Analysis of land cover and landscapes fragmentation based on ESA land cover data, 2022

Supplementary materials will be published a little later

Fragmentation of land cover was estimated for the following three options: 1) all land cover classes fragmented by roads and settlements; 2) all land cover classes fragmented by roads, settlements, and cropland; and 3) tree cover fragmented by all other land cover classes. Lake Sevan, in all cases, was considered as a gap in land cover. For all three options, individual patches were identified, and patch area and perimeter-to-area ratio (Fragstat metric PARA) were calculated using FRAGSTATS 4.3beta [1]. We analyzed the first fragmentation option by provinces and land-scape zones, the second option by landscape zones, and the fragmentation of tree cover by provinces. For large patches divided between provinces and landscapes, we considered parts of them located in the corresponding provinces or landscapes as individual patches. Therefore, the size of most of the largest patches exceeds that counted in provinces and landscape zones.
A single FRAGSTATS calculation takes approximately 8 hours to run on an i5-8260U processor with 16 GB of RAM

[1] McGarigal K, Cushman SA, Neel C, Ene E, 2002, FRAGSTATS: Spatial Pattern Analysis Program for Categorical Maps. Computer software program produced by the authors at the University of Massachusetts, Amherst, viewed June 04, 2024.

1. Fragmentation of land cover of Armenia

For the three fragmentation options studied, the average patch area changes by a factor of tens (Figure 6). The largest average patch area was found when land cover is fragmented by built-up areas and roads (6 km²), and the smallest was found for tree cover fragmentation (0.05 km²). The average PARA changes much less. The lowest PARA value was found for tree cover patches, despite their small average size, which means that woody patches have a smoother boundary than non-woody vegetation.

Figure 6. Average values of fragmentation indicators for the three land cover fragmentation options. The indicators are shown in different scales, indicated on the corresponding axes.

Croplands, identified by the ESA land cover data, fragment the land cover 30-fold in comparison to roads and built-up areas, increasing the number of patches from 4,185 to 119,292 and reducing their average area from 6.1 km² to 0.2 km². The number of the smallest patches increases significantly, while the number of the largest (>1000 km²) and medium (10-100 km²) patches decreases. The total area occupied by the largest patches (>1000 km²) is also reduced (Tables S4, S5 in Supplementary File).
The fragmentation of tree cover is significantly higher than that of the entire natural land cover. The average patch size is 0.05 km² (Table S5 in Supplementary File), which is 3.8 times smaller than that for the entire natural land cover fragmented by roads, built-up areas, and croplands. There are no forest patches larger than 1000 km², and only nine patches are larger than 100 km².
The distribution of patches by area is extremely uneven. Most patches are very small, less than 0.001 km² (Table S4, Figure S2A in Supplementary File). At the same time, there are several large, non-fragmented patches with areas of hundreds or thousands of square kilometers, which account for more than 50% of tree cover and more than 90% of total area of natural land cover classes (Figure 7 and Figure S2B, Table S5 in Supplementary File). Most of the largest patches are located on provincial borders and include a variety of landscapes (Figure S3 in Supplementary File).

Figure 7. Individual patches for the three fragmentation options. (A) All land cover classes fragmented by roads and built-up areas. (B) All land cover classes fragmented by roads, built-up areas, and croplands. (C) Tree cover fragmented by all other land cover classes.

2. Land cover fragmentation in administrative provinces

The degree of provincial fragmentation by roads and built-up areas varies across different indicators and size groups of patches (Figure 8, Tables S6, S7 in Supplementary File).
The average values of fragmentation indicators in provinces correlate with indicators of the degree of anthropogenic transformation of the territory, obtained from ARMSTAT (annually plowed area, population density, rural population density) and from ESA land cover data (cropland and built-up area). However, these dependencies are statistically reliable not for the entire set of patches, but only for patches of a certain size: for average patch size, the dependence is observed for patches ranging from 0.1 km² to 100 km², and for PARA, it is observed for patches larger than 10 km² (Table S8 in Supplementary File). In the size groups of patches that are most “sensitive” to anthropogenic transformation, the least fragmented province according to indicator of patch area is Vayots Dzor, and according to PARA, province Tavush (Figure 8C,D). The number of large patches and the share of provincial area occupied by them are typically lower in provinces where the average patch size is smaller (Tables S9, S10, Figures S4, S5 in Supplementary File).

Figure 8. Indicators of land cover fragmentation by roads and built-up areas in provinces. (A) average patch size for patches of all sizes. (B) average PARA for patches of all sizes. (C) average patch size for patches larger than 10 km². (D) average PARA for patches larger than 100 km².

Tree cover is less fragmented in Tavush, according to the indicators of average patch area, and in Lori, according to PARA (Figure S6, Table S11 in Supplementary File). The number of large forest patches and the share of provincial area occupied by them are typically lower in provinces where the average forest patch size is smaller (Tables S12, S13, Figures S7 – S9 in Supplementary File). There is a strong positive correlation between the average patch size, forest area, and the proportion of forest area in provinces (0.936** and 0.978**, respectively). Thus, all indicators generally follow the tree cover area in the provinces. The pattern of tree cover fragmentation across provinces is significantly different from the fragmentation of the entire natural land cover. We did not find a correlation between the corresponding indicators.

3. Fragmentation of landscape zones

The pattern of relationships between indicators of land cover fragmentation and anthropogenic transformation across landscapes differs from that in provinces. In this analysis, we used indicators of anthropogenic transformation based solely on land cover data (cropland and built-up area), as there is no statistical data for landscape zones. Correlations between the average patch size and anthropogenic indicators are rarely identified, while for PARA, they are statistically significant for patches larger than 1 km² for fragmentation by roads and built-up areas, and for patches larger than 0.1 km² for fragmentation by croplands (Table S14 in Supplementary File).
The analysis of the general set of patches of all sizes shows that when landscapes are fragmented by roads and built-up areas, the largest average patch size was found in the alpine zone, followed by the subalpine and high-altitude zones (Figure S10A, Table S15 in Supplementary File). However, in this case, the average patch size depends not only on anthropogenic fragmentation but also on the total area of the landscape zone. Therefore, the least disturbed high-altitude landscape does not have the largest average patch, since the total area of this zone is small. The same is true for submountain semidesert. According to PARA, high-altitude and alpine zones are the least fragmented (Figure S10D, Table S15 in Supplementary File). Croplands increase the number of the smallest patches (Figure S11, Table S16 in Supplementary File) and reduce the average patch size by 85–99% in all landscapes except for high-altitude snows and sub-mountain semi-desert (Figure S10A–C in Supplementary File). Even in the alpine zone, it decreases by 96%, although croplands occupy only 0.3% of the area. Even this small impact produces a number of very small grassland patches, significantly decreasing the average patch size, and has an even stronger effect on PARA, which in the alpine zone increases almost 600 times and becomes similar to PARA values in most other landscapes (Figure S10D–F in Supplementary File). It is worth mentioning here that land cover data can inaccurately distinguish between croplands and natural grasslands, and the presence of croplands in the alpine zone may be a result of such inaccuracy.
Indicators for patches larger than 1 km², which are more sensitive to anthropogenic impact, give a different picture, which is more like reality. Croplands have almost no effect on average patch size in the alpine zone (Figure 9A–C). In the subalpine zone, the average patch size even increases. This may be explained by the fact that some of the smallest patches between roads and buildings are recognized in the land cover as croplands. The average patch size decreases to the greatest extent in the most transformed landscapes, which are the mountain-valley semi-desert and middle mountain steppe (see Section 3.1, Figure 4). Average PARA is also greatest in these two most transformed zones, and croplands increase PARA values most strongly in them (Figure 9D–F).

Figure 9. Fragmentation indicators for patches larger than 1 km2 in landscape zones. (A) Average patch area due to fragmentation by roads and built-up areas. (B) Average patch area after adding croplands as a fragmenting factor. (C) Reduction in average patch area (%) following the addition of croplands as a fragmenting factor. (D) Average PARA due to fragmentation by roads and built-up areas. (E) Average PARA after adding croplands as a fragmenting factor. (F) Increase in average PARA (%) following the addition of croplands as a fragmenting factor.

The largest average patches are in the subalpine zone, followed by the middle mountain meadow steppe and alpine zone, and only then by high-altitude snows (Figure 9A,B, Table S15 in Supplementary File). However, in this case, the average patch size depends not only on the degree of anthropogenic transformation but also on the total area of the landscape zone. Therefore, the least disturbed high-altitude landscape does not have the largest average patch, since the total area of this zone is small.
The patterns of indicators for the number of patches of different sizes and the area they occupy generally follow PARA and the average patch size (Figures S11, S12, Tables S16, S17 in Supplementary File).
The impact of croplands on landscape fragmentation is more clearly visible for patches larger than 0.01 km² along a hypothetical transect “semi-desert – steppe – mountains – forests” in an approximate direction from southwest to northeast. The following patterns can be discussed (Figure 10):
1) When landscapes are fragmented by roads and built-up areas (pale color shading in panels A and B), the average patch size (panel A) and average PARA (panel B) decrease in all four analyzed size groups of patches as anthropogenic transformation increases (panel C) in the direction from mountains in the SW to semi-deserts and in the NE to forests. However, PARA values decrease with further movement in the SW direction from the steppes to semi-deserts and in the NE from the forest to the forest shelter belt.
2) The addition of croplands as a fragmenting factor has the following effects on the average patch size (lines in the panel A): no effect on small patches >0.01 km² (blue lines); increase for patches larger than 0.1 km² in the subalpine zone (red lines); reduction for patches larger than 1 km² (yellow lines) and for patches larger than 10 km² (green lines).
3) Croplands has the following effects on average PARA values (lines in the panel B): the largest increase even in alpine zone for small patches >0.01 km² (blue lines); for groups of larger patches (>0.1 km², >1 km², and >10 km², red, yellow, and green lines), PARA values begin to increase in the more transformed zones; generally, PARA increases most strongly in landscapes with a high proportion of croplands (middle mountain steppe and mountain-valley semi-desert); PARA does not increase as much in landscapes with smaller proportion of cropland area (dry steppe and forests).

Figure 10. Changes in average values of fragmentation indicators along a hypothetical transect “semi-desert – steppe – mountains – forests” in an approximate direction from southwest to northeast. Changes for different patch size groups are shown in different colors. The pale shading shows the values for fragmentation by roads and built-up areas, and the lines show the values when adding fragmentation by croplands. (A) Changes in average patch size. (B) Changes in average PARA. (C) The share of the area transformed by humans.

In addition to ecosystem rarity and area, important criteria are anthropogenic threats and degradation of ecosystems. To assess these criteria, we used estimates of land cover fragmentation (Section 3.4). The approach of the IUCN Red List of Ecosystems considers ecosystem fragmentation as an important threat, which increases their protection rating [28,29]. However, here we are comparing provinces where ecosystems persist, rather than the ecosystems themselves. Therefore, it can be considered that high average fragmentation in a province reduces its importance for ecosystem conservation. As demonstrated earlier, fragmentation indicators of patches larger than 10 km² correlate with the degree of anthropogenic transformation of provinces (Section 3.4.2). The indicators of rural population density and the proportion of annually plowed land are clearly correlated with each other (0.878**) and as well as with livestock density, with correlation coefficients of 0.936** and 0.975**, accordingly. Thus, fragmentation of patches larger than 10 km² is associated with the whole complex of indicators of threats to ecosystems from agriculture.
The total fragmentation index was calculated as the sum of the normalized average PARA and average patch area indices. PARA was considered a negative index, while the average patch size was considered a positive index. In the provinces of Aragatsotn, Ararat, Armavir, Gegharkunik, Kotayk, and Shirak, the total fragmentation index is negative (Figure 12A, Table S21 in Supplementary File), as relatively small average patch sizes are overshadowed by large PARA values (Figure 12B). In the provinces of Lori, Syunik, Tavush, and Vayots Dzor, the total fragmentation index is positive, as these areas feature larger average patch sizes and lower PARA values. A positive value in this case does not imply that fragmentation has a beneficial effect on ecosystems; rather, it serves only to compare the levels of provincial fragmentation. A comparison of the indices of provincial importance in maintaining ecosystem diversity and fragmentation shows that fragmentation most strongly reduces the importance of Armavir, followed by Kotayk, Ararat, and Gegharkunik. The province of Shirak has the lowest significance according to the criteria analyzed. Relatively low fragmentation makes the provinces of Vayots Dzor, Tavush, and Lori more valuable for ecosystem conservation compared to others (Figure 12B).

Figure 12. Normalized indices of fragmentation and importance in maintaining ecosystem diversity for the provinces of Armenia. (A) Comparison of the degree of provincial fragmentation with their importance in maintaining ecosystem diversity. (B) Comparison of the PARA index with the average patch area.