279
invading trees (Bradley 2014). However, this can be very time-consuming, costly,
and perhaps less systematic and viable across large areas or for early detection.
There are several analysis considerations for mapping tree IAS. A first, and perhaps most important, aspect is that tree IAS detection is prone to higher classification error (Bradley 2014) than other classifications, given the similarity in the
spectral characteristics of trees to each other relative to other plant functional types.
Spectral similarity between invasive and native trees may influence accuracy (Lass
and Prather 2004), so ensemble classifications are recommended as well as other
approaches that maximize spectral differences such as taking into account phenology. The examples cited above illustrate the value of a good field sampling design
(Ramsey III et al. 2002) that covers the diversity of canopy structures (Hudak and
Wessman 1998) and community compositions within the area of interest, since heterogeneity affects overall classification accuracy. In all of the studies discussed
here, we observed a trade-off between omission and commission errors, where classification accuracy seems to be positively correlated with commission errors. Thus,
we recommend that several accuracy metrics should be reported rather than just
overall accuracy to give a better understanding of which species contribute to commission errors and which areas are more uncertain in IAS distribution maps.
12.2.2 Rangelands and Grasslands
Grasslands cover approximately one-third of the Earth’s surface (Latham et al.
2014), account for at least 30% of primary production by terrestrial vegetation
(Grace et al. 2006), and, after forests, are the largest terrestrial carbon sinks
(Anderson 1991; Derner and Schuman 2007; Grace et al. 2006). There are two main
classes of grasslands, tropical/subtropical (also known as savanna) and temperate,
which can further be described by three different subclasses: human generated,
highly managed natural, and rangelands (Ali et al. 2016). Regardless of classification, these regions serve as a major source of animal feed and are heavily influenced
by changes in climate and fire dynamics (D’Antonio and Vitousek 1992; Brooks
et al. 2004). Contrary to popular belief, grasslands and rangelands harbor large
amounts of biodiversity (Murphy et al. 2016); however, they are under threat as IAS
continue to invade. This threatens biodiversity not only through direct losses by IAS
replacing native grasses but also through indirect impacts to ecosystems by changing fire regimes (D’Antonio and Vitousek 1992; Balch et al. 2013), supporting wind
erosion (Weisberg et al. 2017), and serving as a facilitator for plant viruses (Ingwell
and Bosque-Pérez 2015).
IAS in grasslands may be monitored directly or indirectly because not all species
or all grassland ecosystems are good candidates for RS measurements. IAS in grasslands can be difficult to monitor. They are often indistinguishable from native plants
due to spectral similarities or the nature in which they grow—in small patches,
mixed with native vegetation (Shafii et al. 2004). Often indirect methods are most
appropriate because they do not rely solely on discrimination between similar
12 Remote Detection of Invasive Alien Species
invading trees (Bradley 2014). However, this can be very time-consuming, costly,
and perhaps less systematic and viable across large areas or for early detection.
There are several analysis considerations for mapping tree IAS. A first, and perhaps most important, aspect is that tree IAS detection is prone to higher classification error (Bradley 2014) than other classifications, given the similarity in the
spectral characteristics of trees to each other relative to other plant functional types.
Spectral similarity between invasive and native trees may influence accuracy (Lass
and Prather 2004), so ensemble classifications are recommended as well as other
approaches that maximize spectral differences such as taking into account phenology. The examples cited above illustrate the value of a good field sampling design
(Ramsey III et al. 2002) that covers the diversity of canopy structures (Hudak and
Wessman 1998) and community compositions within the area of interest, since heterogeneity affects overall classification accuracy. In all of the studies discussed
here, we observed a trade-off between omission and commission errors, where classification accuracy seems to be positively correlated with commission errors. Thus,
we recommend that several accuracy metrics should be reported rather than just
overall accuracy to give a better understanding of which species contribute to commission errors and which areas are more uncertain in IAS distribution maps.
12.2.2 Rangelands and Grasslands
Grasslands cover approximately one-third of the Earth’s surface (Latham et al.
2014), account for at least 30% of primary production by terrestrial vegetation
(Grace et al. 2006), and, after forests, are the largest terrestrial carbon sinks
(Anderson 1991; Derner and Schuman 2007; Grace et al. 2006). There are two main
classes of grasslands, tropical/subtropical (also known as savanna) and temperate,
which can further be described by three different subclasses: human generated,
highly managed natural, and rangelands (Ali et al. 2016). Regardless of classification, these regions serve as a major source of animal feed and are heavily influenced
by changes in climate and fire dynamics (D’Antonio and Vitousek 1992; Brooks
et al. 2004). Contrary to popular belief, grasslands and rangelands harbor large
amounts of biodiversity (Murphy et al. 2016); however, they are under threat as IAS
continue to invade. This threatens biodiversity not only through direct losses by IAS
replacing native grasses but also through indirect impacts to ecosystems by changing fire regimes (D’Antonio and Vitousek 1992; Balch et al. 2013), supporting wind
erosion (Weisberg et al. 2017), and serving as a facilitator for plant viruses (Ingwell
and Bosque-Pérez 2015).
IAS in grasslands may be monitored directly or indirectly because not all species
or all grassland ecosystems are good candidates for RS measurements. IAS in grasslands can be difficult to monitor. They are often indistinguishable from native plants
due to spectral similarities or the nature in which they grow—in small patches,
mixed with native vegetation (Shafii et al. 2004). Often indirect methods are most
appropriate because they do not rely solely on discrimination between similar
12 Remote Detection of Invasive Alien Species