Long-Term Conflict and Environmental Change

Abstract

The Karabakh conflict reignited in the winter of 2020 and a new peace accord transferred several key regions from Armenian to Azeri control. At least 90,000 people were displaced and tensions in the region remain high. This chapter uses data from remote sensing to assess the long-term environmental impact of the Karabakh conflict from the 1990s through 2020. We look specifically at changes in land use and land cover. These include analyses of forest disturbance, cropland abandonment and patterns of surface water. We conclude that forest disturbance and cropland abandonment are likely to generate expanded habitat for Anopheles sacharovi—a key malaria vector in the region. Ongoing tensions also suggest the conflict is not yet resolved and the potential for new hostilities is high.

5.1 Resurgent Conflict and Rebordering, 2020

The 1994 cease fire agreement did not establish a peacekeeping patrol along the line of contact, and both countries occupied the line with forces. Low intensity violence was a regular feature in the region with dozens of people killed or injured in border skirmishes almost every year. In 2016 four days of fighting led to hundreds of deaths near the line of demarcation (Crisis Group 2022a, 2022b). In a 2017 report, Thomas de Waal described the area transforming from “a string of hastily dug trenches separating the two armies” in 1994 to “the most militarized zone in Europe, bristling with artillery, long range missile launchers, attack helicopters and military drones” in 2017 (de Waal 2017).

Intense fighting erupted again in September 2020 and continued for approximately six weeks. By October, Azeri forces approached the Lachin Corridor and threatened Armenia’s control of surrounding regions. In retaliation, Armenia launched an offensive along Azeri supply routes. Soon after, Azeri military forces surrounded the historic town of Shusha forcing Armenian inhabitants to flee to nearby Stepanakert. Days of intense street fighting followed with “building to building close combat” (Spencer and Ghoorhoo 2021).

Before fleeing the area, many Armenians intentionally set fire to their properties “finding solace in knowing their enemies could never sleep there” (Cookman 2020). They also killed their livestock or released their animals to roam free along high mountain roads. One reporter described witnessing a tragic “large scale migration of people from a land that … literally changed owners overnight.” One reporter estimated that approximately 90,000 people had been displaced during the 2020 conflict out of a total of about 150,000 in the territory (Kucera 2020). The conflict once again left thousands of refugees stranded without communications, heat, hot water or adequate food. Displaced people expressed confusion, shock and uncertainty about where the new territorial boundaries were located (Cookman 2020).

In the 1980s, Shusha had been home to approximately 23,000 Azeri residents and the town had great symbolic importance as “the cradle of Azeri culture” (Spencer and Ghoorhoo 2021). Most of these residents were killed or forced into exile during the Armenian offensive in 1992. At that time the town was resettled by Armenians but remained underpopulated with only about 5000 inhabitants in 2020 (Spencer and Ghoorhoo 2021). By November the town was fully under Azeri military control and a “lopsided” peace deal was signed that transferred Armenian controlled areas surrounding Karabakh back to Azerbaijan (Spencer and Ghoorhoo 2021). More than 7000 casualties were reported, including soldiers and civilians. Hundreds more were wounded (Global Conflict Tracker 2022). An estimated 30,000 Armenians were forced out of their homes in the area on short notice (de Waal 2021).

The 2020 peace accord was brokered by Russia, and included an agreement that seven districts adjacent to Karabakh that were integrated into augmented Armenia after the 1994 cease fire would be returned to Azerbaijan. Karabakh lost substantially more of its croplands than the area that was recaptured by Azerbaijan. The area of augmented Armenia delineated after the 1994 cease fire was approximately 1683 km². The smaller area of Soviet era Nagorno-Karabakh was 4495 km², with 4262 km² within the line of contact after the cease fire of 1994. After the 2020 conflict, the remaining area of Karabakh was reduced to approximately 3259 km², shrinking the entire area under Armenian control by about 72% and shrinking Nagorno-Karabakh by about 24%.

Russian troops are scheduled to remain in place until 2025 to protect Azeri transit corridors between Karabakh and Armenia, as well as between the enclave of Nakhichevan and Azerbaijan (Sestanovich 2020). As detailed by Thomas de Waal, this accord “radically changes the geopolitical configuration of the region, giving Moscow a central role it last held in the Soviet era” (2021). De Waal goes on to state that the peace accord does not appear stable, with both countries continuing to act “as if they are still at war” (2021). Azerbaijan has allegedly resisted turning over Armenian captives and some Armenian forces remain in Karabakh due to uncertainty about the exact terms of the peace accord.

The revised 2020 territorial boundaries also create trade and transit problems. Much of the Karabakh region has been cut off from Armenia but is not yet integrated into the economy or polity of Azerbaijan. According to de Waal, Azerbaijan’s President Aliyev “has indicated that he intends to keep up a policy of isolating Karabakh from the outside world and severing its political connections with Armenia” (2021). Armenia has also refused to surrender some land claimed by Azerbaijan, including seven villages in the Kazakh District (de Waal 2021). There have been reports of widespread destruction of Armenian churches, cemeteries and cultural monuments in the Nakhichevan region as well as in reclaimed areas of Karabakh (Isayev 2022; Maghakyan 2022; Nutt 2022).

These tensions led to new hostilities in the fall of 2022. Nearly 300 people were killed after Azerbaijan shelled several villages inside Armenia (Crisis Group 2022c). The shelling continued for two days along Armenia’s eastern border until it was halted by a fragile cease fire with both sides accusing each other of instigating territorial aggression (Crisis Group 2022c). Both sides claimed to be unclear about where the actual border between the two countries currently falls. Several analysts have claimed that Russia’s 2022 invasion of Ukraine has further destabilized the Karabakh region by distracting Russia’s attention and revealing the extent of Russia’s military weakness to leaders in Azerbaijan and Armenia (Glantz 2022). The threat of a new offensive has been described as “very high” and prospects for peace are looking “increasingly slim” (Crisis Group 2022c).

5.2 Long-Term Conflict and Environmental Change

The Karabakh region has endured more than 30 years of unresolved conflict and territorial instability with repeated rounds of rebordering, forced migration and resettlement. To assess the cumulative impact of these events on patterns of land use and land cover, we conducted an extensive geospatial analysis that included the following components: (1) a topographic analysis to measure elevation and wetness; (2) a cropland analysis to assess changes in land use and extent of croplands disturbed or abandoned during the conflict; (3) a forest disturbance analysis to assess the impact of the conflict on tree cover; (4) a brief analysis of land and village abandonment and how these factors may have increased mosquito breeding sites after the cessation of formal hostilities. This geographic analysis reveals that multiple environmental disturbances from the conflict changed the landscape in ways that created extensive habitat for Anopheles sacharovi—a key malaria vector in the region.

During times of armed conflict there are often a series of direct and indirect environmental consequences. Direct effects of conflicts can result from an increase in fires and clearcutting, such as was visible in the 2020 Karabakh conflict when 889 forest fires were recorded between October 22 and November 3 (ACAPS 2020). Indirect effects are often more subtle but can be longer lasting. One relevant example is the deforestation that took place in Armenia as a result of the energy embargo imposed by Azerbaijan during the 1990s. The energy crisis lasted several winters and Armenians cut thousands of trees as their source for heating.

In addition to changes in forests and croplands, surface water is another important aspect of the land which is relevant to this conflict and its associated epidemic of malaria. Water shortages and droughts are major sources of concern in Azerbaijan (Palazzo 2020). Several important water reservoirs are located on the recaptured lands, such as the large water reservoir in Mataghis which feeds water to the Sarsang water facility and is currently still controlled by Yerevan-backed separatists. Another reservoir in Mataghis is very important for the irrigation of croplands in the southern part of our study region. Before the 2020 conflict, the irrigation canal from Sarsang that fed these croplands was not providing water to Azerbaijani lowland croplands (Palazzo 2020). Abundant water from the mountainous regions fed large river systems for both Armenia and Azerbaijan.

Overall, our analysis reveals agricultural infrastructure sustained severe damage throughout the primary conflict region. An estimated 22 miles of irrigation channels, which provided water to more than 47,000 acres of farmland, were damaged or rendered dysfunctional due to the disabling of water reservoirs and pumping stations. Abandoned irrigation channels can easily get clogged causing salinization and fertilizer contamination. Disruptions in operations of these irrigation channels in these areas have the potential to create abundant mosquito habitat. While direct relationship between malaria and area of irrigated land is hard to pinpoint, the vast development of irrigated water in these arid regions is an important component for maintaining mosquito density. Populations in and around these areas, including IDPs would be under increased potential for malaria and other vector borne diseases if left unmaintained and abatement procedures were not implemented.

The croplands in post-configuration 2020 Nagorno-Karabakh are generally found on higher elevations, with steeper slopes and low topographic water indices, making these areas less suitable for cultivation than the croplands within the recaptured area. In addition, there was minimal forest loss in both regions (<30km² per year), but major forest disturbance, which can represent forest thinning or other damage (approximately 200km² appears disturbed in the combined region). Additional forest disturbance occurred in the summer of 2020. Since satellite images were recorded before the conflict (June–August 2020), this is most likely not related to the fighting that occurred in the fall. A substantial increase in the number of fires—889 vs approximately 90 in previous years—was detected in the region as a direct result of the conflict. Impacts of these fires should be visible in disturbance data from 2021 (ACAPS 2020).

5.3 Topographic Analysis

The average and median elevations for the 1994 era greater Karabakh/augmented Armenia (which Armenians call Artsakh) are 1277m and 1105m respectively (Fig. 5.1). The average and median slopes are 12.5° and 11.2° (Fig. 5.2A, top). This mountainous area only has some gentler slopes in the southern part on the border with Iran and in the eastern part near Stepanakert/Khankendi. The average elevation of the post-reconfiguration 2020 Nagorno-Karabakh region (PR-2020) is a bit lower, 1119m for the mean elevation and 1026m for the median elevation, but the slopes for PR-2020 are steeper than the overall slopes: the mean slope is 13.9° and the median slope is 12.9°.

Fig. 5.1

The digital elevation model reveals that most areas of 1994 era Greater Karabakh/augmented Armenia are highly elevated, with the highest elevations in the West along the border with Armenia. There are lowland areas in the southern part along the border with Iran. Urban areas are gray

Fig. 5.2

A (top) and B (bottom) The slope (top) and the topographic wetness index (bottom) show that only the far southern area of 1994 era Greater Karabakh/augmented Armenia has suitable slopes (<6°) for croplands. Those areas are also the wettest and most suitable for irrigated agriculture. Urban areas are depicted in gray/red respectively

Figure 5.1 depicts the slopes for 1994 era greater Karabakh/augmented Armenia and PR-2020. Only the most southern part of the 1994 era greater Karabakh/augmented Armenia has gentle slopes, as well as some smaller areas just north of Stepanakert/Khankendi and surrounding Agdam. Most of Agdam was destroyed in 1993 and then abandoned. This town was supposed to be transferred to Azerbaijan as part of the 2020 conflict resolution and might be repopulated in the future. The topographic wetness index (TWI, Fig. 5.2B) shows that most of the wetter areas are located around the gentle slopes in southern 1994 era greater Karabakh/augmented Armenia. These areas correspond with the irrigated parts of the region, and there are many towns in the areas along the rivers with higher TWI. TWI is very low for the PR-2020 region, except for a small area just north of Stepanakert/Khankendi.

In addition to changes in forests and croplands, surface water is another important aspect of the land which is relevant to this conflict and its associated epidemic of malaria. Water shortages and droughts are major sources of concern in Azerbaijan (Palazzo 2020). Several important water reservoirs are located on the recaptured lands, such as the large water reservoir in Mataghis which feeds water to the Sarsang water facility and is currently still controlled by Yerevan-backed separatists. Another reservoir in Mataghis is very important for the irrigation of croplands in the southern part of our study region. Before the 2020 conflict, the irrigation canal from Sarsang that fed these croplands was not providing water to Azerbaijani lowland croplands (Palazzo 2020). Abundant water from the mountainous regions fed large river systems for both Armenia and Azerbaijan.

5.4 Cropland Abandonment

One recent geographic study evaluated the effect of the Karabakh conflict on land surface by comparing the areas within conflict zones with other areas that did not see direct conflicts (Baumann et al. 2015). These researchers found that 9% of the croplands were abandoned between 1987 and 2000, and only 17% of those fields were recultivated by 2010. Much higher abandonment rates were found in the areas where the actual fighting took place. Cropland abandonment was common after the collapse of the Soviet Union, even in areas without conflict, and a strong correlation of abandonment with elevation and slope has been shown (Müller et al. 2013).

The 1994 era Karabakh/augmented Armenia also had a good deal of irrigated cropland, with open irrigation canals crossing the landscape. Our remote sensing analysis showed significant amounts of cropland to the west and south of Yerevan along the Araks River (Fig. 5.3A). The lowland areas of the Kura River valley in western and southern Azerbaijan, including border regions with Iran, also indicate a high percent of croplands (Fig. 5.3B). Other areas with croplands in and around central NK and along the south western border with Armenia showed areas of cropland.

Fig. 5.3

A (top) and B (bottom) Percent of irrigated cropland by pixel (100x100m, Copernicus Global Land Service dataset) for 2015 in the South Caucasus region (A) and Nagorno-Karabakh region (B). A high percentage of irrigated cropland was found to the west and south of Yerevan, throughout the Kura and Araks River valleys in Azerbaijan and central portions of NGK. The boundaries established in 1994 and at the end of the conflict in the fall of 2020 are shown. Cities and towns are highlighted in red to indicate areas of settlement

To understand the changes in cropland, we used two land cover datasets. The first is the v3.0.1 Copernicus land cover data (Buchhorn et al. 2020), which is available for the years 2015 through 2019 at 100m spatial resolution. This dataset is available at the global scale and provides data both in discrete classes and as continuous field layers with proportional estimates of the land cover types. The data is based on the PROBA-V 100m time series and using high-quality land cover training sites and ancillary datasets has been classified with an accuracy of 80% (Buchhorn et al. 2020). We use the land cover classification where each grid cell is assigned a land cover class to determine the changes in the percentage of croplands between 2015 and 2019. We also use the proportional data to understand in which regions we find more than 10% cropland loss. The second dataset is a regional land cover classification generated for the Caucasus at 30m spatial resolution for the years 1987, 1995, 2000, 2005, 2010 and 2015 (Buchner et al. 2020). Overall accuracy was a little over 80% for this dataset as well, with slightly higher user accuracies for forests than for croplands.

Most of the augmented Karabakh region is either herbaceous vegetation or forested (Fig. 5.4), with croplands confined to the far eastern and southern parts of the region. The reduced Nagorno-Karabakh region negotiated in the 2020 conflict resolution is predominantly forested, with croplands just north of Stepanakert/Khankendi and in the far southeastern part of the region.

Fig. 5.4

2019 Copernicus land cover data at 100m spatial resolution. Green is closed forest, pink is croplands, red is urban and yellow is herbaceous vegetation. The underlying hill shade is based on the SRTM DEM data. Note the very low number of urban areas along the line of contact in Artsakh, compared with the relatively dense number of towns across the line of contact in Azerbaijan. This map shows a substantial amount of cropland in the area between Nagorno-Karabakh and the line of contact. High resolution imagery reveals a mixed story in this region

There is one year of overlap (2015) between these two land cover datasets. To identify cropland changes in the region, we first compared the cropland percentages by rayon for the two different datasets in the year 2015 (Fig. 5.5).

Fig. 5.5

Cropland area (left, in km²) and cropland percentage (right %) of 13 rayons that make up 1994 era greater Karabakh/augmented Armenia. There is a significant relation between the two cropland area datasets (p<0.01) with an R² _adj of 0.5537, although there is more cropland visible in the Buchner dataset than in the Copernicus dataset

We found a 5.7% relative loss of cropland area in the entire region of 1994 era greater Karabakh/augmented Armenia between 1987 and 2015 (32.1% to 30.3%), with a 19% relative loss of cropland area in the post-reconfiguration in 2020 (PR-2020) region (29.2% to 23.7%) and just 1.2% in the area that was recaptured by Azerbaijan. This percentage includes croplands that are classified as mixed, meaning that they are classified as cropland in some years and not in other years, without showing consistent gain or loss. Only 2.6% and 2.3% of the area for 1994 era greater Karabakh/augmented Armenia and PR-2020, respectively, was classified as cropland the entire time, with 2.7% of the area captured by Azerbaijan in 2020 cropped the entire time. Most of the change occurred in the eastern part of 1994 era greater Karabakh/augmented Armenia (Fig. 5.6A and B), which also has the most cropland (Fig. 5.4). The PR-2020 area has much smaller amounts of cropland than the area recaptured by Azerbaijan in 2020.

Fig. 5.6

A (top) and B (bottom): Cropland change between 1987 and 2015 (Figure A top, Buchner et al. 2020), and cropland change between 2015 and 2019 (Figure B bottom, Copernicus)

The largest cropland relative loss (47%) occurred in Tartar rayon, in the northeastern part of 1994 era greater Karabakh/augmented Armenia. But several other rayons saw losses of more than 10%. Jabrayil and Fuzuli rayons saw a slight increase in croplands (5.6% and 10.9%), respectively. The Copernicus data reveals a similar pattern in the short period between 2015 and 2019, with the largest decline in percent cropland in Tartar rayon (11.5%) and Agdam rayon (10.6%), surrounding the ghost town of Agdam. Three rayons saw significant increases in croplands: Shusha (43.8%), Kalbajar (28.6%) and Zangilan rayon (14.4%). This dataset shows virtually no change in cropland class, for example from cropland to another land cover class, but 31% of the PR-2020 area revealed a loss in cropland proportion that was more than 10%, for example from 85% cropland in a 100m grid cell, to <75% cropland. The area recaptured by Azerbaijan in 2020 revealed a change greater than 10% in 34% of the region.

The PR-2020 region lost 16.6% of its croplands between 1987 and 1995 as a result of the conflict. The area recently recaptured by Azerbaijan lost 5.9% of its cropland, but much of this was regained in later years. The croplands lost by 1995 had the highest elevation and by far the highest slopes of all the cropland areas lost for both the PR-2020 region and the area recaptured by Azerbaijan. All croplands had higher elevations and steeper slopes in the PR-2020 region than elsewhere in augmented Armenia (Fig. 5.7).

Fig. 5.7

A (top), B (middle) and C (bottom): Median elevation and slope for the remaining area of post-reconfiguration Nagorno-Karabakh (PR-2020) and the area recaptured by Azerbaijan in the 2020 conflict. Elevation for the areas that are never cropped is higher in the recaptured area, while the elevation for all the cropped areas is higher in PR-2020 (top). The slope of all areas is also greater for the PR-2020 region, indicating a rougher terrain in general (middle). Crop loss represents the area where there was at least 10% cropland loss between 2015 and 2019 according to the Copernicus land cover data (middle). While the slope of the area lost in these last few years is virtually identical to the slope of the area lost by 2015, the elevation is lower, indicating a loss of area from 2015–2019 that is better suitable for croplands. The topographic wetness index (TWI) is always higher for the recently recaptured area, once more indicating better cropland suitability for that region than for PR-2020 (bottom)

The topographic wetness index is also higher for all cropland types in the area recaptured than in PR-2020. The areas that are always cultivated have the highest median TWI, while the areas that are never cultivated have the lowest TWI. Areas abandoned by 1995 had much lower TWI for both regions than areas abandoned at later times (Fig. 5.7A, B and C). Croplands are located on less suitable lands in the PR-2020 region than in the recaptured areas. Close-up imagery shines a light onto the different cropland regions along the line of conflict, for example close to Agdam (Fig. 5.8).

Fig. 5.8

Croplands along the line of contact between greater Karabakh/augmented Armenia and Azerbaijan. Burned fields are visible along the line of contact. The fields in Azerbaijan are highly fragmented, which happened after the collapse of the Soviet Union. Despite the abandonment of the town of Agdam in the west of the image, some cultivated croplands are still visible in this region

5.5 Forest Disturbance

Developed by NASA, the earth-orbiting satellite called Landsat 5 was launched in 1984. We currently have 38 years of global 30m multi-spectral data recorded by four different satellites: Landsat 5, Landsat 7, Landsat 8 and Landsat 9. These satellites record data at a variety of different spectral resolutions and are capable of recording data outside of the regular blue, green and red light that we as humans observe, making them particularly useful for analysis of vegetation changes.

To calculate the forest disturbance index for all years between 2000 and 2020, we select all available images with less than 20% cloud cover between June and August for every year. We then calculate the summer mean for all available images in each year. Next, we process that data into a disturbance index (de Beurs et al. 2016; Healey et al. 2005). We first calculate the Brightness, Greenness and Wetness Tasseled Cap indices (Crist 1985). These indices are linear transformations of the original reflectance data to form indices that are more straightforward to link to the land surface. Next, we standardize these indices against forested areas with at least 80% tree cover and no change between 2000 and 2019, using the Hansen Global Forest Change data (Hansen et al. 2013). We also standardize by rayon and aspect, for example, to ensure that north facing slopes are only compared to other north facing slopes. For the aspect standardization we look at north facing, south facing and flat areas (slope <6°). Finally, the disturbance index is calculated as follows:

DI = Brightness – (Greenness + Wetness)

This index can capture forest disturbance because disturbed forests typically are brighter, less green and less wet than healthy forests.

We compared these forest disturbance data from 2000–2020 with a regional land cover classification developed for the Caucasus (Buchner et al. 2020), which also uses Landsat data, but classifies the data in distinct land cover classes for the years 1987, 1995, 2000, 2005, 2010 and 2015. As a result of the energy embargo, some forested areas were cut after the 1993 conflict to provide a heating source. We evaluated forest loss with two different datasets. First, we compare the total forest percentages by rayon for the overlapping years: 2000, 2005, 2010 and 2015. It is important to realize that these forest cover/disturbance datasets are not generated equally. The Buchner data provides a forest/non-forest classification, while the disturbance index data only incorporates pixels that are at least 80% forest according to the Hansen global forest data and then investigates disturbance of those forests. Next, we evaluate the forest loss with the Buchner data which covers the years 1987, 2000, 2005, 2010 and 2015, and we evaluate forest disturbance data for the years 2000–2020.

A comparison of the undisturbed forests and the areas classified as forest by Buchner et al. (2020; Fig. 5.9) reveals generally strong agreement between the two datasets for all years (Table 5.1).

Fig. 5.9

Percentage of undisturbed forest and the percentage of forest from Buchner et al. (2020) by rayon in 1994 era greater Karabakh/augmented Armenia. The red line provides the 1:1 line. There is a significant relationship between these variables with a slope of 1.13, and an intercept of –2.5% (R²_adj = 0.933)

Table 5.1 Regression between the percent undisturbed forest according to the disturbance data (DI) and the Buchner forest data (BF)

If we compare the forest area lost (e.g. forest is classified as another class) and the area disturbed over time, it is clear that while the area lost is small (<30km² in total, Fig. 5.10A, top), there are many more forests that are disturbed (~200km², Fig. 5.10B, bottom). We see a strong decline in deforestation after the initial shock in the 1990s and early 2000s, with the total area of deforestation declining from less than 30km² per year, to less than 10km² per year by 2010. There is a slight increase in deforestation in 2015.

Fig. 5.10

A (top) and B (bottom): Buchner et al. (2020) forest loss in post 2020 reconfiguration (PR-2020) and in the area recaptured by Azerbaijan (top). Note how the amount of forest loss decreases over time. With the largest area of loss from 1987 to 1995 and the second largest area of loss from 1995 to 2000. Forest disturbance between 2000 and 2020 (bottom). Disturbance and forest loss are not the same, disturbance of a forest is possible even if the forest itself is not lost. Note the large amount of disturbance in 2020

The amount of deforestation in the area recaptured by Azerbaijan in 2020 is much greater than the amount of deforestation in the remaining area of Nagorno-Karabakh (PR-2020). On the other hand, Fig. 5.11 reveals relatively stable amounts of forest disturbance, with approximately the same area of forests disturbed in PR-2020 as in the recaptured area. It is interesting to note that while the area of clear-cut forests may be relatively small in PR-2020, there is still a lot of forest disturbance. In addition, the latter data reveals a steep increase in forest disturbance in the area recaptured by Azerbaijan. This 2020 disturbance was recorded with data from June–August 2020, before the actual 2020 conflict began.

Fig. 5.11

A (top) and B (bottom): Forest loss data from Buchner et al. (2020) compared with average elevation (top) and slope (bottom)

The elevation of the forested areas is slightly higher than the elevation of the regions that are not forested (Fig. 5.11A, top). Forests at lower elevations were cut first (1995), with forests at higher elevations cut later, although this trend was stronger for the recaptured area than for PR-2020. The slope of the forested areas is greater in both regions than the slope of areas with other land cover types. Both regions reveal that the slope of the areas with forest loss increases steadily, with the lowest slopes for the areas that are lost by 1995, and the highest slopes for areas lost by 2015 (Fig. 5.11B, bottom). In most cases, the slope of the forests that are lost in PR-2020 is slightly higher than the slope of the forests lost in the recaptured area.

We used the disturbance index data between 2000 and 2020 to determine how long areas were disturbed and found that 69% of the forests are never disturbed. On the flip side, 3% of the forests are disturbed more than 16 years, with 0.65% of those pixels being disturbed the entire time period. This might indicate that these forests were misclassified at the beginning of the study period and in fact were never forest. Just below 12% of the forests were disturbed only one year in the full study period, and many of the pixels that were disturbed in 2020 fall in this category. About 4% of the forests were disturbed between two and six years, indicating either disturbance followed by recovery or a disturbance later in the study period. Figure 5.12 provides the slope of the forests by disturbance length. Those areas that were never disturbed had the steepest slopes, while areas that were briefly disturbed had the gentlest slopes. Figure 5.13 illustrates the extent of forest disturbance before and after the resurgence of armed conflict in 2020, and the widespread forest fires that occurred in the areas recaptured by Azerbaijan.

Fig. 5.12

Slope of the disturbed forests by length of disturbance. The areas that were never disturbed and are continuously classified as forests have the greatest slopes. Gentler slopes are found for areas with less than three years’ disturbance. Areas with more than 16 years of disturbance are most likely misclassified areas that were never forests to begin with (<3% of the forests)

Fig. 5.13

Forest disturbance (brown) observed in the summer of 2020, before the latest conflict occurred. Large forest disturbances can be seen just south of Shusha in the area that was later captured by Azerbaijan and was a hotspot for 2020 fires (ACAPS 2020). There is also a patch of disturbance northeast of Stepanakert/Khankendi. Urban centers are plotted in dark gray

An overview of the socio-economic and environmental impacts of the areas affected by the 2020 conflict is currently available through ACAPS (ACAPS 2020). This report also discusses the increase in fires in 2020 and the subsequent deterioration of air quality but does not investigate the ongoing changes in land cover in the region (ACAPS 2020). The fighting and political instability during October and November 2020 also appears to have led to an 8- and 30-fold increase in the number of cases of SARS-CoV-2 infections in both Armenia and Azerbaijan, respectively, compared to before the conflict (Kazaryan et al. 2020). Case numbers in Nagorno-Karabakh are hard to determine, but news outlets reported dramatic increases in Stepanakert and Shusha during the war as residents were forced to live in close quarters and underground shelters (CTV News 2020; AP News 2020). After 30 years, conflict continues and residents are still susceptible to political, economic, and social instability, as well as new disease outbreaks.

The process of rebordering that began with the Karabakh conflict is still unresolved. Armenia and Azerbaijan both maintain “rival topographies” with place names, landmarks and administrative boundaries rendered differently in Armenian and Azerbaijani maps (Broers 2021: xiii). This cartographic uncertainty even extends to the digital realm. In December 2021 the Government of Azerbaijan formally petitioned Google maps to remove Armenian place names from the Karabakh region (Asbarez 2021). Several mapping apps continue to show different place names depending on the viewer’s location. As one recent report described,

Yandex Maps shows a road neatly aligned to the international border, while Google puts long stretches squarely in Azerbaijani territory. The community-derived Open Street Map also shows the road crossing into Azerbaijan but not quite in the same way that Google does, placing a bit more of the curving road north of Shurnukh on the Azerbaijani side. None of these platforms have access to special information on the border demarcation process, but all of them do a good job at creating the illusion of geographic authority. (McGlynn 2021)

This has created problems for travelers who rely on centralized internet sources such as GoogleMaps for navigating the region only to discover that the maps can be quite different depending on where they are accessed (McGlynn 2021). In December 2021 the Government of Azerbaijan formally requested Google to remove Armenian place names from Karabakh maps (Asbarez.com 2021). All this geopolitical and cartographic uncertainty has made Karabakh tricky to navigate and much of the region remains underpopulated.