Applying Crime Pattern Theory and Risk Terrain Modeling to Examine Environmental Crime in Cambodia

Abstract

The spatial-temporal analysis of crime has significantly evolved. One innovative technique recently developed is risk terrain modeling (RTM). RTM, however, has yet to be used for environmental crime. This research applies RTM and draws from crime pattern theory to examine illegal activities in two protected areas in Cambodia. Findings suggest that pathways, edges, areas with suitable targets, conservation posts, landcover, and prior incidents are related to fauna- and flora-related illegal activities, though this relationship varies by season, units of analysis, and study area (i.e., patrol-based compared with official designation). Implications for theory and policy are outlined.

Keywords

Asia conservation criminology crime science environment crime environmental criminology spatial analysis

Introduction

Despite its wide-reaching social, cultural, economic, and ecological impact, the study of environmental crime, as well as its prevention and enforcement, is a “fringe” area within criminology. As defined by the Environmental Investigation Agency (2008), “[e]nvironmental crimes can be broadly defined as illegal acts which directly harm the environment” (p. 1). Environmental crime has largely been studied by researchers in other disciplines who utilize perspectives that are typically not informed by criminological research. Criminology, however, is a rich field that has much to offer in the broader study of environmental crime. This is evidenced by the development of distinct subdisciplines, including green (White & Heckenberg, 2014) and conservation criminology (Gibbs et al., 2010), as well as the application of established criminological perspectives such as environmental criminology (Moreto & Pires, 2018). Despite the strong theoretical foundation available within criminology, some have commented on the need to further conduct empirical assessments to cement the utility and value of a criminological study of environmental crime (see Nobles, 2019), particularly from a quantitative orientation (see Lynch et al., 2017).

One area that quantitative criminology can contribute to the study of environmental crime is through the spatial analysis technique known as risk terrain modeling (RTM).¹ To date, RTM has not been used to examine the spatial-temporal factors that may influence environmental crime. Given its applicability in forecasting crime in urban settings, as well as its potential use to allocate enforcement and crime prevention resources, RTM presents a novel analytical approach to study illegal activities in protected areas (PAs).

Guided by crime pattern theory (Brantingham & Brantingham, 1993a), this study contributes to the literature by using RTM to examine the environmental factors that influence fauna- and flora-related illegal activities in PAs in Cambodia. We also add to the growing RTM literature examining the impact of seasonality (e.g., Szkola et al., 2019) by explicitly comparing dry and wet seasons. Furthermore, given the nature of the data, we also develop and compare data-driven study areas to officially designated study areas. Specifically, we sought to complete three research objectives: first, to use RTM to assess spatial risk factors that influence fauna-related illegal activities during the dry and wet seasons in Cambodian PAs. Second, to use RTM to assess spatial risk factors that influence flora-related illegal activities during the dry and wet seasons in Cambodian PAs. Third, to compare RTM results between patrol-based study areas to complete study areas (i.e., designated PA boundaries) to identify any potential differences in spatial forecasting results.

The Promise and Challenge of Protected Areas

According to the International Union for Conservation of Nature (IUCN), “a protected area is a clearly defined geographical space, recognised, dedicated and managed, through legal or other effective means, to achieve the long-term conservation of nature with associated ecosystem services and cultural values.”² The establishment of PAs represents a tangible achievement for conservation and naturalist goals, and the resources dedicated to the long-term preservation of these assets directly benefit the indigenous species as well as the surrounding environments. PAs are generally funded through public means, ensuring some level of public visibility and access. Furthermore, PAs are readily identifiable as zones of flourishing growth, often focused around endangered (and therefore economically valuable) natural resources.

Despite the potential benefit to wild spaces, wildlife, and local communities, PAs have been criticized for effectively removing resources from local communities that have been used sustainably and have historically relied on such resources (Duffy, 2010). Moreover, increased attention on wildlife crimes, like the poaching of endangered species, has led to magnified scrutiny on local communities living near PAs, even though certain forms of wildlife crime can be driven by external entities and influenced by both legal and illegal market dynamics (Felbab-Brown, 2017). Notably, justification for the increased militarization of PA enforcement (see Duffy et al., 2019) has tended to point to the increasing arms race between rangers and poachers, potentially discouraging community involvement in PA stewardship and ownership while also effectively reducing community–ranger relations and ranger legitimacy (see Moreto et al., 2017; Moreto & Gau, 2017). In general, this results in a biocentric-focus that enforces a Western-biased view of conservation that fails to account for ground-level realities faced by local communities, including circumstances where they are pressured to retaliate against wildlife (i.e., human–wildlife conflict; see Moreto, 2019), as well as broader forces beyond their control, such as population growth, land conversion, and political instability (Duffy, 2010).

The primary method used to manage and monitor PAs is through anti-poaching or law enforcement ranger patrols. Such patrols are often conducted by rangers³ who are tasked with performing a variety of activities, including patrol and human–wildlife conflict resolution. In addition, rangers also engage in ranger-based data collection (RBDC) during patrols, which results in the collection of data for a variety of PA management and monitoring purposes. Such data form the basis for this study.

Theoretical Framework: Crime Pattern Theory

The theoretical foundation of RTM is largely driven by concepts and theories found in environmental criminology. Recently, environmental criminology has been proposed to be a viable framework for environmental crimes (see Moreto & Pires, 2018). Environmental criminology is comprised of three main approaches (routine activity approach, the rational choice perspective, and the geometric theory of crime) and one meta-theory (crime pattern theory). For this study, we utilize crime pattern theory as our theoretical framework.

The geometric theory of crime emphasizes the role that the natural and built environment, places/activity nodes, boundaries/edges, and pathways have on determining individual activity spaces (i.e., the locations that individuals move to and from) and awareness spaces (i.e., knowledge of places and pathways near activity spaces) (Brantingham & Brantingham, 1993b). Pathways (e.g., streets, trails, etc.) facilitate access to locations where individuals spend a considerable amount of time (e.g., home, work, etc.). The decision to utilize and become familiar with a pathway is based on accessibility, efficiency, and the established routine activities of an individual (Cohen & Felson, 1979). These activities are often influenced by boundaries or edges (Brantingham & Brantingham, 1993b). Edges, in particular, can be both physical and social barriers, or they can be indicators of transition. For example, a fence and river represent a built and natural edge, respectively. A border between two countries represents an indicator of transition and can be both a physical and social barrier as well.

Activity nodes, edges, and pathways work in unison to develop an individual’s activity and awareness spaces. An individual’s comfortability and familiarity with the surrounding environment will be dictated by their activity and awareness space. Through experience and time, the individual can develop a crime template, which is “a checklist of items or circumstances that must be present or absent” to successfully complete a criminal act (Andresen, 2014, p. 52). Through such familiarity, an individual can become sensitized to signs in the immediate environment that signal when it is an opportune time to commit a crime, also referred to as a triggering event (Andresen, 2014).

Within crime pattern theory, two additional factors must be considered, with the first being the existence of places known as crime generators and crime attractors (Brantingham & Brantingham, 1995). Crime generators are activity nodes where large numbers of individuals go for noncriminal activities (e.g., a shopping mall). Crime occurs simply because of the increased likelihood of potential offenders and targets converging in time and space. Such crimes are largely opportunistic. Crime attractors, on the other hand, are places where offenders go to explicitly engage in criminal activity (e.g., known drug hot spots). Second, research suggests that crime journeys are short (Townsley & Sidebottom, 2010). This distance decay is not surprising because costs (e.g., time and effort) increase as one moves further away from their activity nodes (Rengert et al., 1999). Collectively, this dynamic setting forms the environmental backcloth in which an offender operates, or the “elements that surround and are part of an individual and that may be influenced by or influence his or her criminal behaviour” (Brantingham & Brantingham, 1993b, p. 6).

Environmental Crime in Protected Areas: Situational Risk Factors and Correlates

There have only been a handful of criminological studies that have examined the spatial correlates of environmental crime (see e.g., Kurland et al., 2018; van Doormaal et al., 2018; Weekers et al., 2019). Yet within the broader conservation sciences literature, there have been several studies that have spatially assessed the distribution and correlates of environmental crime. We separate the following section into fauna- and flora-related risk factors and correlates. We include crimes that involve natural resources, like illegal logging, within these two broad categories given our research objectives as well. It is important to note that while there may be other risk factors and correlates related to environmental crime in PAs, we limit our analyses to the risk factors and correlates that were identified in the literature for this study.

Fauna-Related Risk Factors and Correlates

Distance to cultivated area, settlements, and boundary of the protected area

The distance to cultivated locations, human settlements, and PA boundaries have been linked to increased poaching activity. One study found that illegal activities occurred closer to human habitation within a study site in central Africa, possibly due to lower travel costs between the targets and poachers, as well as diminished detection from law enforcement (Plumptre et al., 2014). This coincides with research that has also found that poaching activities are more likely to occur along the perimeter of the PA (Plumptre et al., 2014; Watson et al., 2013). Although this points to the potential decision-making of offenders (i.e., costs associated with traveling further into the PA), it is possible that these areas simply provide increased opportunities for humans and wildlife to converge in time and space. In a study of wire-snaring patterns in a Zambian national park, Watson and colleagues (2013) determined that the likelihood of a presence of a wire-snare increased when closer to crops or cleared land. The growth and expansion of communities and diminishing resources and space for wildlife may result in competition, precipitating human–wildlife conflict, and resulting in retaliatory killings of wildlife species (Moreto, 2019).

Density of suitable targets

Higher density of targets has been found to increase the likelihood of poaching activities. An investigation of elephant poaching in Kenya found that elephant population density was a significant predictor in poaching behaviors, which could be due to having a greater “harvest” relative to the effort involved (Rashidi et al., 2016). In a study by van Doormaal and colleagues (2018), rhinoceros poaching in the northeastern part of South Africa was more likely to occur in areas of higher rhinoceros’ density. However, high-density targets may only increase the likelihood of certain poaching activities over others. For instance, in a study of patrol data in Queen Elizabeth Conservation Area in Uganda, commercial animal poaching was found to be impacted by the density of possible targets but was not a factor in predicting noncommercial poaching (Critchlow et al., 2015).

Density and proximity to sources of water

Animals, and poachers alike, require water on a routine basis. Not surprisingly, empirical studies have found a spatial relationship between animal poaching and the location and density of water sources. For example, the density of rivers, streams, and waterholes were found to be correlated with the density of elephant poaching incidents (Maingi et al., 2012). Regarding commercial and noncommercial animal poaching, Critchlow and colleagues (2015) found that the probability of poaching incidents was elevated near rivers. In addition, in an examination of wire-snaring in and around national parks in Zambia, Watson and colleagues (2013) found that snares were more common near permanent water sources.

Proximity to ecotourism lodges

In terms of PAs, tourist zones or lodges could act as a potential mechanism for guardianship against animal poaching. The rationale behind this would be that the management of these areas or facilities have a vested interest in maintaining the presence of wildlife to sustain tourism. Thus, individuals responsible for these areas might be more willing to report illegal activity to the appropriate authorities. Empirical findings, however, are mixed. Examining primate densities in Taï National Park, a PA in Côte d’Ivoire, N’Goran and colleagues (2012) found a positive correlation between the density of primates and the distance to the tourist site within the park, while van Doormaal and associates (2018) found that the presence of private facilitates and tourist lodges did not deter illegal border crossings for rhinoceros hunting in South Africa.

Proximity to ranger outposts

One might expect the presence of ranger outposts or bases to act as the primary example of guardianship and would deter incidents of poaching. Yet, to date, several studies have found the opposite. Maingi and colleagues (2012) found a negative correlation between elephant poaching densities and the distance to ranger outposts. Through a camera-trap survey in Khao Yai National Park, Jenks and colleagues (2012) found that poachers still operated within close proximities to both ranger headquarters and remote outposts. Researchers have posited that there may be collusion or wrongdoing on part of the rangers, or more simply, these results are a function of patrol-driven data collection. In other words, these data are collected while on routine patrols and rangers typically patrol close to their outposts, which subsequently would lead to more instances of poaching being recorded near these locations (Maingi et al., 2012; Plumptre et al., 2014). Furthermore, Jenks and colleagues (2012) offered another possible explanation for these findings: Poachers may be utilizing roadways that are often nearby these locations to access the PA.

Proximity to roads

The impacts of roads on wildlife has been of interest to conservationists and criminologists alike. Laurance et al. (2009) determined that roads can serve as barriers for the movement of animals while also increasing access to possible poaching targets. However, the research is mixed on whether the accessibility of roads increases the risk of poaching. Some evidence does suggest roads can be a significant factor in predicting the likelihood of poaching fauna (Watson et al., 2013), whereas others have found that proximity to roads was not an important predictor (Critchlow et al., 2015; Rashidi et al., 2016).

The Role of Seasonality

Apart from spatial patterns of animal poaching, two studies have also attempted to examine temporal trends of these incidences. With reference to the frequency (Maingi et al., 2012) and location (Rashidi et al., 2016) of elephant poaching, both studies have identified notable differences between the wet and dry seasons. In south-eastern Kenya, Maingi and colleagues (2012) found that elephant poaching incidents were more frequent during the dry season (January and February; June–October). Rashidi and colleagues (2016) observed that poaching incidents appeared in different locations during different time periods within south-east Kenya. Because the wet season impacts the suitability of the environment, poachers are unable to reach locations that are accessible during the dry season. In addition, the wet season may provide more cover for poaching targets, depending on the types of flora or vegetation present in certain areas (Maingi et al., 2012).

Flora-Related Risk Factors and Correlates

Much of the literature examining environmental crime has centered on crimes involving fauna. Despite limited research associated with flora products (including timber), there have been studies that point to spatial risk factors and correlates of flora crime. Analyzing the illicit market in American ginseng, Young et al. (2011) determined that poaching risk increased if the areas had previously been poached, if there were roads nearby, and if the habitat was of good quality (i.e., containing valuable targets). Kurland and colleagues (2018) determined that target abundance (density), availability of roads, and proximity to illegal markets all impacted the likelihood of specific redwood areas to be targeted for burl poaching. This was further expanded in a subsequent exploration of choice structuring properties by Marteache and Pires (2020), which found that factors that likely contribute to burl selection include risk of detection, proximity to burl shop, familiarity with the area, proximity to roads to allow for easy access to possible targets, the height of the burl relative to the ground (to allow for easy removal from the tree), and slope of the target relative to their vehicle (higher or similar slope to the road). While there is overlap between characteristics that drive fauna poaching (such as target density), this line of research suggests that there may be different factors that impact flora poaching.

Risk Terrain Modeling (RTM)

Spatial analytical methods featuring crime as the dependent variable have been prevalent over the past quarter century, both in the academic literature and in wide practice. Most spatial forecasting and spatial risk analysis methodologies share a common theme of developing quantitative estimates from existing (past crime) incident data, weighing according to geographic proximity and sometimes incorporating temporal elements. Beyond this foundation, however, subcategories of methodology differ with respect to their instrumental purposes, statistical basis, and associated limitations.

RTM extends previous research on developing crime opportunity surfaces (see Groff & La Vigne, 2001, 2002) for crime forecasting purposes (see Gorr & Olligschlaeger, 2002). Unlike predictive models that rely on retrospective analyses (e.g., hot spots analysis) and on data that are based on solely the location of recent or historical crime events, RTM offers the advantage of assessing relatively stable environmental and contextual risk factors that go beyond incident-based data. As such, RTM is uniquely situated to test key propositions outlined in crime pattern theory, including the environmental backcloth (see Moreto et al., 2014).

RTM involves the construction of a raster probability surface representing event risk based on spatial proximity to known criminogenic locations, which have a user-specified intensity value related to crime association (Caplan, Kennedy, & Miller, 2011). In other words, RTM is an analytical approach that identifies, conceptualizes, and operationalizes factors that are theoretically or empirically associated with the phenomenon of interest to forecast future events. Notably, RTM has been found to be more effective at forecasting crime incidents compared with traditional hot spot techniques (e.g., kernel density estimation; see Drawve, 2016).

The relative simplicity and user-friendliness of the technique and its related software (discussed later) provide an opportunity for practitioners to readily utilize the approach with minimal resources as well. Prior studies have demonstrated the applicability of RTM to assess a variety of criminal activities, like residential burglary (Dugato et al., 2018; Moreto et al., 2014), organized crime homicide (Dugato et al., 2017), violent crime (Caplan et al., 2013a), maritime piracy (Caplan, Moreto, & Kennedy, 2011), and terrorism (Onat & Gul, 2018). RTM, however, has not been used within the context of environmental crime.

Current Study

This study examines environmental crimes in two sites in Cambodia: Mondulkiri Protected Forest (MPF) (renamed as Srepok Wildlife Sanctuary in 2016) and Phnom Prich Wildlife Sanctuary (PPWS) (see Figure 1). The former has an estimated total area of 3,720 km², while the latter covers approximately 2,218 km². These two locations were chosen specifically due to their biodiversity, accessibility to data needed for this study, and the variation in illegal activities and threats that occur within the two sites. Specifically, both MPF and PPWS are subject to poaching for subsistence and to support both domestic and international wildlife trade, illegal logging of high-value timber, and mining, among other threats.⁴ Given the similarities in threats and the patrolling activities and strategies utilized in both sites, as well as the contiguous nature of the PAs, we decided to combine the PAs as a single study area for this study.

Figure 1.

Full study area boundaries.

For this study, we sought to address three research objectives: first, to use RTM to assess spatial risk factors that influence fauna-related illegal activities in Cambodian PAs during both dry and wet seasons. Second, to use RTM to assess spatial risk factors that influence flora-related illegal activities in Cambodian PAs during both dry and wet seasons. Finally, to develop and compare patrol-based study areas to designated study areas within the scope of RTM to identify possible differences in results.

Data and Method

Data were obtained from the Spatial Monitoring and Reporting Tool (SMART) platform and provided by the World Wide Fund for Nature (WWF). SMART is a software mapping program that facilitates the collection, storage, communication, and analysis of RBDC, including data about patrol activities and routes, illegal activities, and observations of wildlife. Data are collected through a mobile data collection application. SMART not only provides insight on the spatial and temporal distribution of illegal activities within PAs, but also provides site managers with the ability to assess ranger performance.⁵

It is important to highlight some of the inherent limitations of the data. An important caveat of data obtained from RBDC is that all observations, including those of illegal activities, are biased in that they only refer to observations that have been identified by rangers (see Gavin et al., 2010). In essence, RBDC data only reflect illegal activities that have been identified throughout the course of a patrol, and not necessarily where all illegal activities have occurred (see Moreto et al., 2014). RBDC also suffers from limitations that other more “traditional” forms of crime data suffer from: problems with accuracy of GPS devices (when such technology is available), errors in data entry, officer discretion, and the fact that evidence of a crime occurring may not be overly identifiable. Despite these limitations, and as shown earlier, RBDC offers some of the best available data to assess illegal activities in PAs and is sufficient for our current research objectives. Indeed, our third research objective attempts to assess the comparability of study areas, one based on site boundaries and the other based on patrols.

Analysis

Dependent Variables

The dependent variables for this study are split up into two primary classifications: flora and fauna incidents. These variables come from the SMART data described above. For this study, we had access to data from 2005 to 2015 for flora-related illegal activities, and 2006 to 2015 for fauna-related illegal activities. Due to the low number of observed illegal activities in any given year, we included data from 2008 to 2015. As will be discussed later, we utilized flora-related illegal activities between 2005 and 2007, and 2006–2007 for fauna-related illegal activities as a measure of previous incidents. With reference to flora incidents, these include instances where rangers encountered potential suspects, equipment, felled logs, or other signs of illegal logging or flora poaching within the PAs (n = 1,376). For fauna, data points included incidents in which poachers were confronted, weapons or equipment spotted, animal carcasses found, or other signs of poaching identified by rangers (n = 492). As a temporal element to this study, each dependent variable was divided based on occurrences during the wet and dry season for Cambodia. The wet season lasts from May 1 to October 31 (fauna, n = 320; flora, n = 737) and the dry season from November 1 to April 30 (fauna, n = 172; flora, n = 639).

Risk Factors

For this study, three of the 15 risk factors were obtained from the SMART data described above. These include animal sightings, previous flora incidents, and previous fauna incidents. All three of these variables were available as point shapefiles. Animal sightings refer to all instances in which a ranger observed fauna while on routine patrol. Previous incidents for flora contain illegal activity incidents from the years 2005 through 2007. Previous fauna incidents contain illegal activity for the years 2006 and 2007. Data for the year 2005 for this risk factor were not available. The data for the remaining variables were provided by the WWF.

Several of these variables were provided as polygon shapefiles, including landcover and ecotourism zones. To assess different types of landcover, this shapefile was first split up into five unique environment types—woodland, semi-evergreen, non-forest, deciduous, and evergreen. For use within RTMDx, these data needed to be converted to point shapefiles through the following process. First, a raster grid (150 m × 150 m) was displayed over the entire study area. Cells that intersected with these polygons were extracted to provide a grid surface of only the areas of interest. From there, centroids of these individual raster cells were created, thus providing a coverage of equally spaced points where polygons existed. Plotting these points throughout the entire coverage of the polygons allows for the assessment of the density of these factors. In addition, three of the provided shapefiles were in polyline format, which were subsequently covered to points. Aside from the previously mentioned variables, three were already provided as point shapefiles, thus no manipulation of the data was necessary. These variables include villages, waterholes, and conservation posts.

RTMDx Parameters

For this study, RTMDx Version 1.0 (Caplan & Kennedy, 2013) was utilized. Although prior research utilizing the RTM approach has incorporated different analytical software, including ArcGIS (Caplan, Kennedy, & Miller, 2011), we decided to use the RTMDx software for our purposes. Within this software, there are six primary input fields necessary, including a shapefile for the study area boundary, block length, raster cell size, model type (aggravating or protective), outcome event, and risk factors. For this study, a shapefile with the two adjacent PAs of interest, MPF and PPWS, was used as the primary study area boundary. To address the third research objective, a shapefile of just the areas patrolled within the two PAs was also created. This was accomplished by first creating a 150-m by 150-m raster over the study area and extracting raster cells that intersected with the areas patrolled. The output raster was then dissolved to produce a single polygon to allow for use within RTMDx. In addition, a 50-m dissolved buffer was added to this shapefile to smooth out the edges. The overall coverage of this patrol-based study area, compared with the primary study area, was approximately 82% (see Figures 1 and 2).

Figure 2.

Patrol-based study area boundaries.

Considering the remote nature of this study area and the lack of city blocks typically found within RTM research, the block length for the initial analysis with the primary study area was chosen based on practical implications for rangers. Thus, a block length of 300 m was utilized. Prior research has found that the mean visibility profile for rangers was estimated to be 200 m during patrol (Kakira, 2010). As this was based solely on one study, and given the variability in terrain and environment, we adopted a more expanded patrol profile for our purposes, and therefore decided to utilize a 300-m block length. Caplan and colleagues (2013b) recommend a raster cell size that is half the size of the block length, so a raster cell size of 150 m was chosen. The 150-m cell size was sufficient to address our first two research objectives. Unfortunately, our hardware was unable to run RTMDx on the 150-m cell size to address our third research objective. We attributed this to the necessary processing power to run RTMDx on the patrol-based study area we created. Thus, we had to increase the unit of analysis to a 1,000-m raster cell size with a 2,000-m block length.

For the outcome event, four different dependent variables were utilized and tested, including flora-related illegal activity incidents for the wet and dry season, as well as fauna-related illegal activity during the wet and dry season. This allows for comparison of significant risk factors between flora and fauna, as well as a temporal component to compare risk factors present for each classification during the wet and dry season, respectively. This study assessed each of these outcome events under aggravating and protective models. Within RTMDx, an aggravating model is an assessment of whether specified risk factors correlate spatially with the presence of the dependent variable of interest. A protective model assumes the opposite—that the risk factors correlate with the absence of the chosen dependent variable (Caplan et al., 2013b).

Within the selection of risk factors, there are three primary inputs to select for each chosen variable: the maximum spatial influence (SI), operationalization, and analysis increments. For each risk factor, a maximum SI of three blocks was utilized, along with whole-block increments. Beyond these specifications, RTMDx allows for the selection of three primary operationalizations: density, proximity, or proximity and density. The density operationalization refers to concentrations of factors that increase the risk of the outcome of interest. The proximity operationalization assumes that a close proximal distance to the risk factor increases the risk of the outcome of interest. By selecting proximity and density, RTMDx will test both operationalizations and choose the best fit for the chosen risk factor (Caplan et al., 2013b). Because this option effectively doubles the size of the models being run, risk factors for this study were chosen to be either density or proximity. These selections were based on theoretical as well as empirical justifications mentioned previously in our literature review.

Each of the five landcover variables—woodland, semi-evergreen, non-forest, deciduous, and evergreen—was operationalized as density. Because certain animals may be more detectable in certain areas, such as non-forest versus an evergreen forest, the density of these landcovers could influence the likelihood of illegal activity (Plumptre et al., 2014). Ecotourism zones were also operationalized as density due to previous research suggesting higher animal concentrations exist near tourist sites, pointing to such locations as an effective mechanism of guardianship against illegal activity. Roads were operationalized as proximity due to research that suggests illegal activity in PAs is often near such pathways because poachers use these roads for transit (Critchlow et al., 2015; Rashidi et al., 2016; Watson et al., 2013). Villages were operationalized as density because of previous research that shows concentrations of settlements are related to illegal activity incidents in surrounding areas (Rashidi et al., 2016). We surmise that this may be due to the increased potential number of offenders, the journey-to-crime of offenders residing in such locations, and because of increased likelihood of human–wildlife conflict.

As previously discussed, flora and fauna alike tend to cluster near sources of water (Maingi et al., 2012; Rashidi et al., 2016). This means there could be an increased likelihood of a poacher hunting for a potential target around a crime attractor, like a waterhole (see Moreto & Pires, 2018). Because of this, waterholes were operationalized as density. Similarly, rivers are also a source of water, but because of the extended distance that an individual river can be, they were operationalized as proximity. Previous research has also shown that poachers typically offend near edges like park boundaries (Watson et al., 2013), thus the boundaries for this study were operationalized as proximity. Because known or high target densities, for both flora (Young et al., 2011) and fauna (Critchlow et al., 2015; Rashidi et al., 2016; van Doormaal et al., 2018), have been shown to correlate with future illegal activity, animal sightings, previous flora incidents, and previous fauna incidents were all operationalized as density. Finally, conservation posts were operationalized as proximity because previous empirical research has shown that illegal activity often occurs in close distance to these locations (Jenks et al., 2012; Maingi et al., 2012; Rashidi et al., 2016).

Table 1 displays each risk factor along with the input parameters for RTMDx. Twelve of the 15 risk factors displayed are used in both model classifications (flora compared to fauna) for this study, with three that are used in either the flora or fauna models. This provides a total of 13 risk factors for the flora models and 14 for the fauna models.

Table 1.

Operationalization of Risk Factors in RTMDx.

Risk factor	n	Operationalization	Spatial influence	Increment	Model
Woodland	12,028	Density	3 blocks	Whole-block	All
Semi-evergreen	113,926	Density	3 blocks	Whole-block	All
Non-forest	28,415	Density	3 blocks	Whole-block	All
Evergreen	26,467	Density	3 blocks	Whole-block	All
Deciduous	182,512	Density	3 blocks	Whole-block	All
Roads	4,188	Proximity	3 blocks	Whole-block	All
Villages	30	Density	3 blocks	Whole-block	All
Waterholes	1,173	Density	3 blocks	Whole-block	All
Rivers	63,505	Proximity	3 blocks	Whole-block	All
PA boundary	2,399	Proximity	3 blocks	Whole-block	All
Ecotourism zones	2,846	Density	3 blocks	Whole-block	All
Conservation post	13	Proximity	3 blocks	Whole-block	All
Animal sightings	14,000	Density	3 blocks	Whole-block	Fauna
Fauna previous	123	Density	3 blocks	Whole-block	Fauna
Flora previous	287	Density	3 blocks	Whole-block	Flora

Note. PA = protected area.

Within these parameters and using a stepwise regression technique, RTMDx tests the varying degrees of SI and presents the best model specifications as the final output with only the most significant risk factors (Caplan et al., 2013b). This best model specification is chosen based on the Bayesian information criteria (BIC) from each model tested. From these model specifications, the results from a total of 24 models are presented below.

Results

For each of the following outputs, RTMDx used a negative binomial model. Within these outputs, RTMDx provides several items of interest. Specifically, the best model specifications include an SI, coefficient, and relative risk value (RRV) for each risk factor identified. In addition, RTMDx provides an output for the maximum relative risk score (RRS) as identified within the risk surface. In the context of this study and with an aggravating model specification, a higher RRS in an individual cell indicates a heightened level of risk associated with an illegal incident. For a protective model specification, a higher RRS would be indicative of an elevated risk of the absence of the outcome of interest. Thus, the maximum RRS indicates the highest value identified across the entire risk surface. It should be noted that in separate models, two variables—villages and ecotourism zones—were tested with the proximity operationalization with the same results obtained. For the first two research objectives, Table 2 displays the results from the 150-m raster cell size models within the full study area boundaries. The protective model types for these specifications yielded no significant risk factors associated with the outcomes of interest, thus only the aggravated models are presented.

Table 2.

Summary of Results for 150-m Raster Cell Size Models.

Risk factor	Operationalization	Spatial influence (m)	Coefficient	RRV
1: Output for fauna model during wet season (aggravated)
Animal sightings	Density	300	1.7462	5.7328
Roads	Proximity	600	0.7976	2.2203
Intercept			−7.1184
2: Output for fauna model during dry season (aggravated)
Animal sightings	Density	300	2.1756	8.8075
Intercept			−7.6296
3: Output for flora model during wet season (aggravated)
Previous incidents	Density	600	1.2801	3.5970
Roads	Proximity	600	1.1082	3.0289
Intercept			−6.1968
4: Output for flora model during dry season (aggravated)
Previous incidents	Density	600	1.3805	3.9769
Conservation posts	Proximity	900	1.3580	3.8884
Roads	Proximity	600	0.9672	2.6305
Intercept			−6.3256

Note. 1: BIC: 3,265.3 / RRS: 12.7; 2: BIC: 1,938.9 / RRS: 8.8; 3: BIC: 6,758.6 / RRS: 10.9; 4: BIC: 6,055.9 / RRS: 40.7. RRV = relative risk value; BIC = Bayesian information criteria; RRS = relative risk score.

From the aggravated fauna models, two risk factors were found to be significantly correlated with illegal hunting activity during the wet season: the density of animal sightings and proximity to roads. Animal sightings were found to have an RRV of approximately 5.7, meaning that the risk of the outcome event is about 5.7 times as likely when compared with areas outside of the SI of 300 m. Roads, on the contrary, have an RRV of approximately 2.2 with an SI of 600 m. The dry season yielded only one significant risk factor: animal sightings. The SI for this factor remained the same between the wet and dry seasons. Across these two models, the highest maximum RRS was identified within the wet season (RRS = 12.7). This can be interpreted as cells that contain this value are approximately 12.7 times more likely to experience an illegal incident than a cell with an RRS of 1.

From the aggravated flora models with a raster cell size of 150 m, the density of previous illegal activity incidents was the riskiest factor associated with the outcome of interest (wet: RRV = 3.5970; dry: RRV = 3.9769). Both had a maximum SI of 600 m. Within both the wet and dry season, proximity to roads was identified as being correlated with illegal flora-related activity. When comparing the wet and dry seasons, only the dry season models found that proximity to conservation posts was significantly associated with illegal activity (RRV = 3.8884). From these aggravated flora models, the highest maximum RRS of the 150-m raster cell size models were identified for the dry season iterations (RRS = 40.7).

We now turn to discussing the results for our third research objective. Table 3 displays the results from the models of the full study boundaries with a raster cell size of 1,000 m. From these results, one protective model type produced significant results. Specifically, the flora-related model during the dry season found two protective factors to be significantly correlated with illegal flora-related activity. These factors include one landcover variable, density of deciduous forests (RRV = 1.7177), and the proximity to PA boundaries (RRV = 1.6468). As discussed above, within RTMDx, protective models assume that the chosen risk factors correlate with the absence of the outcome of interest (Caplan et al., 2013b). Thus, the density of deciduous forests and the proximity to study area boundaries are negatively associated with the presence of illegal flora-related activity. These findings are notable within the scope of RTM and spatial crime mapping in general given that much of the criminological literature has tended to focus on aggravating (i.e., criminogenic) factors as opposed to factors that may reduce illegal activities. From this protective model, the maximum RRS was relatively low (RRS = 2.8). In the context of these results, it is uncertain whether this is a factor of the conceptualization of the protective model specifications for RTMDx, or simply a result of relatively uninfluential risk factors.

Table 3.

Summary of Results for 1,000-m Raster Cell Size Models (Full Study Area).

Risk factor	Operationalization	Spatial influence (m)	Coefficient	RRV
1: Output for fauna model during wet season (aggravated)
Woodland	Density	6,000	0.8451	2.3282
Animal sightings	Density	2,000	0.8291	2.2913
Previous incidents	Density	6,000	0.7695	2.1586
Intercept			−3.4474
2: Output for fauna model during dry season (aggravated)
Animal sightings	Density	2,000	1.3566	3.8830
Previous incidents	Density	6,000	0.8104	2.2487
Intercept			−4.0466
3: Output for flora model during wet season (aggravated)
Woodland	Density	6,000	0.7270	2.0688
Previous incidents	Density	6,000	0.6786	1.9711
Evergreen	Density	6,000	0.3922	1.4803
Roads	Proximity	2,000	0.3825	1.4659
Intercept			−2.6773
4: Output for flora model during dry season (aggravated)
Previous incidents	Density	6,000	0.8766	2.4026
Roads	Proximity	2,000	0.5909	1.8056
Evergreen	Density	4,000	0.5449	1.7244
Intercept			−2.8535
5: Output for flora model during wet season (protective)
Deciduous	Density	4,000	0.5410	1.7177
PA boundaries	Proximity	2,000	0.4989	1.6468
Intercept			−1.9242

Note. 1: BIC: 1,830.9 / RRS: 11.5; 2: BIC: 1,167.0 / RRS: 8.7; 3: BIC: 3,543.7 / RRS: 8.8; 4: BIC: 3,254.3 / RRS: 7.5; 5: BIC 3,601.1 / RRS: 2.8. RRV = relative risk value; BIC = Bayesian information criteria; RRS = relative risk score; PA = protected area.

With reference to the aggravated models for the full study area boundaries and the 1,000-m raster cell size, several risk factors were identified by RTMDx as being significantly correlated with the outcomes of interest. With respect to the fauna models, the riskiest factor during the wet season was the density of woodland land-types (RRV = 2.3282). Following that, in terms of RRV, are animal sightings (RRV = 2.2913) and previous incidents (RRV = 2.1586). These two factors were also found to be the only risk factors significantly associated with illegal fauna-related activity during the dry season (animal sightings, RRV = 3.8830; previous incidents, RRV = 2.2487). The wet season model also produced a maximum RRS of 11.5.

With the flora-related models, the RTM for the wet season found four risk factors to be significantly related. These factors include the density of woodland land-types (RRV = 2.0688), previous incidents (RRV = 1.9711), evergreen forests (RRV = 1.4803), and the proximity to roads (RRV = 1.4659). The first three of these risk factors had a maximum SI of 6,000m, while roads had a maximum SI of 2,000 m. In contrast with this model, woodland land-types were not identified as being significantly correlated with illegal activity during the dry season. Instead, only density of previous incidents (RRV = 2.4026), proximity to roads (RRV = 1.8056), and density of evergreen forests (RRV = 1.7244) were found to be significantly correlated with the outcome of interest. As can be seen with the RRVs, the proximity to roads was identified as being riskier than the density of evergreen forests within this model. In addition, the maximum SI of the density of evergreen forests was reduced to 4,000 m for the dry season models.

The models for the patrol-based study area boundaries are displayed in Table 4. Overall, these models found similar results to the full study area boundaries discussed above. Much like the previous models, only one protective model produced risk factors that were significantly correlated with the outcome of interest. Specifically, this was the model for flora-related illegal activity during the wet season. Like the full study area model, the two factors related to the absence of illegal activity were the density of deciduous forests (RRV = 1.9068) and the proximity to PA boundaries (RRV = 1.7194). Unlike the previous model, however, the maximum SI for the proximity to study area boundaries was 2,000 m, compared with 4,000 m.

Table 4.

Summary of Results for 1,000-m Raster Cell Size Models (Patrol-Based Study Area).

Risk factor	Operationalization	Spatial influence (m)	Coefficient	RRV
1: Output for fauna model during wet season (aggravated)
Woodland	Density	6,000	0.8815	2.4146
Animal sightings	Density	2,000	0.8719	2.3914
Previous incidents	Density	6,000	0.6635	1.9415
Intercept			−3.4330
2: Output for fauna model during dry season (aggravated)
Animal sightings	Density	2,000	1.486	4.3414
Intercept			−3.9013
3: Output for flora model during wet season (aggravated)
Woodland	Density	6,000	0.7776	2.1762
Previous incidents	Density	6,000	0.5770	1.7807
Roads	Proximity	2,000	0.5504	1.7340
Intercept			−2.6453
4: Output for flora model during dry season (aggravated)
Roads	Proximity	2,000	0.6616	1.9378
Previous incidents	Density	6,000	0.5510	1.7350
Woodland	Density	6,000	0.5043	1.6558
Evergreen	Density	4,000	0.4882	1.6294
Intercept			−2.8775
5: Output for flora model during wet season (protective)
Deciduous	Density	4,000	0.6454	1.9068
PA boundaries	Proximity	4,000	0.5420	1.7194
Intercept			−1.8377

Note. 1: BIC: 1,863.0 / RRS: 11.2; 2: BIC: 1,175.1 / RRS: 4.3; 3: BIC: 3,546.6 / RRS: 6.7; 4: BIC: 3,284.1 / RRS: 9.1; 5: BIC 3,602.2 / RRS: 3.3. RRV = relative risk value; BIC = Bayesian information criteria; RRS = relative risk score; PA = protected area.

The aggravated models for fauna-related illegal activity as the outcome of interest produced similar results to the previous full study area boundary models. With regard to the wet season, and like the full study area boundary model, three risk factors were identified as being significantly correlated with illegal activity—the density of woodland land-types (RRV = 2.4146), the density of animal sightings (RRV = 2.3914), and the density of previous incidents (RRV = 1.9415). The maximum SIs of each of the risk factors remained the same between the two different models as well (woodland, SI = 6,000 m; animal sightings, SI = 2,000 m; previous incidents, SI = 6,000 m). The model for the dry season did, however, provide some differing results than the full study area model. Specifically, only the density of animal sightings (RRV = 4.3414) was identified as being correlated with the presence of illegal activity. As with previously discussed models, the highest maximum RRS was identified within the wet season fauna model (RRS = 11.2).

For the aggravated flora-related models, the wet season model only identified three significant risk factors, as opposed to four from the full study area boundary model. These risk factors included the density of woodland land-types (RRV = 2.1762), the density of previous incidents (RRV = 1.7807), and the proximity to roads (RRV = 1.7340). In comparison to the full study area boundary, the density of evergreen forests was not significantly correlated to the presence of flora-related illegal activity within the patrol-based study area. For the dry season model, there were several marked differences between the risk factors that were significantly correlated with the presence of illegal activity. Notably, there were four identified risk factors, as opposed to three for the full study area boundary. The riskiest factor for the patrol-based model was the proximity to roads (RRV = 1.9378), followed by the density of previous incidents (RRV = 1.7350), the density of woodland land-types (RRV = 1.6558), and the density of evergreen forests (RRV = 1.6294). In contrast, the full study area boundary found woodland land-types not significantly correlated with the outcome of interest, and the riskiest factor was the density of previous incidents, as opposed to the proximity to roads.

Discussion

In this study we applied crime pattern theory and used RTM to examine illegal activities in two sites in Cambodia. In addition, we explicitly incorporated a temporal element in our analyses by separating our models into dry and wet seasons. Based on our findings, seasonality appeared to be an important component that influenced the influence of risk factors included in this study. For example, in our initial fauna-based model, density of animal sightings and proximity to roads were identified to be risk factors during the wet season, while only animal sightings were identified during the dry season. For our initial flora-based model, both density of illegal activity and proximity to roads were identified to be significant risk factors in both dry and wet models, while proximity to conservation posts was identified to be a risk factor only in the dry model.

We also compared RTM results at different units of analysis. Findings demonstrate that our models did display variation between the 150-m and 1,000-m model specifications. For the fauna-based dry model, the 1,000-m analysis found density of woodland, density of animal sightings, and density of previous incidents to be significantly associated, while the 150-m analysis found density of animal sightings and proximity to roads to be determining risk factors. Similarly, for the flora-based dry models, the 1,000-m analysis found density of previous incidents, proximity to roads, and density of evergreen to be significant risk factors. This is compared with the density of previous incidents, the proximity of conservation posts, and the proximity to roads showing as significant risk factors for the 150-m analysis for the flora-based dry models. The wet season models also mirrored this variation. Specifically, the 1,000-m fauna-related models identified three significant risk factors—the density of woodland areas, animal sightings, and previous incidents. The 150-m fauna-related models for the wet season, however, identified density of animal sightings and the proximity to roads as significant risk factors. For the flora-related models, the 150-m models identified two risk factors: density of previous incidents and proximity to roads. Conversely, the 1,000-m flora models found four significant risk factors—density of woodland land coverage, previous incidents, evergreen forests, and proximity to roads.

Finally, we compared patrol-based study boundaries to full study area boundaries to assess any differences. For this, we had to operate at a larger unit of analysis (1,000-m cell sizes compared to 150 m). Although the larger unit of analysis may limit its practical use, we felt that its inclusion was conceptually meaningful and provided a nuanced approach in accounting for the patrol-driven nature of environmental crimes documented in PAs. The patrol-based and full study area boundaries displayed variation, although not as dramatic as the comparison between 1,000 and 150 m models. In general, it appears that while a full study area boundary is appropriate to utilize within the scope of RTM, researchers should still be cognizant of potential biases imposed by RBDC and should run patrol-based study boundaries as a comparison.

The results of our study extend the utility of RTM to illegal activities within PAs. Given the user-friendly nature of the approach, RTM may prove to be a useful tool in the management and monitoring of PAs and can be used in conjunction with existing spatial data and software like SMART. Indeed, while SMART is useful in providing data for management purposes (i.e., assessing patrol activity of specific ranger groups), it is limited in its ability to provide actionable information for strategic and tactical decision-making. The data found in SMART, however, can be analyzed using crime forecasting techniques like RTM to further inform site managers.

Our study also demonstrates the applicability of crime pattern theory, and the utility of environmental criminology and crime science more broadly (Moreto & Pires, 2018), in the study of environmental crimes within PAs. By using RTM, we were able to operationalize and assess the environmental backcloth for environmental crimes in Cambodian PAs. Most previous work on crime pattern theory has examined more “traditional” crimes (e.g., burglary) in urban settings while we analyzed illegal activities in a rural setting. By doing so, we demonstrated the applicability of crime pattern theory beyond urban settings. Specifically, our findings point to the importance of pathways and suitable targets for both fauna- and flora-related illegal activities. Animal sightings and previous incidents of both fauna and flora suggest that offenders are likely to focus on locations that provide an abundance of targets, while roads provided reliable means to access and exit the PAs. Importantly, these factors appear to be influenced by the season, further pointing to the need to understand both the spatial and temporal elements of illegal activity in PAs. For example, in our 150-m cell model, roads were a risk factor in the wet season for fauna-related illegal activities but not in the dry season. It may be possible that established pathways (e.g., roads) are more important during the wet season because they provide a trustworthy means to access and exit the PA, whereas backroads or other pathways may be more difficult to transverse during the wet season.

Our study is not without limitations. First, our incident data, which is reliant on RBDC, reflect only illegal activities that have been documented and not illegal activities that occur in locations that are not patrolled. Relatedly, and as shown in our third research objective, RBDC is inherently limited in its ability to account for areas that are not included within the data. This can result in differences in analyses involving boundaries that are determined by patrols or those determined by official designations. Despite these limitations, we believe that by utilizing a patrol-based study area, the assessment of risk factors on known illegal activities still yields important insight on how to manage and monitor PAs. For example, recognizing the role of roads during the wet season informs PA managers to focus their resources on key roadways. In addition, despite the benefits of utilizing RTM, it is possible that additional nuances may be unraveled utilizing other spatial analytical techniques. We recommend that future research assess the risk factors examined in our study using other analytical methods, including spatial regression models. Finally, future studies should assess the applicability of crime pattern theory and RTM in PAs in other countries. Given the variability in terrain and threats, it would be insightful to assess whether our findings hold in other settings, particularly given the variability in terrain, topography, species, and local context.

Management of PAs, with their attendant ecological diversity, economic impact, and vital symbolism, represents an important goal for researchers and policymakers alike. Although only a handful of peer-reviewed studies to date have empirically investigated the spatial risk factors of wildlife crime, this body of literature continues to evolve. The most accessible of these studies offers straightforward assessment of various risk factors that predict illegal activity aimed at fauna and flora. Thus, we believe that in the future, research focused on RTM and other spatiotemporal methodologies to understand environmental crime are likely to yield additional insights that will directly inform policy and practice.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Devin Cowan is a doctoral student within the Criminal Justice Department at the University of Central Florida. Previous BS and MS degrees in Criminal Justice were earned at the University of Central Florida. General research interests include spatial analysis, wildlife crime, and environmental criminology. His recent work has appeared in Policing: A Journal of Policy and Practice and Policing.

William D. Moreto, PhD, is an associate professor in the Department of Criminal Justice at the University of Central Florida, where he is the coordinator for the graduate crime analysis certificate program. His research centers on applying environmental criminology and crime science concepts to wildlife crime, as well as examining wildlife law enforcement from a criminal justice and policing perspective. He is the lead author of Wildlife Crime: An Environmental Criminology and Crime Science Perspective (Carolina Academic Press) and editor of Wildlife Crime: From Theory to Practice (Temple University Press).

Christina Burton is a PhD student of the Department of Criminal Justice at the University of Central Florida. Her research interests include situational crime prevention within wildlife crime, organizational theories and their application to wildlife law enforcement, and the use of spatial analyses to determine predictors of various crime types.

Matt R. Nobles, PhD, is a professor in the Department of Criminal Justice and Doctoral Program in Public Affairs at the University of Central Florida. His research interests include stalking and interpersonal crimes, criminological theory testing, and quantitative methods. His work has appeared in outlets including Justice Quarterly, the Journal of Research in Crime and Delinquency, Criminal Justice and Behavior, the Journal of Quantitative Criminology, Crime & Delinquency, the Journal of Interpersonal Violence, the American Journal of Criminal Justice, and the American Journal of Public Health.

Rohit Singh has over 15 years of experience in wildlife law enforcement and anti-poaching. He has an MSc degree in Wildlife Sciences and a diploma in International Environmental Law. He currently leads the Zero Poaching Initiative of WWF Wildlife Crime Initiative. He is also president of the Ranger Federation of Asia.

References

Andresen

M. A.

(2014). Environmental criminology: Evolution, theory, and practice. Routledge.

Belecky

Singh

Moreto

(2019). Life on the frontline 2019: A global survey of the working conditions of rangers. World Wide Fund for Nature.

Brantingham

P. L.

Brantingham

P. J

. (1993a). Environment, routine, and situation: Toward a pattern theory of crime. In Clarke

R. V.

Felson

(Eds.), Routine activity and rational choice (pp. 259–294). Transaction Publishers.

Brantingham

P. L.

Brantingham

P. J.

(1993b). Nodes, paths, and edges: Considerations on the complexity of crime and the physical environment. Journal of Environmental Psychology, 13, 3–28.

Brantingham

P. L.

Brantingham

P. J.

(1995). Criminality of place: Crime generators and crime attractors. European Journal on Criminal Policy and Research, 3(3), 5–26.

Caplan

J. M.

Kennedy

L. W.

(2013). Risk terrain modeling diagnostics utility (Version 1.0). Rutgers Center on Public Security.

Caplan

J. M.

Kennedy

L. W.

Miller

(2011). Risk terrain modeling: Brokering criminological theory and GIS methods for crime forecasting. Justice Quarterly, 28(2), 360–381.

Caplan

J. M.

Kennedy

L. W.

Piza

E. L.

(2013a). Joint utility of event-dependent and environmental crime analysis techniques for violent crime forecasting. Crime & Delinquency, 59(2), 243–270.

Caplan

J. M.

Kennedy

L. W.

Piza

E. L.

(2013b). Risk terrain modeling diagnostics utility user manual (Version 1.0). Rutgers Center on Public Security.

10.

Caplan

J. M.

Moreto

W. D.

Kennedy

L. W

. (2011). Forecasting global maritime piracy utilizing the risk terrain modeling (RTM) approach to spatial risk assessment. In Kennedy

L. W.

McGarrell

E. F.

(Eds.), Crime and terrorism risk: Studies in criminology and criminal justice (pp. 98–115). Routledge.

11.

Cohen

L. E.

Felson

(1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44, 588–608.

12.

Critchlow

Plumptre

A. J.

Driciru

Rwetsiba

Stokes

E. J.

Tumwesigye

. . . Beale

C. M.

(2015). Spatiotemporal trends of illegal activities from ranger-collected data in a Ugandan national park. Conservation Biology, 29(5), 1458–1470.

13.

Drawve

(2016). A metric comparison of predictive hot spot techniques and RTM. Justice Quarterly, 33(3), 369–397.

14.

Duffy

(2010). Nature crime: How we’re getting conservation wrong. Yale University Press.

15.

Duffy

Massé

Smidt

Marijnen

Büscher

Verweijen

. . . Lunstrum

(2019). Why we must question the militarisation of conservation. Biological Conservation, 232, 66–73.

16.

Dugato

Calderoni

Berlusconi

(2017). Forecasting organized crime homicides: Risk terrain modeling of Camorra violence in Naples, Italy. Journal of Interpersonal Violence. https://doi.org/10.1177/0886260517712275

17.

Dugato

Favarin

Bosisio

(2018). Isolating target and neighbourhood vulnerabilities in crime forecasting. European Journal on Criminal Policy and Research, 24(4), 393–415.

18.

Environmental Investigation Agency. (2008). Environmental crime: A threat to our future.

19.

Felbab-Brown

(2017). The extinction market: Wildlife trafficking and how to counter it. Oxford University Press.

20.

Gavin

M. C.

Solomon

J. N.

Blank

S. G.

(2010). Measuring and monitoring illegal use of natural resources. Conservation Biology, 24(1), 89–100.

21.

Gibbs

Gore

M. L.

McGarrell

E. F.

Rivers

III . (2010). Introducing conservation criminology: Towards interdisciplinary scholarship on environmental crimes and risk. The British Journal of Criminology, 50(1), 124–144.

22.

Gorr

Olligschlaeger

(2002). Crime hot spot forecasting: Modeling and comparative evaluation. National Institute of Justice.

23.

Groff

E. R.

La Vigne

N. G.

(2001). Mapping and opportunity surface of residential burglary. Journal of Research in Crime & Delinquency, 38(3), 257–278.

24.

Groff

E. R.

La Vigne

N. G.

(2002). Forecasting the future of predictive crime mapping. Crime Prevention Studies, 13, 29–58.

25.

Jenks

K. E.

Howard

Leimgruber

(2012). Do ranger stations deter poaching activity in national parks in Thailand? Biotropica, 44(6), 826–833.

26.

Kakira

L. M.

(2010). Monitoring law enforcement effort and illegal activity in selected protected areas: Implications for management and conservation, Democratic Republic of Congo [Unpublished dissertation]. Universiteit Gent.

27.

Kurland

Pires

S. F.

Marteache

(2018). The spatial pattern of redwood burl poaching and implications for prevention. Forest Policy and Economics, 94, 46–54.

28.

Laurance

W. F.

Goosem

Laurance

S. G. W.

(2009). Impacts of roads and linear clearings on tropical forests. Trends in Ecology & Evolution, 24(12), 659–669.

29.

Lynch

M. J.

Barrett

K. L.

Stretesky

P. B.

Long

M. A.

(2017). The neglect of quantitative research in green criminology and its consequences. Critical Criminology, 25(2), 183–198.

30.

Maingi

J. K.

Mukeka

J. M.

Kyale

D. M.

Muasya

R. M.

(2012). Spatiotemporal patterns of elephant poaching in south-eastern Kenya. Wildlife Research, 39(3), 234–249.

31.

Marteache

Pires

S. F.

(2020). Choice structuring properties of natural resource theft: An examination of redwood burl poaching. Deviant Behavior, 41(3), 311–328. https://doi.org/10.1080/01639625.2019.1565518

32.

Moreto

W. D.

(2019). Provoked poachers? Applying a situational precipitator framework to examine the nexus between human-wildlife conflict, retaliatory killings, and poaching. Criminal Justice Studies, 32(2), 63–80.

33.

Moreto

W. D.

Brunson

R. K.

Braga

A. A.

(2017). “Anything we do, we have to include the communities”: Law enforcement rangers’ attitudes towards and experiences of community-ranger relations in wildlife protected areas in Uganda. British Journal of Criminology, 57(4), 924–944.

34.

Moreto

W. D.

Gau

J. M.

(2017). Deterrence, legitimacy, and wildlife crime in protected areas. In Gore

M. L.

(Ed.), Conservation criminology (pp. 45–54). John Wiley & Sons Ltd.

35.

Moreto

W. D.

Pires

S. F.

(2018). Wildlife crime: An environmental criminology and crime science perspective. Carolina Academic Press.

36.

Moreto

W. D.

Piza

E. L.

Caplan

J. M.

(2014). “A plague on both your houses?” Risks, repeats and reconsiderations of urban residential burglary. Justice Quarterly, 31(6), 1102–1126.

37.

N’Goran

P. K.

Boesch

Mundry

N’Goran

E. K.

Herbinger

Yapi

F. A.

Kuehl

H. S.

(2012). Hunting, law enforcement, and African primate conservation. Conservation Biology, 26(3), 565–571.

38.

Nobles

M. R.

(2019). Environmental crime and contemporary criminology: Making a difference. American Journal of Criminal Justice, 44(4), 656–669. https://doi.org/10.1007/s12103-019-09483-7

39.

Onat

Gul

(2018). Terrorism risk forecasting by ideology. European Journal on Criminal Policy and Research, 24(4), 433–449.

40.

Plumptre

A. J.

Fuller

R. A.

Rwetsiba

Wanyama

Kujirakwinja

Driciru

. . . Possingham

H. P.

(2014). Efficiently targeting resources to deter illegal activities in protected areas. Journal of Applied Ecology, 51(3), 714–725.

41.

Rashidi

Wang

Skidmore

Mehdipoor

Darvishzadeh

Ngene

. . . Toxopeus

A. G.

(2016). Elephant poaching risk assessed using spatial and non-spatial Bayesian models. Ecological Modelling, 338, 60–68.

42.

Rengert

G. F.

Piquero

Jones

(1999). Distance decay re-examined. Criminology, 37(2), 427–445.

43.

Szkola

Piza

E. L.

Drawve

(2019). Risk terrain modeling: Seasonality and predictive validity. Justice Quarterly, 1–22. https://doi.org/10.1080/07418825.2019.1630472

44.

Townsley

Sidebottom

(2010). All offenders are equal, but some are more equal than others: Variation in journeys to crime between offenders. Criminology, 48(3), 897–917.

45.

van Doormaal

Lemieux

A. M.

Ruiter

. (2018). Understanding site selection of illegal border crossings into a fenced protected area: A rational choice approach. Crime Science, 7, Article 7.

46.

Watson

Becker

M. S.

McRobb

Kanyembo

(2013). Spatial patterns of wire-snare poaching: Implications for community conservation in buffer zones around national parks. Biological Conservation, 168, 1–9.

47.

Weekers

D. P.

Zahnow

Mazerolle

(2019). Conservation criminology: Modelling offender target selection for illegal fishing in marine protected areas. British Journal of Criminology, 59(6), 1455–1477.

48.

White

Heckenberg

(2014). Green criminology: An introduction to the study of environmental harm. Routledge.

49.

Young

J. A.

van Manen

F. T.

Thatcher

C. A.

(2011). Geographic profiling to assess the risk of rare plant poaching in natural areas. Environmental Management, 48(3), 577–587.