2019.
The results follow the same pattern, which appears to indicate widespread disturbance across large parts of the national park.
These results give us an even longer-term assessment of forest health. Furthermore, they indicate that the scale of change observed in this forest is truly anomalous.
By comparing groups of remote sensing satellite data collected months apart, we can also use the program to see the more immediate impact of fires or forest damage. For example, the screenshot below compares a collection of images between April 1, 2022 and April 20, 2022 with a collection just two months later, between June 1 and June 30.
The red color is less visible than in the previous comparison, especially around the southern edge of the protected area along the Siverskyi Donets River. This does not mean that the territory remains unchanged, rather, the scale of changes is minimal in the short studied period.
To the south of the city of Lyman in the protected area you can see a yellow-orange color.
An abnormal moderate or severe change over such a short period can be more easily attributed to a specific cause. Indeed, looking at the same area on MODIS, the fire data shows numerous fires detected around May 2022 in and around the Holy Mountains National Park. Some of these areas coincide with the yellow and orange areas in the image above.
Inspection of one of these areas on satellite images on the dates specified in MODIS provides visual indications of fires in forested areas.
In this case, the tool showed us where to look, narrowing down the entire territory of Ukraine to selected conservation areas, and then to a specific part of one of those areas where anomalous changes were observed.
Again, the cause of these fires cannot be determined using this tool alone, and may need to be established through other open source material or through on-site reporting.
Damage to the forests of the Holy Mountains was detected by first searching the protected area with this tool and then checking it with MODIS and satellite images. However, the tool is also useful when a researcher already knows where to look and wants to verify claims of damage to a specific conservation area.
Eyes On Russia, a conflict monitoring project led by the Center for Information Resilience and Bellingcat, verified attacks, troop movements and battles by analyzing social media footage. Its researchers have already reviewed thousands of incidents during Russia’s invasion of Ukraine based on information from open sources.
By importing the Eyes on Russia dataset into QGIS, a free and open-source geographic information system (GIS) program, we can analyze which verified conflict events have occurred within or near protected areas. We can do this by running a “polygon point count” in QGIS, which counts the number of geographic points – in this case conflict event locations – in each protected area.
For this analysis, we created a one-kilometer buffer around the perimeters of protected areas to capture conflict events that may have occurred nearby. Some events may have been geolocated, for example, outside of a national park, but the activity may have had some impact in those areas, such as spreading forest fires or digging trenches.
The protected area that showed the highest number of confirmed conflict events in the first year of the war was around the Dnieper estuary near the city of Kherson. The territory stretches along and around the river that flows to the Black Sea.
The Kinburn Peninsula in southern Ukraine is located at the mouth of the Dnipro River. It is an area of great strategic importance, given its proximity to the Black Sea and the port of Mykolaiv. It is also an ecologically important area where the Black Sea Biosphere and the Svyatoslav Ivory Coast National Nature Park are located.
By looking at publicly available data on conflict events collected by the conflict monitoring group Eyes On Russia, it is possible to see both the locations of the fires and whether there may have been a link between military action and damage to protected areas. The OSINT Forest Area Tracker allows us to assess potential links to military operations and their impact on protected areas in shorter timeframes.
In late July 2022, the Eyes on Russia team reviewed images from open sources showing shelling. The Telegram post contained several images of burning trees, which were said to have been caused by shelling.
“On July 30, from approximately 11:00 a.m., the Kinburn spit was subject to systematic artillery fire: 3-4 fire at one point… The forest caught fire almost immediately.”
Using visual clues in images and fire detection data, open source researcher @davidnewschool geolocated the incident, which occurred near the village of Vasylivka on the Kinburn Spit. https://twitter.com/DavidNewschool
Images shared by David in a post on X, the social network formerly known as Twitter, show the damage to the area. However, there are limits to what true-color satellite images of the type shown at left can show.
Using the dNBR readings in the OSINT Damaged Area Tracker allows us to further see which areas appear to be the most damaged in this protected area. When we compare the images from July 1, 2022 to July 20, 2022 and from August 10, 2022 to August 30, 2022, the tool shows us that this protected area has areas of low to moderate burn severity following the attacks. Shown below are the yellow and orange areas.
In the case of the Holy Mountains, the tool identified damage to the protected area that merits further investigation. In the case of Kinburn Spit, this allowed us to further verify existing open-source claims of an attack that damaged the forest – also enriching our knowledge of the extent of the damage, which was harder to see in actual color satellite images. .
However, both cases demonstrate the importance of confirming the results of the instrument by other sources before drawing any conclusions about the causes of such harm.