Header

UZH-Logo

Maintenance Infos

Dealing with uncertainty in amphibian and reptile population monitoring for conservation


Cruickshank, Sam S. Dealing with uncertainty in amphibian and reptile population monitoring for conservation. 2017, University of Zurich, Faculty of Science.

Abstract

Successful conservation management is underpinned by a solid understanding of species distributions and population trends, which is used both to identify populations under threat, and to monitor the effectiveness of management actions. However, as humans are imperfect, observation errors are introduced when we monitor variables such as occupancy patterns or abundances. These errors, if left unaccounted for, can bias inference in ways which can be detrimental to species conservation. In this thesis I used simulations and case studies from amphibian and reptile populations in Switzerland to discuss several forms of observation error, illustrated how their presence may lead to bias, and presented methods by which such biases can be avoided or mitigated.
In Chapter Two, I introduced the issue of imperfect detection in species monitoring and explained how occupancy modelling can be used to estimate detection probabilities and thereby accurately assess occupancy rates. In recent years, some authors have questioned the benefit of collecting the extra information necessary to apply this method, and have argued that conservation practise is not improved by accounting for detection. By applying IUCN red-list guidelines to nationwide monitoring data for amphibians within Switzerland, I demonstrated explicitly how failing to account for detection probabilities could lead to inappropriate management decisions being made. 5 of 12 species would have been inappropriately assigned to a higher extinction threat category if detection probabilities were ignored. Using this case study, I highlight that presence-only datasets can only ever be used to calculate maximum possible declines, and are inadequate for estimating the true magnitude of population change. In order to do so, monitoring practise must change to systems in which non-detections as well as species presences are routinely recorded.
In Chapter Three, I expanded upon the challenges of designing monitoring programs such that species absence can be reliably inferred. I compared two different frameworks used to determine how much effort must be invested in surveying a site before it can be considered unoccupied, and argued that to correctly interpret a string of non-detections, one must have an expectation of species prevalence. Using nationwide monitoring data for reptiles in Switzerland, I showed that this is problematic because most species have no natural scale at which prevalence should be assessed; consequently survey recommendations depend strongly on the assumptions made. Our results emphasised that for rare species, it will barely ever be possible to invest sufficient survey effort to ensure that undetected populations are not overlooked, and that by incorporating knowledge of species prevalence, monitoring for invasive species will always cease before the species is truly eradicated.
In Chapter Four, I challenged the claim that volunteer-collected data are of low quality by developing a dynamic occupancy model which accounts for false-positive records in addition to imperfect detection. I quantified false-positive error rates for a long-term amphibian monitoring program and demonstrated that false-positives were uncommon for rare species, yet for the most common species monitored, up to 10% of records represented false-positives. I presented guidelines for designing future volunteer programmes such that false-positive records can be readily quantified, but concluded that for our monitoring dataset, ignoring false-positive records would not lead to a quantitative change in occupancy trends for any of the species monitored.
In Chapter Five, I called for caution when using count data to conduct large-scale abundance estimation for amphibian populations. I argued that amphibian populations may pose particular difficulties in abundance estimation because i) detection probabilities are often low, and ii) because it is typical that only a proportion of the total population will be available for detection during surveys. Using simulated datasets, I demonstrated that if either availability for detection or detection probabilities themselves are low, abundance estimates from open N-mixture models cannot be relied upon. Furthermore, I developed a new model formulation for scenarios in which availability is expected to vary predictably over a period of time. Using simulating data representing egg-mass counts, I demonstrated that by mechanistically building the availability process into N-mixture models, it is possible to accurately derive population size estimates even when availability is low.
In Chapter Six, I focussed on errors within photographic datasets used in mark-recapture studies. Failing to correctly identify individuals from photographs, or wrongly matching individuals, can induce severe bias in capture-recapture datasets. In this chapter, I investigated the effects of WILD-ID, a software designed to aid photographic recognition, upon the false-positive and false-negative error rates of 63 volunteers using a test dataset of photographs of yellow-bellied toads. I found that photographic identification software greatly increased the speed of matching as well as leading to a strong decline in both the frequency of false-negative matches, and in the variation in error rates between volunteers. I used this finding to argue that such software should be routinely used in long-term photographic monitoring programmes in order to minimise inevitable variation in matching abilities caused by staff turnover.

Abstract

Successful conservation management is underpinned by a solid understanding of species distributions and population trends, which is used both to identify populations under threat, and to monitor the effectiveness of management actions. However, as humans are imperfect, observation errors are introduced when we monitor variables such as occupancy patterns or abundances. These errors, if left unaccounted for, can bias inference in ways which can be detrimental to species conservation. In this thesis I used simulations and case studies from amphibian and reptile populations in Switzerland to discuss several forms of observation error, illustrated how their presence may lead to bias, and presented methods by which such biases can be avoided or mitigated.
In Chapter Two, I introduced the issue of imperfect detection in species monitoring and explained how occupancy modelling can be used to estimate detection probabilities and thereby accurately assess occupancy rates. In recent years, some authors have questioned the benefit of collecting the extra information necessary to apply this method, and have argued that conservation practise is not improved by accounting for detection. By applying IUCN red-list guidelines to nationwide monitoring data for amphibians within Switzerland, I demonstrated explicitly how failing to account for detection probabilities could lead to inappropriate management decisions being made. 5 of 12 species would have been inappropriately assigned to a higher extinction threat category if detection probabilities were ignored. Using this case study, I highlight that presence-only datasets can only ever be used to calculate maximum possible declines, and are inadequate for estimating the true magnitude of population change. In order to do so, monitoring practise must change to systems in which non-detections as well as species presences are routinely recorded.
In Chapter Three, I expanded upon the challenges of designing monitoring programs such that species absence can be reliably inferred. I compared two different frameworks used to determine how much effort must be invested in surveying a site before it can be considered unoccupied, and argued that to correctly interpret a string of non-detections, one must have an expectation of species prevalence. Using nationwide monitoring data for reptiles in Switzerland, I showed that this is problematic because most species have no natural scale at which prevalence should be assessed; consequently survey recommendations depend strongly on the assumptions made. Our results emphasised that for rare species, it will barely ever be possible to invest sufficient survey effort to ensure that undetected populations are not overlooked, and that by incorporating knowledge of species prevalence, monitoring for invasive species will always cease before the species is truly eradicated.
In Chapter Four, I challenged the claim that volunteer-collected data are of low quality by developing a dynamic occupancy model which accounts for false-positive records in addition to imperfect detection. I quantified false-positive error rates for a long-term amphibian monitoring program and demonstrated that false-positives were uncommon for rare species, yet for the most common species monitored, up to 10% of records represented false-positives. I presented guidelines for designing future volunteer programmes such that false-positive records can be readily quantified, but concluded that for our monitoring dataset, ignoring false-positive records would not lead to a quantitative change in occupancy trends for any of the species monitored.
In Chapter Five, I called for caution when using count data to conduct large-scale abundance estimation for amphibian populations. I argued that amphibian populations may pose particular difficulties in abundance estimation because i) detection probabilities are often low, and ii) because it is typical that only a proportion of the total population will be available for detection during surveys. Using simulated datasets, I demonstrated that if either availability for detection or detection probabilities themselves are low, abundance estimates from open N-mixture models cannot be relied upon. Furthermore, I developed a new model formulation for scenarios in which availability is expected to vary predictably over a period of time. Using simulating data representing egg-mass counts, I demonstrated that by mechanistically building the availability process into N-mixture models, it is possible to accurately derive population size estimates even when availability is low.
In Chapter Six, I focussed on errors within photographic datasets used in mark-recapture studies. Failing to correctly identify individuals from photographs, or wrongly matching individuals, can induce severe bias in capture-recapture datasets. In this chapter, I investigated the effects of WILD-ID, a software designed to aid photographic recognition, upon the false-positive and false-negative error rates of 63 volunteers using a test dataset of photographs of yellow-bellied toads. I found that photographic identification software greatly increased the speed of matching as well as leading to a strong decline in both the frequency of false-negative matches, and in the variation in error rates between volunteers. I used this finding to argue that such software should be routinely used in long-term photographic monitoring programmes in order to minimise inevitable variation in matching abilities caused by staff turnover.

Statistics

Additional indexing

Item Type:Dissertation
Referees:Ozgul A, Schmidt B R, Keller L, Holderegger R
Communities & Collections:07 Faculty of Science > Institute of Evolutionary Biology and Environmental Studies
Dewey Decimal Classification:570 Life sciences; biology
590 Animals (Zoology)
Uncontrolled Keywords:monitoring, survey, occupancy, abundance, detection probability, false positive, absence, availability, mark-recapture
Language:English
Date:2017
Deposited On:19 Dec 2017 13:38
Last Modified:19 Mar 2018 09:25
Number of Pages:193
OA Status:Closed

Download

Full text not available from this repository.
Get full-text in a library