-
This
project evaluates the use of airborne remote sensing data acquired by a
Compact Airborne Spectrographic Imager (CASI) and an Airborne Laser Terrain
Mapper (ALTM) in Countryside Survey 2000. The focus is on environmental
monitoring at an extent and scale that is intermediate to the field and
satellite surveys.
-
The
overall project aim was to assess how the airborne sensors may be used in
conjunction with field survey and satellite sensors in future Countryside
Surveys.
-
Pairs
of example survey squares were studied in each of the Arable, Pastural,
Marginal and Upland Landscapes of Britain - as defined in Countryside Survey
1990. Each pair was divided into a trial and a check square, to allow the
development, refinement, and validation of methods and their subsequent
'blind' testing.
-
Analysis
focussed on 1 km squares using airborne digital data acquired during summer
1999. These data posed two sets of limitations. First, the independent
acquisition of the CASI and ALTM data meant that integration was not an
automated process. Second, the affects of atmospheric attenuation, and
differences in viewing geometry and solar illumination angle restricted the
transfer of spectral training data across and between sites.
-
A
data processing flow-line was developed for the four trial squares. For the
ALTM data, pre-processing involved creating a Digital Surface Model from the
point-sample elevation data, and subsequently separating the terrain and
vegetation canopy height information. For the CASI data, pre-processing
involved: image normalisation, geometric correction, flight-line mosaicking,
and spectral segmentation. The integrated CASI and ALTM data were then used
for per-parcel classification and knowledge-based correction.
-
The
final product of classification was a vector data-base in which each parcel
contained information on land-cover, canopy relative height and terrain
context.
-
Many
of the data processing methods developed for the trial squares were semi or
fully automated and were thus directly transferable to the check squares.
These can broadly be considered as operational.
-
Because
the classification of trial squares was restricted to 1 x 1 km areas,
spectral training data were identified for an insufficient number of
land-cover types to represent those present in the check squares. It was
therefore not possible to assess fully the 'blind' classification of check
squares using only the spectral training data from the trial squares.
-
Mapping
CS squares by the 'blind' classification of airborne remotely sensed data
would need libraries of spectral signatures for land-cover types to be
developed. This could only become operational if more complex methods of
image spectral normalisation than used in this project were developed. This
may be addressed in future Research and Development by the EA and NERC
airborne remote sensing facilities.
-
The
data processing flow-line for trial squares was shown to be applicable for
the classification of a 3 x 3 km site, setting the core CS square into a
wider landscape setting.
-
The
product of airborne data classification was not directly comparable with
field survey or Land Cover Map 2000. Each survey approach differed in:
spatial detail; extent of coverage; landscape features and land-cover types
mapped; variables recorded; and cost.
-
For
the farmed Landscape types (Arable, Pastural, Marginal) correspondence of
the classified airborne data with edited field survey data (at Broad Habitat
level) was 80-89% for the trial squares and 69-80% for the check squares.
-
Non-correspondence
related to: differences in the reported size and boundary location of
land-cover parcels; mis-registration between the two data-sets; a greater
subdivision of landscape parcels in the airborne data; distinctions between
land-cover mapped from the airborne images and land-use mapped in the field
survey; and errors in airborne data classification and in the field survey.
-
For
the two Upland squares correspondence between classified airborne data and
field survey was 69-72%. The discrepancies related to registration and
classification errors, and to the classification of a finer spatial mosaic
of habitats in the airborne data than the field survey parcel boundaries
portrayed.
-
Classification
of land-cover in the Arable trial square using 1998 and 1999 CASI data
demonstrated the repeatability of Broad Habitat mapping from airborne
digital data, with an 89% correspondence.
-
The
correspondence between the classified airborne data and LCM2000 varied
between 23% and 74%. Although the process of classifying the airborne
digital data largely followed the methods developed for LCM2000, differences
between the two data products exist (e.g. in the dates and spatial
resolution of the imagery, in the use of generalised soil sensitivity and
drift maps for LCM2000).
-
Neither
LCM2000 nor the classified airborne data represent the absolute ground
truth; both contain errors in land-cover identification, and these had a
direct influence on the correspondence statistics.
-
The
cost of using airborne digital data to supply land-cover information for
individual 1 km CS squares is prohibitive compared with field survey. This
is because the spatial dispersion of CS squares is too great for the
operational logistics of airborne data acquisition to be cost-effective for
1 km squares.
-
A
greater economy of scale exists for airborne survey compared with field
survey when acquiring data for CS squares in their wider landscape context.
The restrictions imposed by the operational logistics of airborne survey,
which prohibit the cost-effective data capture for 1 km squares, are
significantly reduced for 3 x 3 km squares and removed for 5 x 5 km squares.
-
Airborne
data thus offers the potential for mapping land-cover and landscape
3-dimensional structure (at a spatial resolution of 3 m or better), placing
the core CS square into a wider landscape context.
-
Single-date
airborne data cannot, however, supply information on land-use, species
composition, woodland type or age. Airborne survey could not easily give
national coverage (either census or sample) in a single target year.
-
If
the emphasis of future Countryside Surveys remains on data acquisition in a
single year, then the recommended use of integrated airborne data is to
provide landscape-scale information for a sub-sample of key target sites. If
a higher proportion of CS squares are to be surveyed in high spatial detail
by remote sensing, high resolution satellite data may represent an
alternative to airborne digital data.
-
National
coverage of CS squares in their wider landscape context could be achieved by
integrated airborne and field survey on an annual rolling basis.
