SPIN Internship 2019

Monitoring Heathlands at the RSPB Lodge Reserve

Using hyperspectral imagery to classify vegetation and quantify habitat coverage.

by | Oct 22, 2019 | Environment

Loreena Jaouen

Geospatial Analyst at 2Excel geo

 

Loreena has an MSc in Remote Sensing and Environmental Mapping from University College London (UCL) and a BSc in Natural Sciences (Biology and Physical Geography) from Durham University. She has experience in both optical and SAR remote sensing, including image processing, multi-temporal change detection and image classification. Loreena has a particular interest in conservation, environmental monitoring and landscape management.

 

As part of the 2019 Space Placement in Industry (SPIN) programme, I had the pleasure to intern at 2Excel geo for 8 weeks. During this time, I collaborated with the RSPB on a project that aimed to classify vegetation and quantify habitat coverage and fragmentation using 2Excel geo’s hyperspectral airborne capability.

I met up with Adrian Hughes, Head of GIS Services at the RSPB Lodge Reserve in Sandy, Bedfordshire in early August. We discussed the ways in which remote sensing could help support the RSPB conservation work and explored the potential for airborne monitoring of heathlands. Since 2005, the RSPB has been actively working on heathland restoration in Sandy, which is especially vital to ground-nesting birds such as nightjars, curlews and golden plovers. The Countryside Stewardship (CS) offers funding to land managers who have made environmental improvements to their land. In order to be eligible for funding, an area must meet the requirements set out by the CS, which differ based on habitat type. For management of lowland heathlands, the followings must apply to the landscape:

  • A variety of heather age and structure through management (grazing, burning or cutting)
  • 1-10% bare ground (must be present – valuable for insects and reptiles)
  • Trees and scrubs below 15%
  • Bracken below 10% in dense canopy
  • Full range of age classes of Western or European gorse

Each hectare of lowland heathland that meets the requirements generates £274 of funding, and these requirements provide the baseline for management objectives.

In Sandy, extensive growth of birch and bracken fragments mean the landscape and young trees must be frequently cut down and removed (Fig. 1 and 2). The RSPB is responsible for tracking the restoration progress by submitting a yearly report on heathland condition to Natural England. At present, heather coverage is estimated in situ using line survey techniques, but remote sensing could allow for more efficient and consistent monitoring and provide a reserve-wide map for further fragmentation and connectivity analysis. This could be particularly useful for targeting areas of heavy encroachment to inform reserve management and could also be used in valuing the landscapes.

Figure 1

Birch and bracken fragment in the landscape at one of the RSPB Lodge Reserve heathlands in Sandy, Bedfordshire.

Figure 2

The RSPB actively works to regenerate heathlands. Removing birch trees is one example of their management activities.

Figure 1

Birch and bracken fragment the landscape at one of the RSPB Lodge Reserve heathlands in Sandy, Bedfordshire.

Figure 2

The RSPB actively works to regenerate heathlands. Removing birch trees is one example of their management activities.

Adrian and I agreed that classified maps of heather, birch and bracken would be produced from hyperspectral airborne imagery, as well as Sentinel-2 multispectral satellite imagery for comparison. Sentinel-2 data is free and open-source. It can provide an image of the reserve every 6 days, allowing for regular change detection analysis. The downside is that Sentinel-2 only offers 13 relatively broad spectral channels and the lower 10-30m spatial resolution could be limiting when classifying similar vegetation types.

Airborne imagery of the reserve had been acquired by Steven Case, our collection manager, in June 2018 at an altitude of 1000m. This provided us with an orthophoto of the site, a Digital Surface Model (DSM) and a 186-band hyperspectral image (see previous article). We drove down to the RSPB Lodge Reserve on the 13th of August with the full dual ASD spectrometer kit (Fig. 3). The goal was to record the spectral signatures of heather, birch and bracken for classification of the airborne imagery. 

Figure 3

Our backpack and boom to carry one of the ASD units for taking dual field ASD measurements.

Figure 3

Our backpack and boom to carry one of the ASD units for taking dual field ASD measurements.

This visit gave me the chance to get to grips with the ASD kit and associated fieldwork protocol. I was also quickly introduced to the challenges of dealing with intermittent clouds when characterising surfaces in the field. Several canopy measurements were taken for each species of interest, covering a wide range of ages and densities to produce average reference spectra (Fig. 4).

Figure 4

Average canopy field spectra collected at the RSPB Lodge Reserve in Sandy.

Figure 5

Selected heathlands at the RSPB Lodge Reserve (white outlines).

Figure 4

Average canopy field spectra collected at the RSPB Lodge Reserve in Sandy.

Figure 5

Selected heathlands at the RSPB Lodge Reserve (white outlines).

Two heathland compartments were chosen for the analysis (Fig. 5). The Sentinel-2 imagery was classified on a pixel basis using vegetation indices and a threshold ruleset (Fig. 6). For the airborne imagery, the degree of match between the reference spectra and each pixel was used, followed by aggregation to the object-level (Fig. 7). Coverage percentages and metrics such as the Perimeter-Area Ratio were then calculated to inform on heather quantity and fragmentation.

Figure 6

Final classification of the Sentinel-2 imagery for Heathland 1 (left) and Heathland 2 (right).

Figure 7

Final classification of the airborne imagery for Heathland 1 (left) and Heathland 2 (right).

Figure 6

Final classification of the Sentinel-2 imagery for Heathland 1 (left) and Heathland 2 (right).

Figure 7

Final classification of the airborne imagery for Heathland 1 (left) and Heathland 2 (right).

The outcomes from the maps were compared with the RSPB ground estimates (Fig. 8). The airborne classification performed well with a 95% accuracy in heather identification against 66% for Sentinel-2. As expected, birch and bracken proved more difficult to distinguish in the airborne imagery. Sentinel-2 did not allow for classification of birch and bracken, but green vegetation could be identified to provide some idea of heather coverage and fragmentation.

Figure 8

Bar graphs showing the relative percentage coverage of all classes for Sentinel-2, airborne and ground estimates, as well as heather fragmentation metrics (top: Heathland 1; bottom: Heathland 2).

Figure 8

Bar graphs showing the relative percentage coverage of all classes for Sentinel-2, airborne and ground estimates, as well as heather fragmentation metrics (top: Heathland 1; bottom: Heathland 2).

A poster of the results was presented in early September at the National Centre for Earth Observation (NCEO) annual conference in Nottingham (Fig. 9). This was a good opportunity to promote remote sensing as a tool for conservation and discuss the methods and implications with a wide variety of people across the environmental sector, from academia to industries.  Going forward, this work should prove very useful to the RSPB in their conservation efforts. The methods considered in this project should be transferable to other areas and habitat types, and similar classified maps could provide landowners with the ability to quantify habitat condition with a high degree of accuracy. This may prove beneficial for valuing the environment and assess Net Biodiversity Gain as part of the 25 Year Environment Plan.

Figure 9

Poster presentation of the results at the National Centre for Earth Observation (NCEO) annual conference 2019

Figure 9

Poster presentation of the results at the National Centre for Earth Observation (NCEO) annual conference 2019