Rangers and modellers collaborate to build and evaluate spatial models of African elephant poaching
Introduction
Monitoring trends within socio-ecological systems (species populations, illegal harvest rates, etc.) is essential for adaptive management, helping managers understand and manage change (Nichols and Williams, 2006). Evaluating anti-poaching strategies, for example, requires reliable measurement of real poaching trends. Data on biodiversity and threats are however difficult to gather at relevant scales, and are often biased and imprecise (Field et al., 2007). Time and resource constraints often mean that monitoring data are collected by people doing other jobs, such as wildlife rangers detecting snares while on patrol or fisherman providing records of bycatch species landed. Such opportunistic data present unique challenges to interpretation (Keane et al., 2011). A drop in the detection of poachers' snares, for example, may reflect a shift in patrolling to a ‘non-hotspot’ area, rather than an actual change in poaching levels.
Another challenge to interpreting observational data is the complexity of the underlying mechanisms generating the data. The behaviours of data generators (e.g. poachers), data collectors (e.g. rangers) and species of concern (e.g. elephants) are likely to interact in complex ways and their relative influence is difficult to disentangle. Dobson et al., (2018), for example, show how deterrence of poachers by rangers can confound inferred trends on the prevalence of illegal activity. Imperfect detectability of illegal activity (like bushmeat snares in thick forest; O'Kelly et al., 2018), and patrol observations that are biased towards certain areas (Critchlow et al., 2015), may similarly confound true patterns.
Participatory modelling is a promising way to design quantitative models that are robust to uncertainty arising from the bias and complexity discussed above (Voinov and Bousquet, 2010). Bringing together people familiar with the system of interest provides essential qualitative context to modelling (Milner-Gulland and Shea, 2017). These may be fisherman, wildlife rangers, or protected area managers that have a grounded understanding of how a system works (e.g. where elephant poaching happens) and how data are collected (e.g. what affects ranger movements). Participatory or collaborative modelling involves using the qualitative insights of on-the-ground practitioners and stakeholders in both the design and validation stages of statistical/mathematical modelling (Voinov and Bousquet, 2010). Quantitative models are vulnerable to the data and assumptions used to build them, while qualitative insights are often subjective or incomplete. Combining multiple lines of evidence (statistical outputs and interview responses) is a useful way of addressing this uncertainty. Engaging practitioners in modelling may also create a sense of ownership that amplifies its real-world relevance (Basco-Carrera et al., 2017).
Globally, tens of thousands of park rangers spend significant amounts of time on patrol, encountering plants, animals, and illegal activities. Such data are becoming an increasingly important source of information for both science and conservation (Gray and Kalpers, 2005; Moore et al., 2018). The MIKE programme (Monitoring of the Illegal Killing of Elephants), is a high-profile example of the use of data collected by ranger patrols to inform local and international conservation policy (CITES Secretariat, 2019). MIKE covers 60 sites across Africa, within which >19,000 elephant carcasses have been detected by rangers to date. The data have been used in high profile global and continental analyses (Wittemyer et al., 2014; Hauenstein et al., 2019), but less so at the local site level. In this paper, we investigate spatial patterns in poached elephant carcasses detected by rangers at a MIKE site in the Zambezi Valley, Zimbabwe. We combine quantitative models with interviews with widlife rangers and their supervisors to address the following research questions:
- (1)
What spatial patterns are evident in poached elephant mortalities at the case study site?
- (2)
How are these patterns influenced by monitoring bias?
Section snippets
Study area
The Chewore Safari Area MIKE site (3390 km2; hereafter Chewore) in Zimbabwe is part of the World Heritage Site comprising three adjacent protected areas (PAs): Mana Pools National Park and the Chewore and Sapi Safari Areas (Fig. 1). The elephant population in the broader Zambezi Valley declined by an estimated 42% (19,981 to 11,656) between 2003 and 2014, primarily due to poaching (Dunham, 2015; ZPWMA, 2015). Chewore is divided into two management units (north and south) and is also a sport
Results
In all scenarios, the random forests and generalised boosted models performed best at predicting poached carcass distribution (AUC/TSS scores, Fig. S2 Supporting information). The ensemble model in each scenario performed markedly better than the single models (Fig. S2). NDVI and tree cover were correlated (r = 0.69). We excluded NDVI because it varies widely between seasons whereas the models averaged 17 years of data. All other predictor pairs had r < 0.6.
The effect of each predictor on
Discussion
Uncertainty is recognised as an important topic within socio-ecological systems research. These systems comprise complex and uncertain linkages between human behaviour and natural systems (Milner-Gulland and Shea, 2017). In line with this, applied ecologists are developing more robust tools for dealing with one particular class of uncertainty: observation uncertainty, the discrepancy between the true and observed states of the natural system under management (Bunnefeld et al., 2017). However,
Data archiving
We intent to archive the data on which our analysis is based in the Dryad digital repository, pending necessary permissions from the Zimbabwe Parks and Wildlife Management Authority.
CRediT authorship contribution statement
Timothy Kuiper: Conceptualization, Methodology, Formal analysis, Writing - original draft, Writing - review & editing. Blessing Kavhu: Conceptualization, Methodology, Writing - original draft, Writing - review & editing, Data curation. Nobesuthu A. Ngwenya: Conceptualization, Writing - original draft, Writing - review & editing, Data curation. Roseline Mandisodza-Chikerema: Conceptualization, Writing - original draft, Writing - review & editing, Data curation. E.J. Milner-Gulland:
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We thank the Research Council of Zimbabwe and the Zimbabwe Parks and Wildlife Management Authority (ZPWMA) who provided permission to carry out this research. TRK is funded by the Commonwealth Scholarship Commission (PhD scholarship ZACS-2017-648). Danica Kuiper and Sally Kuiper are thanked for their help with logistics during field work. Richard Maasdorp and Lynne Taylor provided invaluable on-the-ground knowledge to help plan field work. The rangers and managers of Chewore Safari Area are
References (38)
- et al.
Ensemble forecasting of species distributions
Trends Ecol. Evol.
(2007) - et al.
Collaborative modelling or participatory modelling? A framework for water resources management
Environ. Model. Softw.
(2017) - et al.
Monitoring does not always count
Trends Ecol. Evol.
(2010) - et al.
Monitoring for conservation
Trends Ecol. Evol.
(2006) - et al.
Robust estimation of snare prevalence within a tropical forest context using N-mixture models
Biol. Conserv.
(2018) - et al.
Modelling with stakeholders
Practical solutions for making models indispensable in conservation decision-making
Divers. Distrib.
(2013)- et al.
Ecosystem-scale effects of megafauna in African savannas
Ecography
(2016) - et al.
Selecting pseudo-absences for species distribution models: how, where and how many?
Methods Ecol. Evol.
(2012) Spatial analysis of aerial survey data reveals correlates of elephant carcasses within a heavily poached ecosystem
Report to CITES CoP 18 on Monitoring the Illegal Killing Of Elephants (MIKE). CoP18 Doc. 69.2
Spatiotemporal trends of illegal activities from ranger-collected data in a Ugandan national park
Conserv. Biol.
Improving law-enforcement effectiveness and efficiency in protected areas using ranger-collected monitoring data
Conserv. Lett.
On the evaluation, monitoring and management of law enforcement patrols in protected areas
Detection deterence from patrol data
Conserv. Biol.
Collinearity: a review of methods to deal with it and a simulation study evaluating their performance
Ecography (Cop.)
Drought threatens Mali elephants
Pachyderm
Exploring the relationship between NDVI and African elephant population density in protected areas
Afr. J. Ecol.
Cited by (21)
Crowdfunding marine and coastal protected areas: Reducing the revenue gap and financial vulnerabilities revealed by COVID-19
2023, Ocean and Coastal ManagementArtificial illumination influences niche segregation in bats
2021, Environmental PollutionCitation Excerpt :This led to a final set of six environmental predictors used to build the SDMs in the ALMNP for both species: elevation, average maximum temperature of the warmest month (BIO 5), precipitation of the warmest quarter (BIO 18), distance from broadleaved forests, distance from riparian vegetation, and artificial illumination (Supplementary Materials – Table S1). We We built SDMs through an ensemble forecasting approach, as implemented in the R package “biomod2” (https://cran.r-project.org/bin/windows/base/; Thuiller et al., 2009), a method widely used in spatial ecology (e.g., Hao et al., 2019; Bertolino et al., 2020; Kuiper et al., 2020; Raffini et al., 2020; Smeraldo et al., 2021). We considered five modelling techniques (Thuiller et al., 2009; Jiguet et al., 2010), as follows: 1) Generalized Linear Models (GLM); 2) Generalized Boosted Models (GBM); 3) Random Forests (RF); 4) Multivariate Adaptive Regression Spline (MARS), and 5) Maximum Entropy Models (MAXENT).
Economics of conservation law enforcement by rangers across Asia
2023, Conservation Letters