A systematic review of adaptive wildlife management for the control of invasive, non‐native mammals, and other human–wildlife conflicts

1. We are entering an era where species declines are occurring at their fastest ever rate, and the increased spread of non-native species is among the top causes. High uncertainty in biological processes makes the accurate prediction of the outcomes of management interventions very challenging. Adaptive management (AM) offers solutions to reduce uncertainty and improve predictability so that the outcomes of interventions can continuously improve. 
2. We quantitatively assess the extent to which AM is used for managing vertebrates, with a focus on invasive non-native species (INNS). Using the Web of Science, we evaluated 3992 articles returned by the search terms ‘adaptive management’ or ‘adaptive harvest management’ against seven recommended elements of AM (engagement with stakeholders, defining objectives, forecasting and estimating uncertainty, implementing management, monitoring populations, adjusting management in response to monitoring, and improving forecasting and reducing uncertainty in response to monitoring populations). 
3. The use of AM for vertebrates was reported in 56 (1%) of the evaluated studies; including four for managing INNS. Of these, ten studies excluding INNS and no studies of INNS management implemented all seven recommended elements of AM. Those elements infrequently implemented were: the use of analysis or models to forecast and represent uncertainty (44%) and the feedback of monitoring data to improve forecasting and reduce uncertainty (25%). 
4. Complete active AM has rarely been implemented and reported for managing INNS, despite the significant advantages it offers. Among studies purporting to have implemented AM, most did not use analyses or models to forecast and represent uncertainty, while most defined objectives, implemented management, and monitored populations. 
5. Improvements to ongoing control programmes and much broader adoption of the AM approach are required to increase the efficiency and success of INNS management campaigns and reduce their negative impacts on native species.


INTRODUCTION
Anthropogenic impacts on biological systems are widely accepted to be the cause of recent mass global species declines (Sarukhan et al. 2005), often leading to extinctions (Barnosky et al. 2011). Among the top drivers of species declines are: habitat loss and fragmentation (Collinge & Forman 2009), invasive non-native species (INNS, Vitousek et al. 1997, Simberloff 2010, Blackburn et al. 2019) and climate change (Thomas et al. 2004). Effective and efficient management to prevent these declines is essential (Simberloff 2010), especially where financial resources are limited. Effective management of INNS is challenging, as species can become numerous and widespread before they are detected, and their impacts (often on native species and ecosystems) may require novel and long-term management methods. Examples of invasive mammal species subject to long-term management in the UK include American mink Neovison vison, grey squirrel Sciurus carolinensis, fallow deer Dama dama and muntjac deer Muntiacus reevesi. Relatively recent eradications include muskrat Ondatra zibethicus and coypu Myocastor coypus.
In complex ecological systems, predicting the optimal management intervention among a range of options is difficult, often due to a range of unknowns and uncertainties (Ward et al. 2020). Uncertainties may arise from environmental variation, sampling variation or the response of populations to the management methods. Adopting an adaptive, rather than a fixed approach to management allows management actions to change based on new knowledge while maintaining the same objective (Franklin et al. 2007). Adaptive management (AM) has frequently been proposed as the most effective method for managing ecological resources (here we use the term to include INNS species which may need to be managed or eradicated), where there is high uncertainty (Holling 1978). AM is defined as "flexible decision making that can be adjusted in the face of uncertainties as outcomes from management actions and other events become better understood. Careful monitoring of these outcomes both advances scientific understanding and helps adjust policies or operations as part of an iterative learning process" (National Research Council 2004). The philosophy of AM was first described for adaptive decision making in fisheries management (Beverton & Holt 1957); it was later defined as a term and formalised as a conceptual framework (Holling 1978, Walters & Hilborn 1978.
The aim with using AM is to reduce uncertainty by gaining knowledge of the system using either an active or a passive approach (Williams 2011a). In active AM, learning about the system is a distinct objective, and multiple approaches are tested simultaneously and may result in a sub-optimal management intervention being implemented to improve learning about the system. In passive AM, learning about managing the ecological resource is a distinct objective, and therefore a known sub-optimal management intervention is not implemented (Larson et al. 2013). A passive approach is often adopted where it is impossible to implement multiple interventions concurrently, managers are unwilling to implement suboptimal management interventions, or there are insufficient funds for the extra time that may be needed for learning about the system (Gregory et al. 2006, Hughes et al. 2007. While a passive approach is more likely to result in effective management than where no AM is applied, it does not facilitate learning about the critical features of the system, and putative relationships between environmental variables and the ecological resource may be misinterpreted, hampering evaluation of the effects of management actions on the ecological resource. One of the most well-known examples of AM for harvesting populations is to support the hunting of the North American mallard Anas platyrhynchos (Nichols et al. 2007). The resource objectives were to maximise the long-term cumulative harvest utility, and the system objectives were to learn about the relationship between harvest rate and survival rate.
The management options were liberal, moderate and conservative hunting regulations; daily bag limits were specified for each option. The underlying models which were updated included two models of different levels of density dependence of mallard reproductive rates and a maximal and minimal model of the hunting mortality on total annual survival (Nichols et al. 2007). Population estimates ranged from 6 to 12 million birds over 15 years. AM enabled the additive hunting mortality and weak density-dependent reproductive rate model to be identified as the optimal model, resulting in population sizes being predicted more accurately, and hunting limits could be more liberal (Johnson et al. 2002 invertebrates, and domestic animals. We removed articles that included AM or adaptive harvest management as a keyword but did not state that they used AM. Filtering was conducted by sequentially reviewing the title or abstract of the article, or the whole article, as necessary. We identified 183 articles citing AM for vertebrates: 25 (14%) were review articles without case studies, 63 (34%) were theoretical studies where a framework was described and/or the study suggested AM should be used but it was not implemented, 27 (15%) reported on the initial stages or parts of an AM study for vertebrates, and the remainder (68, 37%) were studies of AM of vertebrates (including review articles with a case study). Within these, we combined articles that reported on different aspects of the same AM study into a single record, resulting in 56 unique studies. We recorded the taxa of the study species, the region of the study, and the publication year of the article. Studies were further reviewed to categorise AM as active or passive, and the reason for choosing AM. We defined the type of AM as active if the objective was system orientated, and passive if the objective was resource orientated.
We assessed if each study reported the seven recommended elements of AM described by Williams (2011b) as follows: (1) engagement with stakeholders, (2) identification of objectives, (3) forecasting and representing uncertainty, (4) implementation of more than two management options (Williams 2011b states that a range of management options should be implemented, which we interpreted as more than two), (5) implementation of a regular monitoring programme, (6) management practices that are adjusted in response to the results of monitoring, and (7) monitoring data that are used to improve forecasting ability and reduce uncertainty. We assessed studies against meeting all these elements and against our reduced criteria of meeting elements 2, 3, 5, 6 and 7 (see Discussion for an explanation of our removal of elements 1 and 4). Of fully implemented studies that included (2) objectives and (6) modified management in response to monitoring, we calculated the percentage that was successful (defined as meeting their stated objectives).

RESULTS
Among the 56 AM studies of vertebrates, the term AM was used in the title in 38%, as a keyword in 45%, and in the abstract in 70% of studies. Ten studies implemented all seven recommended elements of AM described by Williams (2011b), while 12 implemented at least our five reduced elements (excluding stakeholder engagement and the implementation of a range of management options; Tables 1, 2). Among these 12 studies, examples of AM were active (three studies) and passive (nine). Since 1997, there has been an increasing trend in the number of AM studies produced per year (Fig. 1). AM studies are most common in North America (43% of studies), Europe (25%) and Australasia (16%; Fig. 2). The highest percentages of studies that implemented our reduced elements of AM were in North America (25%), Australasia (22%); and Europe (21%). AM was most commonly used to manage mammals (45% of studies), birds (23%) and fish (18%), only two studies were conducted on reptiles or amphibians (Fig. 3). The highest percentages of studies that implemented our reduced elements of AM were studies conducted on birds (38%), and fish (30%) defined objectives and management modified in response to monitoring (n = 16), 50% were successful, 25% were partially successful, and 19% were unsuccessful in meeting their objectives (the remaining 6% did not state the outcomes in relation to their objectives); the two studies on managing INNS were both successful.

DISCUSSION
Improving the effectiveness and efficiency of management is increasingly important for the conservation and sustainable use of wildlife; a particular challenge is INNS management where numbers and impacts are increasing in the face of limited resources (Vitousek et al. 1997, Simberloff 2010. When operating under high uncertainty, AM is an effective way of managing human-wildlife conflicts (Holling 1978, Walters 1990. We found that the use of the term AM has been increasing in the literature. However, when we examined the application of AM, we found that most studies had not implemented all the elements of AM recommended by Williams (2011b). Only 10 of the 56 studies identified as using AM for vertebrate management or conservation reported full implementation of the recommended AM elements described by Williams (2011b), and none of these described Our study showed further quantitative evidence of the misrepresentation of the term AM, despite the publication of detailed descriptions of AM (Williams 2011a, b). For example, the percentage of articles implementing our reduced elements of AM was 22% for articles published before 2012 and 25% after 2011. A lack of uptake of AM may be due to confusion about what AM is how it should be implemented (e.g. Lee 1999, Rist et al. 2013).
We found that the number of studies that reported the implementation of all seven recommended elements of AM occurred after 2000, but partial implementation was typical.
Most studies defined objectives and conducted regular monitoring, while few studies forecasted the effects of management and estimated uncertainty at the start of the study, and continued to do so with the addition of new monitoring information.
Ongoing evaluation is one of the defining characteristics of AM. We found several studies reported as AM for vertebrates in which the authors assessed the effect of management on an ecological resource at the end of a multi-year study (e.g. Whitehead et al. 2008), rather than assessing and re-evaluating the effects of management on the resource and whether they had met their objectives throughout the study. In AM, predictive models or analyses are used to forecast and estimate uncertainty. These models or analyses are then continually updated with new monitoring information. A common misconception is that AM is something that is assessed once at the end of the study, rather than a cyclic, reiterative process where monitoring, management, prediction, and evaluation are continually conducted, enabling management to be modified and improved continually. To avoid misunderstanding, we advocate that the term AM is used only to describe programmes that implement all seven of the elements defined by Williams et al. (2011b), or our five reduced elements (i.e. defining objectives, forecasting and estimating uncertainty, implementing regular monitoring, adjusting management in response to monitoring, and updating forecasts and uncertainty estimates in response to new monitoring data).
We found that more of the recommended AM elements were likely to be implemented where the managed resource has a monetary value, such as harvest management, and that none of the INNS studies fully implemented all the recommended AM elements or our reduced AM elements. The required commitment of funds for long-term, replicated studies, often required for AM, may explain the lack of use (e.g. Gregory et al. 2006, Hughes et al. 2007). Failure of AM studies at the planning stage, possibly due to lack of funds, has been previously identified (Walters 1997). Collaboration between scientists and practitioners facilitates the use of analyses or models to understand underlying structures or processes in management practice, an element of AM implemented in only 44% of the studies we reviewed. Models or analyses are essential for two reasons: they play a crucial role in representing structural or process uncertainty, and they link potential management actions to the ecological resource consequences (Williams 2011a, b). Implementing analyses or models for complex ecological environments is perhaps one of the most significant barriers in AM, reflected by the fact that contrasting models to forecast resource changes through time were identified in only 14% of studies.
One of the elements of AM recommended by Williams (2011b) was a reasonable range of management actions, which we quantified as being more than two management managing INNS may limit the uptake of volunteers, and their motivation, continuity, and training are potential drawbacks in the use of a volunteer workforce for INNS management.
Engagement with stakeholders is a recommended aspect of AM, and while we agree with this recommendation, we also acknowledge that engagement is study dependent and not necessarily an essential element of AM. Hence, we excluded engagement from our reduced list of AM elements for analysis. Engaging with stakeholders and including diverse perspectives is likely to result in unbiased objectives and avoid inappropriate or unnecessary constraints on management (Beierle 2002, Williams 2011b. However, if, at the outset, AM is not designed into a project, then engagement with all appropriate stakeholders is unlikely to be achieved. Working with and achieving consensus among stakeholders with diverse aims is challenging, but is required if common objectives are to be reached. Achieving collaboration requires careful management, as achieved by Bryce et al. (2011) andJohnson et al. (2002).
Among the studies reviewed here, 45% did not achieve engagement with stakeholders, which may be one of the reasons why they were not more successful. It may be that stakeholders were engaged but not reported in articles; however, if this is the case, it demonstrates the lack of importance afforded by authors to this stage of the AM process. Late engagement with stakeholders can lead to increased conflict when applied to the management of mammals, particularly the removal of INNS (Crowley et al. 2017). An example of this is the attempted control of grey squirrels in Italy, which was delayed due to the actions of animal rights activists which eventually resulted in the abandonment of the study (Bertolino & Genovesi 2003 recommend that practitioners define a clear AM framework at the outset, so that modifying management, forecasting, and reducing uncertainty in response to monitoring data will not be afterthoughts, but will instead be structured into the study. As stated above, AM is a cyclical process of continuing to reduce uncertainty and improve the forecasting of the effects of management, which is adjusted in light of new knowledge gained by monitoring populations to achieve the study objectives, rather than a single assessment of the effects of management on a population or a trial-and-error process (which AM is often confused with       Table 2. The 12 studies of adaptive management (AM) of wildlife in which at least five elements of AM were included. Ten studies implemented all seven elements of adaptive management (AM) as recommended by Williams (2011b): (1) engagement with stakeholders, (2) identification of objectives, (3) forecasting and estimating uncertainty, (4) implementation of more than two management options, (5) implementation of a regular monitoring programme, (6) management practices adjusted in response to the results of monitoring and (7) monitoring data feedback to improve forecasting ability and reduce uncertainty. Two studies (indicated with *) implemented our reduced number of five elements of AM, excluding (1) and (4). Type of AM Reason for AM Country Taxa Species