Habitat Heterogeneity
Back to the Monitoring Hub
Increase heterogeneity of montado mosaic habitat
Data collection in progress
Greater Côa Valley
Portugal
Problem: A trend of land abandonment and declining agricultural activities over recent decades has resulted in the spread of homogeneous scrubland and forest across large areas of the Greater Côa Valley. This makes the landscape vulnerable to intense, large-scale wildfires, endangering people’s lives and properties, as well as wildlife.
Action: Rewilding Portugal has released free-living herds of Sorraia horses. Through grazing this aims to create more open habitats, such as scattered scrubland and open forest, and increase plant diversity. Establishing a natural grazing regime by these large herbivores can thus increase landscape heterogeneity and mitigate wildfire risks.
Indicator: This indicator measures changes in the heterogeneity of vegetation cover and structure at sites where natural grazing has been introduced, to understand if it results in greater ecological diversity.
Methods
Habitat heterogeneity was assessed at two rewilding sites: Ermo das Águias and Vale Carapito. Both these sites were grazed by cattle prior to the start of the project. 16 Sorraia horses were introduced to Ermo das Águias in 2022. 10 horses were introduced to Vale Carapito in 2021, reduced to 5 since 2022. The whole area of Vale Carapito (approximately 54 ha) was surveyed by unmanned aerial vehicles (UAVs), whereas at Ermo das Águias, due to the larger area of this site (312 ha), six 470 x 150 m plots were surveyed (Fig. 1), although due to data collection issues only results from plots 1-2 and 4-6 are presented.
Figure 1: Natural grazing areas (orange outline) at Vale Carapito and Ermo das Águias, and plots surveyed by UAVs at Ermo das Águias (green); at Vale Carapito the entire natural grazing area was surveyed.
The first survey was carred out in 2020 at Vale Carapito and in 2021 at Ermo das Águias, with images collected at both sites before the introduction of horses. Images were collected from UAVs twice, in spring and summer, using a Mavic Pro 2 UAV with an RGB camera. Raw images were processed using the Structure from Motion approach in Pix4D Mapper 4.7.5, to classify areas of different ground cover types (trees, rocks, bare soil, grass, and shrubs; Fig. 2).
Figure 2: RGB orthomosaic drone image from one of the surveyed plots at Ermo das Águias (above), and the resulting groundcover classification map (below).
Second surveys were carried out in July 2024 using a Mavic 3M UAV, collecting both RGB and multispectral images, and data were again processed using the software Pix4D to classify areas of different ground cover. Vegetation classifications were verified using both a desk-based approach (using orthomosaic and Canopy Height Model UAV data) and using surveys in the field. Accuracy of 2024 classifications was 97%, but no information about the accuracy of the first surveys was available.
The software Fragstats was used to analyse the images once ground cover classification was complete. Shannon’s diversity and evenness of different groundcover types was calculated, along with the number of patches of different types, and patch and edge density (the number of patches or edges in the landscape, divided by the total area of the landscape). In addition, several metrics were calculated for each groundcover type separately: number of patches, patch and edge density, percentage of total landsape, and an Interspersion and Juxtaposition Index (reflecting patch adjacency, where higher values indicate a more equal adjacency of all patch types).
It was not possible to carry out a quantitative comparison of heterogeneity between 2021 and 2024 at either site, because of issues with image classification accuracy from 2021. Therefore, results presented here are based only on visual comparison of classified images. The 2024 images and metrics will be used as a baseline to compare with future results, in order to assess long-term change in habitat structure.
Scrub habitat. Photo: Rewilding Portugal.
A herd of Sorraia horses in the Vale Carapito natural grazing area. Photo: Rewilding Portugal.
Results
Visual comparison of the classified UAV images of Vale Carapito between 2020 and 2024 suggest that the site underwent an overall “greening”, as both the trees and shrubs grew in size (Fig. 3). In general, the area covered by shrubs increased and the area covered by grass decreased. However, this overall change encompassed substantial variation in the change in vegetation across the site. In areas that were heavily used by horses (based on data from radio-collars), trampling limited the growth and spread of shrubs (Fig. 4). However, in areas that were less frequently used by the horses, shrub cover increased significantly (Fig. 4).
Figure 3: Landscape classification results for Vale Carapito in 2020 (left) and 2024 (right).
Figure 4: Comparison of orthomosaics between 2021 and 2024 in Vale Carapito, in areas with low (left) and high (right) horse activity.
At Ermo das Águias, visual comparison of the classified UAV images again showed strong variation between different areas over the period. Among the five plots surveyed, the change in vegetation cover between 2021 and 2024 varied greatly (Fig.s 5a-e). Again, this could often be ascribed to variation in the intensity of horse activity between different areas. In many areas, shrub encroachment increased from 2021-2024, whereas in areas where horse activity was higher, shrub encroachment was limited and more open grassland was retained.
Figure 5a: Landscape classification results for Ermo das Águias plot 1 in 2021 (top) and 2024 (bottom).
Figure 5b: Landscape classification results for Ermo das Águias plot 2 in 2021 (top) and 2024 (bottom).
Figure 5c: Landscape classification results for Ermo das Águias plot 4 in 2021 (top) and 2024 (bottom).
Figure 5d: Landscape classification results for Ermo das Águias plot 5 in 2021 (top) and 2024 (bottom).
Figure 5e: Landscape classification results for Ermo das Águias plot 6 in 2021 (top) and 2024 (bottom).
Summary landscape-scale fragmentation metrics for both areas in 2024 are shown in Table 1. While these metrics were also calculated for 2020/2021, classification accuracy issues meant that they were not considered sufficiently reliable to use to assess change in habitat fragmentation through time. The 2024 results will therefore be used as a baseline to compare against future data to assess long-term change.
Table 1: Landscape-level metrics of Vale Carapito and Ermo das Águias rewilding areas in 2024.
Interpretation
Based on a visual inspection of the images from the UAVs, the main detected vegetation change at both rewilding sites was the natural expansion of both trees and shrubs, especially broom. The effect of the horses was only evident in areas where they had been most active. The presence of horses did not seem to affect shrubs that were already present, but they did limit new growth. It is not yet possible to be sure whether in future years broom will spread to form continuous cover or if the movement of the horses will keep it fragmented. It is therefore crucial to continue monitoring vegetation growth in future years. These results can also be compared to on-the-ground monitoring of the effect of grazing carried out to assess effects on wildfire risk.
Lessons learned
Unfortunately, it was not possible to compare habitat classes and fragmentation statistics between images from 2021 and 2024 due to uncertainty around the accuracy of the 2021 classification. Overall, the quality of the imagery from 2021 was lower than in 2024, as the 2021 images had issues with blurriness, missing data, or low illumination spots, which lowered the classification accuracy. In addition, the 2021 images were based only on RGB bands, which performed poorly at differentiating between similar-looking groundcover types such as broom and moss-covered rocks. The addition of multispectral bands, as well as the addition of more and better representative samples, likely greatly improved classification accuracy in 2024.
The approach used for the 2024 classification was therefore a substantial improvement on the classification used in 2021. However, it not clear whether this method, which used object-based landscape classification and an analysis of its derived fragmentation statistics, is the most effective approach to assess changes in habitat heterogeneity through time. Results obtained were strongly affected both by the classification accuracy and by the classification approach, including the method used for segmenting different patches. Therefore, if this approach is employed to assess change, it is vital to ensure that a) the classification is as accurate as possible and b) classification methods are kept consistent through time.
Partners
