101
retrieve biophysical attributes. This type of product can only be used visually and
the boundaries of sites and habitats are usually digitised manually by monitoring
staff. Aerial photography can also be used as a training and validation dataset for
products generated from satellite sensor data, such as land cover and habitat maps.
Field Data
The decision support tools used for vegetation management require accurate information on the spatial array of different plant communities. The use of Global
Positioning Systems (GPS) allows habitat data to be collected with location accuracy of below 1 m, these data can include habitat type, presence of species and
photographs. Ideally, to accommodate errors in satellite image processing at 2 m
resolution and GPS accuracy, habitats need to be homogenous across a 6 m
2
area to
ensure that data associated with the GPS points are truly representative of that specific location. When using coarser pixel resolution data, this area needs to be
increased. However, with rare or sparse habitats, this may not be possible and the
heterogeneous and nature of vegetation (e.g., those occurring in fine-grained mosaics) needs to be at the forefront of any considerations for collecting training points
and validation datasets.
Staff who are undertaking site monitoring routinely georeference ground data
collections using a GPS. However, these point datasets tend to include only habitats
of interest and very rarely include other classes that are in abundance on sites and
these are often ignored (i.e., are not monitored). For a dataset to be suitable for training and validating satellite products, both types of habitat need to be represented.
For this study, those points were collected from the UAV data and aerial photography and combined with field-measured data. To ensure an even spatial distribution
of points into training and accuracy classes, a sampling grid was generated at different resolutions (5 m, 10 m, 15 m, 20 m) and split accordingly. (Fig. 2).
Index Calculation
For characterizing vegetation health and condition, a number of spectral indices are
commonly used. These include (a) the Normalised Difference Vegetation Index
(NDVI), which represents a ratio of productivity from vegetation, (b) the Normalised
Difference Wetness Index (NDWI), which calculates the proportion of water in
leaves and c) the Plant Senescence Reflectance Index (PSRI), which calculates the
proportion of senescent or non-photosynthetic vegetation. The indices that are available are sensor dependent and the Index Database (2016) is a particularly useful
resource for making a selection. For Worldview-2, there are 134 indices that can be
calculated. For this study, two Worldview-2 images were obtained in the peak and
post-flush period, allowing these indices to be calculated and compared over time.
Mapping Coastal Habitats in Wales
retrieve biophysical attributes. This type of product can only be used visually and
the boundaries of sites and habitats are usually digitised manually by monitoring
staff. Aerial photography can also be used as a training and validation dataset for
products generated from satellite sensor data, such as land cover and habitat maps.
Field Data
The decision support tools used for vegetation management require accurate information on the spatial array of different plant communities. The use of Global
Positioning Systems (GPS) allows habitat data to be collected with location accuracy of below 1 m, these data can include habitat type, presence of species and
photographs. Ideally, to accommodate errors in satellite image processing at 2 m
resolution and GPS accuracy, habitats need to be homogenous across a 6 m
2
area to
ensure that data associated with the GPS points are truly representative of that specific location. When using coarser pixel resolution data, this area needs to be
increased. However, with rare or sparse habitats, this may not be possible and the
heterogeneous and nature of vegetation (e.g., those occurring in fine-grained mosaics) needs to be at the forefront of any considerations for collecting training points
and validation datasets.
Staff who are undertaking site monitoring routinely georeference ground data
collections using a GPS. However, these point datasets tend to include only habitats
of interest and very rarely include other classes that are in abundance on sites and
these are often ignored (i.e., are not monitored). For a dataset to be suitable for training and validating satellite products, both types of habitat need to be represented.
For this study, those points were collected from the UAV data and aerial photography and combined with field-measured data. To ensure an even spatial distribution
of points into training and accuracy classes, a sampling grid was generated at different resolutions (5 m, 10 m, 15 m, 20 m) and split accordingly. (Fig. 2).
Index Calculation
For characterizing vegetation health and condition, a number of spectral indices are
commonly used. These include (a) the Normalised Difference Vegetation Index
(NDVI), which represents a ratio of productivity from vegetation, (b) the Normalised
Difference Wetness Index (NDWI), which calculates the proportion of water in
leaves and c) the Plant Senescence Reflectance Index (PSRI), which calculates the
proportion of senescent or non-photosynthetic vegetation. The indices that are available are sensor dependent and the Index Database (2016) is a particularly useful
resource for making a selection. For Worldview-2, there are 134 indices that can be
calculated. For this study, two Worldview-2 images were obtained in the peak and
post-flush period, allowing these indices to be calculated and compared over time.
Mapping Coastal Habitats in Wales