Habitat Connectivity
Back to the Monitoring Hub
Increase connectivity of woodland across the landscape
Data collection in progress
Greater Côa Valley
Portugal
Problem: Native woodlands in the Greater Côa Valley have been substantially affected by human activity, in particular conversion to agriculture and more frequent wildfires. This has increased fragmentation and reduced woodland connectivity, limiting movement and dispersal of animals across the landscape.
Action: Rewilding Portugal is strategically identifying areas of land to purchase or bring under management for ecological restoration and rewilding.
Indicator: More spatially connected patches of woodland, which act as stepping stones of good quality habitat for wildlife, indicate increased connectivity for species such as lynx and roe deer.
Methods
Connectivity was assessed across a corridor of land that runs north-south in the Greater Côa Valley along the Côa and Águeda Rivers, connecting the Douro Internacional Park in the north with Malcata Nature Reserve in the south (Fig. 1).
Figure 1: Map of the Côa corridor.
A land cover map with spatial resolution of 30 m was created using image data from the Sentinel 2 satellite for the four seasons between August 2020 and May 2021. Images with no or minimal cloud coverage (< 20%) were used. The EUNIS Habitat Classification from the European Environment Agency was used as a reference to classify land cover into seven types (Table 1). Vegetation indexes (NDVI, GNDVI, and SAVI) were calculated to facilitate the identification and classification of each land cover type. Classification was carried out using CART in Google Earth Engine and a training set of points was created to test the model.
Following Simonson et al. (2018), Morphological Spatial Pattern Analysis (MSPA) using Guidos Toolbox was used to assess woodland connectivity. MSPA describes the geometry and connectivity of habitat patches, detecting and quantifying connecting pathways and isolated areas. Components are classified into classes such as Core, Edge, Bridge and Branch, depending how they relate to other areas of surrounding target habitat. The MSPA connectivity analysis was then repeated at a finer scale (based on the same Sentinel satellite images) at nine specific sites: five core areas that Rewilding Portugal had previously identified as a priority for habitat restoration based on local knowledge of the landscape; and four rewilding sites where ecosystem restoration by Rewilding Portugal was underway at the time of the baseline assessment.
This work forms a baseline connectivity assessment. Long-term changes in woodland connectivity will be assessed by revisiting this analysis in future, once sufficient time has elapsed to allow woodland regeneration.
Grassland habitat. Photo: Rewilding Portugal.
Scrub habitat. Photo: Rewilding Portugal.
Results
The initial landcover classification, using the seven EUNIS habitat categories, revealed that most of the Côa corridor is covered by a mosaic of natural scrubland, woodland and grassland habitats (Fig. 2).
Figure 2: Landcover in the Côa corridor, according to EUNIS landcover categories.
The analysis then focused only on the areas of land classified as woodlands, which showed that most woodlands are concentrated near the Côa and Águeda Rivers (Fig.s 2-3). Connectivity was assessed between Rewilding Portugal’s five core areas (turquoise open polygons in Fig. 3) and also the four rewilding sites (blue, purpose, pink and yellow open polygons in Fig. 3). In general, connectivity of woodlands between Rewilding Portugal’s core areas was good, although there were some gaps towards the south of the Côa corridor, in particular in the area linking the core areas and Malcata Nature Reserve (the large area of relatively continuous woodland at the south end of the map). The landcover maps were used to identify three areas where woodland restoration could be targeted to improve north-south and east-west connectivity.
Figure 3. Connectivity map for woodland in the Côa corridor based on results of MSPA analysis, showing Rewilding Portugal’s five core areas (turquoise outline), four rewilding sites (Vale Carapito, Ermo das Aguias, Quinta de Santa Magarida, and Quinta da Azilheira), and the components identified in the MSPA analysis.
Interpretation
The results of this assessment will help guide strategic decision making on which areas of land should be prioritised for restoration in future. Although there were some inaccuracies in landcover classification, the landcover and connectivity maps produced were useful for better understanding of connectivity among woodland patches in the Greater Côa Valley at a landscape scale. The results show that the Côa corridor is mainly composed of continuous natural habitats that are likely to be permeable for animal movements. Woodlands are mainly concentrated along the Côa river and the project core areas are stepping stones that are mostly connected by woodland bridges.
Human activities and wildfire have affected most of the woodland in the Côa Valley, with the result that most of the forest is relatively young. As well as promoting better connectivity and landscape integrity in the Côa Valley, improving woodland connectivity, complexity and functionality could provide habitat and nesting sites for important species like Bonelli’s Eagle, black stork, cinereous vultures and Imperial Eagle.
Woodland recovery is a slow process that will take decades to achieve. Therefore, this assessment will be repeated in the future, to evaluate the long-term effects of restoration actions on woodland connectivity in the landscape.
Lessons learned
The landcover map created had a good resolution for landscape-scale analyses but its low accuracy limited site-level analysis. For example, inaccuracies were seen in the southeast of the corridor, where sparse vegetation was classified as agricultural land. Similarly, woodland cover in and around Quinta de Santa Margarida rewilding area was substantially underestimated in the landcover map and consequent connectivity analysis, meaning connectivity in this area is likely to be higher than shown in Fig. 3. Using more training data, classified by people with knowledge of the local ecosystem, could help improve the model in the future.
In addition, there were insufficient data to further disaggregate the seven initial landcover types, meaning that native woodland could not be separated from other forested areas like plantations. This was a significant limitation of this connectivity analysis, as all woodland is currently classified together, despite some parts containing commercial non-native conifer plantations that may actually act as a barrier for some species. Additionally, more advanced image analyses and more ground data are needed to separate a greater number of landcover classes.
Citations
Simonson, W. D., Allen, H. D., Parham, E., de Basto e Santos, E., & Hotham, P. (2018). Modelling biodiversity trends in the montado (wood pasture) landscapes of the Alentejo, Portugal. Landscape Ecology, 33(5), 811-827.
Partners
