283
here are from land to water: riparian forests with shrubs and trees, emergent reeds
and sedges, floating macrophytes, submerged macrophytes and macroalgae, and
phytoplankton. Differentiating among these functional types with RS is achievable,
but species-level detection within each community is more difficult due to similar
survival strategies. Each functional type has its own challenges regarding species
detection.
Many studies have successfully mapped IAS in aquatic environments using
direct detection. Depending on the objectives of the study and the functional type
being examined, spectral, spatial, and temporal requirements vary. In simple systems, high spatial resolution aerial photos can often be used to map species functional types as well as single species by taking advantage of unique attributes or
phenology (Marshall and Lee 1994; Everitt et al. 1999, 2003). Mapping multiple
species within the same functional types has been less successful using aerial photos. In these situations, more spectral information is needed to differentiate at the
species level due to varying community complexity and species attributes (e.g.,
Khanna et al. 2011). Multispectral data have also been used with varying levels of
success to map IAS in simple systems such as lakes invaded by just one species
(essentially a two-class system; Venugopal 2002) or lakes with floating and some
submerged vegetation (a three-class system; Everitt et al. 2003; Verma et al. 2003;
Albright et al. 2004). Many classification methods have been used within aquatic
ecosystems with varying degrees of success, including unsupervised classifiers,
such as k-means and ISODATA (Ackleson and Klemas 1987; Dogan et al. 2009)
and simple supervised classifiers, such as maximum likelihood and minimum distance (Malthus and George 1997; Vis et al. 2003; Nelson et al. 2006; Jollineau and
Howarth 2008; Phinn et al. 2008; Yuan and Zhang 2008; Dogan et al. 2009), as well
as more advanced machine learning methods (Malthus and George 1997; Nelson
et al. 2006; Hestir et al. 2008, 2012; Everitt et al. 2011; Santos et al. 2012, 2016).
While some studies have been successful and have even been operationalized into
routine monitoring for invasive species management and reporting (sensu Santos
et al. 2009; Santos et al. 2016), in many studies it is difficult to judge classification
efficacy because accuracy assessment is missing or unusual, often not having independent validation data. Overall, machine learning algorithms seemed to have performed best. Within functional types, some specific strategies seem to work best as
well. We highlight these below.
12.2.3.1 Riparian
Riparian plants are often more difficult to differentiate at the species level than emergent and floating plants due to higher number of species and life forms, and a complex canopy structure, similar to forest IAS detection. Riparian IAS sometimes grow
in monocultures, which may be easier to detect (e.g., giant reed, Arundo donax).
Other IAS can grow embedded in the native community similar to grasslands, making them harder to map using RS (e.g., yellow star-thistle, Centaurea solstitialis).
12 Remote Detection of Invasive Alien Species
here are from land to water: riparian forests with shrubs and trees, emergent reeds
and sedges, floating macrophytes, submerged macrophytes and macroalgae, and
phytoplankton. Differentiating among these functional types with RS is achievable,
but species-level detection within each community is more difficult due to similar
survival strategies. Each functional type has its own challenges regarding species
detection.
Many studies have successfully mapped IAS in aquatic environments using
direct detection. Depending on the objectives of the study and the functional type
being examined, spectral, spatial, and temporal requirements vary. In simple systems, high spatial resolution aerial photos can often be used to map species functional types as well as single species by taking advantage of unique attributes or
phenology (Marshall and Lee 1994; Everitt et al. 1999, 2003). Mapping multiple
species within the same functional types has been less successful using aerial photos. In these situations, more spectral information is needed to differentiate at the
species level due to varying community complexity and species attributes (e.g.,
Khanna et al. 2011). Multispectral data have also been used with varying levels of
success to map IAS in simple systems such as lakes invaded by just one species
(essentially a two-class system; Venugopal 2002) or lakes with floating and some
submerged vegetation (a three-class system; Everitt et al. 2003; Verma et al. 2003;
Albright et al. 2004). Many classification methods have been used within aquatic
ecosystems with varying degrees of success, including unsupervised classifiers,
such as k-means and ISODATA (Ackleson and Klemas 1987; Dogan et al. 2009)
and simple supervised classifiers, such as maximum likelihood and minimum distance (Malthus and George 1997; Vis et al. 2003; Nelson et al. 2006; Jollineau and
Howarth 2008; Phinn et al. 2008; Yuan and Zhang 2008; Dogan et al. 2009), as well
as more advanced machine learning methods (Malthus and George 1997; Nelson
et al. 2006; Hestir et al. 2008, 2012; Everitt et al. 2011; Santos et al. 2012, 2016).
While some studies have been successful and have even been operationalized into
routine monitoring for invasive species management and reporting (sensu Santos
et al. 2009; Santos et al. 2016), in many studies it is difficult to judge classification
efficacy because accuracy assessment is missing or unusual, often not having independent validation data. Overall, machine learning algorithms seemed to have performed best. Within functional types, some specific strategies seem to work best as
well. We highlight these below.
12.2.3.1 Riparian
Riparian plants are often more difficult to differentiate at the species level than emergent and floating plants due to higher number of species and life forms, and a complex canopy structure, similar to forest IAS detection. Riparian IAS sometimes grow
in monocultures, which may be easier to detect (e.g., giant reed, Arundo donax).
Other IAS can grow embedded in the native community similar to grasslands, making them harder to map using RS (e.g., yellow star-thistle, Centaurea solstitialis).
12 Remote Detection of Invasive Alien Species