316
and threats and allow targeting of effort, particularly where unexpected changes are
detected. Of importance is that these data provide a historical record of the states
and dynamics of habitats over areas that are far larger than those that can be surveyed from the ground alone. So, the overall message is to find ways to integrate
both approaches, where appropriate, to maximize and optimize the provision of
information in support of nature conservation.
A common approach to demonstrate and convey the benefits of advances in Earth
observation for nature conservation has been to undertake joint studies between
scientists and conservation practitioners, focusing on one or only a few sites. Whilst
successes have been obtained, a better approach would be to develop Earth observation techniques that are more widely applicable beyond these sites or support
national or international efforts. An example of this is the EODESM system (Lucas
and Mitchell (2017), this book), which provides scalable classifications across protected areas using the consistent and globally recognized FAO LCCS taxonomy
and, at the same time, attributes classes with universally recognized environmental
variables. However, when encouraging the use of a specific taxonomy, it is important to convey or demonstrate also how this can be translated to other taxonomies
that may be more familiar or relevant. Examples might be the Ramsar Classification
System for Wetlands, which can be reproduced from the LCCS but requires additional information on context (e.g., marine, estuarine, tundra, alpine). Recognition
of taxonomies used commonly by ecologists is also important when developing or
designing approaches to classification from Earth observation data.
A further criticism of Earth observation products is that these are not fit for purpose or are of insufficient accuracy. As an example, grassland maps may not consider soil acidity measures in their taxonomy. The accuracy of grassland maps may
also be stated as exceeding a value that is often considered acceptable (80% correct
80% of the time), but that of grassland habitats of conservation importance may be
far lower. Hence, there is a need for remote sensing scientists to consider the characteristics of these more specific and critical habitats and how these might be better
discriminated, described and mapped; and also provide realistic measures of accuracy. Perhaps more importantly, there needs to be a dialogue between remote sensing and conservation practitioners in terms of the taxonomies used for classification
to ensure that these meet the needs on the ground, with this occurring before the
commencement of projects. Communication of ideas and methods and collaborations between all parties is essential to ensure that funding and resources are effectively and efficiently utilized.
When more detailed classifications are generated for sites, acknowledgement of
both the actual and potential presence of plant species is needed. As an example, the
training of a classification of Australian open woodland tree species from airborne
hyperspectral data relied on a set of tree species defined through ground observations (Lucas et al. 2008). However, when conducting an independent ground-based
validation of the classifications in other areas, rare tree species that had not been
previously observed in the area were found. These were therefore not mapped as
their spectral characteristics and statistics were not integrated in the training dataset.
This is also the case where plant species are associated with habitat types or strucR. Lucas et al.
and threats and allow targeting of effort, particularly where unexpected changes are
detected. Of importance is that these data provide a historical record of the states
and dynamics of habitats over areas that are far larger than those that can be surveyed from the ground alone. So, the overall message is to find ways to integrate
both approaches, where appropriate, to maximize and optimize the provision of
information in support of nature conservation.
A common approach to demonstrate and convey the benefits of advances in Earth
observation for nature conservation has been to undertake joint studies between
scientists and conservation practitioners, focusing on one or only a few sites. Whilst
successes have been obtained, a better approach would be to develop Earth observation techniques that are more widely applicable beyond these sites or support
national or international efforts. An example of this is the EODESM system (Lucas
and Mitchell (2017), this book), which provides scalable classifications across protected areas using the consistent and globally recognized FAO LCCS taxonomy
and, at the same time, attributes classes with universally recognized environmental
variables. However, when encouraging the use of a specific taxonomy, it is important to convey or demonstrate also how this can be translated to other taxonomies
that may be more familiar or relevant. Examples might be the Ramsar Classification
System for Wetlands, which can be reproduced from the LCCS but requires additional information on context (e.g., marine, estuarine, tundra, alpine). Recognition
of taxonomies used commonly by ecologists is also important when developing or
designing approaches to classification from Earth observation data.
A further criticism of Earth observation products is that these are not fit for purpose or are of insufficient accuracy. As an example, grassland maps may not consider soil acidity measures in their taxonomy. The accuracy of grassland maps may
also be stated as exceeding a value that is often considered acceptable (80% correct
80% of the time), but that of grassland habitats of conservation importance may be
far lower. Hence, there is a need for remote sensing scientists to consider the characteristics of these more specific and critical habitats and how these might be better
discriminated, described and mapped; and also provide realistic measures of accuracy. Perhaps more importantly, there needs to be a dialogue between remote sensing and conservation practitioners in terms of the taxonomies used for classification
to ensure that these meet the needs on the ground, with this occurring before the
commencement of projects. Communication of ideas and methods and collaborations between all parties is essential to ensure that funding and resources are effectively and efficiently utilized.
When more detailed classifications are generated for sites, acknowledgement of
both the actual and potential presence of plant species is needed. As an example, the
training of a classification of Australian open woodland tree species from airborne
hyperspectral data relied on a set of tree species defined through ground observations (Lucas et al. 2008). However, when conducting an independent ground-based
validation of the classifications in other areas, rare tree species that had not been
previously observed in the area were found. These were therefore not mapped as
their spectral characteristics and statistics were not integrated in the training dataset.
This is also the case where plant species are associated with habitat types or strucR. Lucas et al.