Evidence-based interventions designed to reduce wildlife-caused losses are essential for human-wildlife coexistence. The lack of systematic summarization of research effort and evidence makes it challenging for researchers, managers, and policymakers to prioritize interventions for evaluation and implementation. 

Here, we compiled experimental case studies of nonlethal technical interventions designed to reduce the losses of crops, livestock, and fishery catches caused by terrestrial carnivores, elephants, farmland birds, and marine fauna worldwide. Then, we summarized the research effort and the performance of interventions by their sensory stimuli and target animals.

We found that: (i) 54 out of 88 interventions included in this study had statistically effective evidence, where only 39% (21/54) were evaluated with more than three experiments (ii) physical-, sound-, chemical-, and light (or visual) -based interventions were the most in numbers and their performance varied greatly; (iii) farmland birds, seabirds, and cetaceans were the most studied animal groups while there are only a few experiments for elephants; and (iv) the interventions for marine fauna generally had no impact on the target catch of fisheries.

Syntheses and applications: Our results indicated that collective effort is needed to further evaluate interventions using various sensory stimuli and launch incentive programs to motivate the implementation of interventions, particularly related to marine fauna conservation. Our synthesis could be helpful for stakeholders to tackle the negative human-wildlife interactions outlined as Target 4 of the Kunming-Montreal Global Biodiversity Framework.

Compilation of the list of case studies

We collected case studies from the six syntheses (van Eeden et al. 2018b; Khorozyan 2022; Lorand et al. 2022; Montgomery et al. 2022; Huang et al. 2023; Huang et al. 2024), which were based on systematic search schemes and provided the most comprehensive lists of case studies to our knowledge. For terrestrial carnivores, van Eeden et al. (2018b) compiled 114 case studies based on four systematic reviews (Miller et al. 2016; Treves, Krofel & McManus 2016; Eklund et al. 2017; van Eeden et al. 2018a); Lorand et al. (2022) compiled 103 studies; and Khorozyan (2022) compiled 244 studies to summarize the study designs. For elephants, Montgomery et al. (2022) compiled 95 studies. For farmland birds, Huang et al. (2023) compiled 48 studies. For marine fauna, Huang et al. (2024) compiled 119 studies. After removing the duplicated studies, 594 case studies were left for the initial list (Fig, S1).

Inclusion criteria

These six syntheses had different inclusion criteria about target animal groups, types of interventions, study designs, reported data, and statistical approaches (Table S1). According to our research objectives, we set six criteria to make the research effort comparable. If a case study did not meet one of the following criteria, we would exclude it from further review.

Language. The studies had to be published in the working languages of the reviewing team (i.e., English and Chinese).

Problem. The involved animals should be wild and free-ranging animals. The affected properties should be crops, livestock, and fishery catches. Notably, while terrestrial animals are often intentionally controlled, marine megafauna usually is entangled or hooked unintentionally when they feed on baits and captured fishes, leading to injuries or drowning and further damage to fishing gear (Wilson et al. 2014; Gilman et al. 2023).

Nonlethal technical intervention. Interventions must be nonlethal technical measures and adjusting fishing or farming time were excluded.

Comparisons. While the random set-up of a strict experiment is often impossible for case studies in this research field, we required the set-up of experimental comparison to be somehow controlled to reduce various confounding biases in the wild.

Outcome. The effects of interventions on properties should be measured. The measurements include weight of crop yield, area or weight of crop damage, number of livestock lost, and weight or number of captured fish. We excluded experiments based on indirect behavioral indicators, such as the number of visitations, contact rate (e.g., contact rate between seabirds and gears near baited hooks), and respondents’ perceived effectiveness. For translocations, the change of losses of the pre-release site and the post-release site should be measured.

Statistical test. Significance testing of the outcome between the treatments and controls should be reported or is possible to calculate from the reported data.

In general, to include more types of interventions, we relaxed the inclusion criteria on case studies about the reported data compared to previous meta-analysis (van Eeden et al. 2018a; Huang et al. 2023). The meta-analysis approaches are often used to evaluate the overall impact of an intervention based on quantitative effect size (e.g., Hedge’s d and g) calculated from detailed data (e.g., the mean, standard deviation, and sample size) obtained from multiple case studies and may detect statistically significant effects of interventions, even when synthesizing insignificant case studies of low statistical power (Koricheva & Gurevitch 2013; Borenstein et al. 2021). However, case studies that do not report the required data for calculating the effect size were usually excluded, such as the studies for fyke net modification for pinniped (Königson et al. 2007) and umbrella-and-stones system for seabirds (Brown et al. 2010). Besides, many interventions were evaluated only by a single experiment (or one effect size), such as using decoy fields (Hagy, Linz & Bleier 2008), seed coat (Decker, Avery & Way 1990), and caffeine (Avery et al. 2005) for farmland birds, were often excluded from the meta-analysis (Huang et al. 2023). Meanwhile, we tightened the inclusion criteria to that of previous narrative reviews on the study designs and outcome measurement.

With our different criteria from previous syntheses, we revisited the inclusion process of Huang et al. (2023) and Huang et al. (2024) and identified an additional 13 and 8 studies, respectively, which did not meet the previous inclusion criteria of the meta-analysis but met the criteria of this study. Therefore, there were 615 case studies in the final list and 605 of them are available online. The inclusion process of this study was documented in the supplementary material (Fig. S1).

Data extraction and compilation

For each case study, we extracted information on the types of interventions and their outcomes. An intervention was recognized mainly based on the combination of the target animal groups and device or gear its used. We classified intervention outcomes as either statistically effective, insignificant, or counter-productive according to the original case studies. Statistically effective means that an intervention showed statistical significance in reducing the loss or increasing the yield of properties; counter-productive means that an intervention showed statistical significance in increasing the loss or reducing the yield of properties; and statistically insignificant means that an intervention had no statistically significant impact on the properties but this does not necessarily mean the effect of an intervention is absent due to statistical powers (Borenstein et al. 2021). For studies that reported the effects of interventions on multiple livestock (e.g., sheep, goat, and cattle) or target fishes, we extracted the overall statistical test, if available. We treated the level of 0.05 as statistically significant.

We compiled the data about the environment (i.e., the terrestrial or marine), animal groups, sensory stimuli, and geographical information (e.g., countries and major fishing areas). According to management and research practices, we classified the animals into terrestrial carnivores, elephants, farmland birds, seabirds, cetaceans, pinnipeds, and elasmobranchs. The sensory stimuli could be light (or visual, e.g., flashing light and movable predator models), sound (e.g., acoustic pingers), chemical (e.g., anthraquinone), electricity (e.g., electric wire), magnetic (e.g., permanent magnet), multi-sensory stimuli (multimodality, hereafter, e.g., alternative crops or baits), and physical approach (e.g., fortified fences and line weightings to avoid depredations), and combinations of the former stimuli (e.g., toriline as the light-physical and guardian animals as the multimodality-physical) (Fig. 1 and Table S2). We assigned each study location (if available) to the countries’ territory, Economic Exclusive Zone of a specific country, and the major fishing area as defined by the Food and Agriculture Organization.

We classified feeding stations or decoy fields as physical approaches that lead wildlife away. We classified the bird perch and aerial-applied herbicide to habitat as two indirect interventions because they relied on the presence of aggressive birds and the removal of weeds at roost habitats. We only included statistically effective interventions when a study aims to examine the outcome of multiple similar interventions without making any assumptions about which one would work. Examples of such interventions included the colors of fruit-protective bags (Mutisya et al. 2016; Wen et al. 2016)