How to Describe and Evaluate Hotspots of Mortality
Most methods developed to identify mortality hotspots have been developed for
roads (Gomes et al. 2009; Malo et al. 2004). One of the methods includes the
assumption that the expected number of kills per segment (km) follows a Poisson
distribution, and hotspots correspond to segments where the number of casualties is
higher than the upper 95% confidence limit of the mean (Malo et al. 2004; Santos
et al. 2015). The kernel method, identifies clusters of casualties by using a moving
function to weight points of mortality within the influence of the function by their
proximity to the location where density is being calculated (Ramp et al. 2005). The
nearest neighbour hierarchical clustering identifies groups of points based on
“nearest-neighbor-method” criteria (Gomes et al. 2009). Finally, the Getis-Ord Gi*
statistic identifies hotspots by adding the number of casualties associated with a
given segment of the road to the casualties of its neighboring segments, and
compares that value with an overall expected distribution (Garrah et al. 2015).
Integrating the information on rail sectors with high mortality rates with GIS
mapping tools leads to the identification of the locations where mitigation measures
(drift fences, rail passages, traffic regulation, etc.) should be applied (Costa et al.
2015). After obtaining the location of mortality hotspots, it is essential to verify the
accuracy of the estimates, as overestimation can result in false hotspots (i.e., areas
wrongly identified as having high mortality rates; Santos et al. 2015).
Sources of Bias in Wildlife Mortality Estimates
The reduction of mortality is one of the main aims of mitigation measures, hence
high mortality rates and their locations need to be as accurate as possible (Guinard
et al. 2012). As stated, even in systematic surveys, the number of carcasses found
largely underestimates mortality because (1) only a subset of animals killed stay
within the area that is searched by the observer; (2) many carcasses are removed by
scavengers, or decomposed until the survey occurs; and (3) some dead animals
remain undetected because of the observers failure (Korner-Nievergelt et al. 2015).
Thus, to obtain unbiased estimates, the numbers obtained during surveys should be
corrected by taking into account: (1) the proportion of animals killed in the area
searched; (2) carcass persistence probability; and (3) searcher efficiency. In order to
take into account these sources of bias, several mortality estimators were developed
in wind farm studies (Korner-Nievergelt et al. 2015) and some have been applied to
road casualties (Gerow et al. 2010; Teixeira et al. 2013), but rarely to railways,
where these issues also apply.
The proportion of casualties (dead and/or injured that have not moved away)
recorded at the search area can be obtained from the size and spatial distribution of
the total area that can be searched, the spatial distribution of the carcasses, and the
proportion of injured animals (still alive) that manage to move away. However, the
3 Methods to Monitor and Mitigate Wildlife Mortality in Railways
29
Most methods developed to identify mortality hotspots have been developed for
roads (Gomes et al. 2009; Malo et al. 2004). One of the methods includes the
assumption that the expected number of kills per segment (km) follows a Poisson
distribution, and hotspots correspond to segments where the number of casualties is
higher than the upper 95% confidence limit of the mean (Malo et al. 2004; Santos
et al. 2015). The kernel method, identifies clusters of casualties by using a moving
function to weight points of mortality within the influence of the function by their
proximity to the location where density is being calculated (Ramp et al. 2005). The
nearest neighbour hierarchical clustering identifies groups of points based on
“nearest-neighbor-method” criteria (Gomes et al. 2009). Finally, the Getis-Ord Gi*
statistic identifies hotspots by adding the number of casualties associated with a
given segment of the road to the casualties of its neighboring segments, and
compares that value with an overall expected distribution (Garrah et al. 2015).
Integrating the information on rail sectors with high mortality rates with GIS
mapping tools leads to the identification of the locations where mitigation measures
(drift fences, rail passages, traffic regulation, etc.) should be applied (Costa et al.
2015). After obtaining the location of mortality hotspots, it is essential to verify the
accuracy of the estimates, as overestimation can result in false hotspots (i.e., areas
wrongly identified as having high mortality rates; Santos et al. 2015).
Sources of Bias in Wildlife Mortality Estimates
The reduction of mortality is one of the main aims of mitigation measures, hence
high mortality rates and their locations need to be as accurate as possible (Guinard
et al. 2012). As stated, even in systematic surveys, the number of carcasses found
largely underestimates mortality because (1) only a subset of animals killed stay
within the area that is searched by the observer; (2) many carcasses are removed by
scavengers, or decomposed until the survey occurs; and (3) some dead animals
remain undetected because of the observers failure (Korner-Nievergelt et al. 2015).
Thus, to obtain unbiased estimates, the numbers obtained during surveys should be
corrected by taking into account: (1) the proportion of animals killed in the area
searched; (2) carcass persistence probability; and (3) searcher efficiency. In order to
take into account these sources of bias, several mortality estimators were developed
in wind farm studies (Korner-Nievergelt et al. 2015) and some have been applied to
road casualties (Gerow et al. 2010; Teixeira et al. 2013), but rarely to railways,
where these issues also apply.
The proportion of casualties (dead and/or injured that have not moved away)
recorded at the search area can be obtained from the size and spatial distribution of
the total area that can be searched, the spatial distribution of the carcasses, and the
proportion of injured animals (still alive) that manage to move away. However, the
3 Methods to Monitor and Mitigate Wildlife Mortality in Railways
29