275
overall accuracy of a classifier compares with expected accuracy, a random
classification of pixels from the data set. One final consideration for accuracy
assessment is the importance of having independent validation data that were not
used in the mapping procedure. If accuracy is assessed with training data, it only
measures how good the classifier is for those specific data that it was trained on,
but the classifier may not be as accurate with other non-training pixels within
the image.
12.2 Invasive Plants in Natural and Agroecosystems
Each ecosystem and IAS combination presents unique challenges for identification
and mapping using RS. This is due to different landscape configurations, community composition, canopy structures, climates, habitat characteristics, and plant phenology. Each of these characteristics can be used to inform the optimal instrumentation
for IAS detection and mapping. For this reason, we have separated IAS detection
methods by biome and then split into more specific ecosystems and case studies.
12.2.1 Forests
Around one-third of Earth’s land surface is covered by forests. Forests are critical
ecosystems, holding a very large proportion of global biodiversity. They are responsible for a large fraction of the global carbon storage and fluxes, strongly influence
local and global water cycle processes, and provide fundamental goods and services
to humanity (Foley et al. 2007). Globally, there are 26 types of forests, from taiga to
tropical, all characterized by the unique ecological adaptations of trees to local climate, geology, and ecological conditions.
Forests invasions come in two types: (i) tree invasions (13 trees are in the top 100
world’s most invasive alien species, Lowe et al. 2000); and (ii) when other plants,
such as vines and shrubs, or animals invade (Resasco et al. 2007; Cheng et al. 2007;
Santos and Whitham 2010). Detection of invasion by tree species requires the direct
detection of tree canopies (e.g., Asner et al. 2008a, b). Invasion of forests by other
plants or animals can be detected directly, for example, when the IAS covers the
canopy (Cheng et al. 2007), or indirectly, by measuring canopy leaf-off (Resasco
et al. 2007; Wilfong et al. 2009), or through detection of pest impacts (Näsi et al.
2015; Ortiz et al. 2013).
Several studies have used optical RS data to directly detect invasion by tree species. One of the earliest approaches performed texture analysis on simulated satellite panchromatic imagery from historical 2 m aerial photography to map the
invasive acacia (Acacia mearnsii) in South Africa (Hudak and Wessman 1998).
Ramsey III et al. (2002) used 0.5 and 1.0 m color-infrared aerial photographs to map
Chinese tallow (Sapium sebiferum) in Louisiana and Texas. They used a k-means
12 Remote Detection of Invasive Alien Species
overall accuracy of a classifier compares with expected accuracy, a random
classification of pixels from the data set. One final consideration for accuracy
assessment is the importance of having independent validation data that were not
used in the mapping procedure. If accuracy is assessed with training data, it only
measures how good the classifier is for those specific data that it was trained on,
but the classifier may not be as accurate with other non-training pixels within
the image.
12.2 Invasive Plants in Natural and Agroecosystems
Each ecosystem and IAS combination presents unique challenges for identification
and mapping using RS. This is due to different landscape configurations, community composition, canopy structures, climates, habitat characteristics, and plant phenology. Each of these characteristics can be used to inform the optimal instrumentation
for IAS detection and mapping. For this reason, we have separated IAS detection
methods by biome and then split into more specific ecosystems and case studies.
12.2.1 Forests
Around one-third of Earth’s land surface is covered by forests. Forests are critical
ecosystems, holding a very large proportion of global biodiversity. They are responsible for a large fraction of the global carbon storage and fluxes, strongly influence
local and global water cycle processes, and provide fundamental goods and services
to humanity (Foley et al. 2007). Globally, there are 26 types of forests, from taiga to
tropical, all characterized by the unique ecological adaptations of trees to local climate, geology, and ecological conditions.
Forests invasions come in two types: (i) tree invasions (13 trees are in the top 100
world’s most invasive alien species, Lowe et al. 2000); and (ii) when other plants,
such as vines and shrubs, or animals invade (Resasco et al. 2007; Cheng et al. 2007;
Santos and Whitham 2010). Detection of invasion by tree species requires the direct
detection of tree canopies (e.g., Asner et al. 2008a, b). Invasion of forests by other
plants or animals can be detected directly, for example, when the IAS covers the
canopy (Cheng et al. 2007), or indirectly, by measuring canopy leaf-off (Resasco
et al. 2007; Wilfong et al. 2009), or through detection of pest impacts (Näsi et al.
2015; Ortiz et al. 2013).
Several studies have used optical RS data to directly detect invasion by tree species. One of the earliest approaches performed texture analysis on simulated satellite panchromatic imagery from historical 2 m aerial photography to map the
invasive acacia (Acacia mearnsii) in South Africa (Hudak and Wessman 1998).
Ramsey III et al. (2002) used 0.5 and 1.0 m color-infrared aerial photographs to map
Chinese tallow (Sapium sebiferum) in Louisiana and Texas. They used a k-means
12 Remote Detection of Invasive Alien Species