29
Use More Widely Available Image Data
User requirements in vegetation remote sensing can be diverse and demanding. As a
result, there is a tendency to use state-of-the-art image data (hyperspectral and/or
very high spatial resolution) to fulfil requirements as much as possible. Although this
has its scientific merit, obtaining good quality data of such types can be quite cumbersome: data providers are few, demand is high and data are costly. Less expensive
images may be purchased from the archive, but this gives less control over timing of
imagery. Moreover, clouds and year-to-year phenological variations may make it
virtually impossible to obtain truly comparable image pairs of different years suitable for change detection analysis. At the same time, there is a range of cheaper or
even free data products whose potential has not been fully tested, at least not for
detailed Natura 2000 habitat conservation: Landsat, Aster and, since 2015, Sentinel-2
all provide multispectral data of a somewhat lower spatial resolution (10–60 m) than
is mostly desired but it is hard to imagine that this excludes any application on Natura
2000 habitats, even at a local level (e.g., Feilhauer et al. 2014). Moreover, many
countries routinely acquire aerial orthophotos of very high spatial resolution (<2 m
pixel resolution), which are sometimes available for free to government agencies and
NGOs, and which could be combined with satellite data. Even with these free data
sources, acquiring the perfect data set (e.g., in terms of timing, cloud cover etc.) may
still be a challenge but at least these data have the advantage of not consuming substantive components of the available budget before the work even starts.
Limit the Dependence on Ground Reference Data
Obtaining good reference data (i.e., the right variables, collected at the right time, in
the right way and the right format, and in sufficient amount) is often a problem, both
for training and for validation. Existing data may turn out to be (wholly or partly) of
limited use for various reasons. For example, they may be too old, do not cover the
core/entire area or are inconsistent in content and coverage. Collecting new data
may be too expensive or impractical because of, for example, inaccessible terrain
and sub-optimal seasons. Generally, it is advisable to rely on easily accessible reference data (e.g., from online map services like Google Earth) or limit the reference
data dependency altogether to enhance transferability. The best way to achieve this
is probably to reduce the complexity of the reference dataset as much as possible, to
a level where it balances fitness for purpose with technical feasibility. Ontologybased methods (e.g., Nieland et al. 2015) may also boost the chances of successfully
re-using existing, non-tailor-made datasets. However, more research is needed to
ensure these methods are applicable for non-experts.
Towards a Mature Age of Remote Sensing for Natura 2000 Habitat Conservation: Poor…
Use More Widely Available Image Data
User requirements in vegetation remote sensing can be diverse and demanding. As a
result, there is a tendency to use state-of-the-art image data (hyperspectral and/or
very high spatial resolution) to fulfil requirements as much as possible. Although this
has its scientific merit, obtaining good quality data of such types can be quite cumbersome: data providers are few, demand is high and data are costly. Less expensive
images may be purchased from the archive, but this gives less control over timing of
imagery. Moreover, clouds and year-to-year phenological variations may make it
virtually impossible to obtain truly comparable image pairs of different years suitable for change detection analysis. At the same time, there is a range of cheaper or
even free data products whose potential has not been fully tested, at least not for
detailed Natura 2000 habitat conservation: Landsat, Aster and, since 2015, Sentinel-2
all provide multispectral data of a somewhat lower spatial resolution (10–60 m) than
is mostly desired but it is hard to imagine that this excludes any application on Natura
2000 habitats, even at a local level (e.g., Feilhauer et al. 2014). Moreover, many
countries routinely acquire aerial orthophotos of very high spatial resolution (<2 m
pixel resolution), which are sometimes available for free to government agencies and
NGOs, and which could be combined with satellite data. Even with these free data
sources, acquiring the perfect data set (e.g., in terms of timing, cloud cover etc.) may
still be a challenge but at least these data have the advantage of not consuming substantive components of the available budget before the work even starts.
Limit the Dependence on Ground Reference Data
Obtaining good reference data (i.e., the right variables, collected at the right time, in
the right way and the right format, and in sufficient amount) is often a problem, both
for training and for validation. Existing data may turn out to be (wholly or partly) of
limited use for various reasons. For example, they may be too old, do not cover the
core/entire area or are inconsistent in content and coverage. Collecting new data
may be too expensive or impractical because of, for example, inaccessible terrain
and sub-optimal seasons. Generally, it is advisable to rely on easily accessible reference data (e.g., from online map services like Google Earth) or limit the reference
data dependency altogether to enhance transferability. The best way to achieve this
is probably to reduce the complexity of the reference dataset as much as possible, to
a level where it balances fitness for purpose with technical feasibility. Ontologybased methods (e.g., Nieland et al. 2015) may also boost the chances of successfully
re-using existing, non-tailor-made datasets. However, more research is needed to
ensure these methods are applicable for non-experts.
Towards a Mature Age of Remote Sensing for Natura 2000 Habitat Conservation: Poor…