Abstract Globally, tens of thousands of wildlife rangers patrol wide areas and record evidence of poaching activity such as elephant carcasses and snares. Such data have significant potential to inform conservation, but patrols are non-random in space and time, so conclusions from raw patrol data may be biased. Here we model spatial patterns of elephant poaching based on detections of carcasses by ranger patrols in the Zambezi Valley, Zimbabwe (201 carcasses, 2000–2017), using different methodological scenarios to correct for patrol bias. We follow a participatory modelling framework, using interviews with practitioners (rangers and managers) to help build and evaluate these models. We found that poaching patterns in the bias-corrected scenarios differed among themselves and from the uncorrected scenario. Practitioners interrogated the credibility of the predictions in each scenario and thus helped discern true poaching patterns from those explained by patrol bias. We uncovered proximity to water as the strongest driver of poaching, likely reflecting both poacher and elephant behaviour. Our results show that it is essential to account for observer bias before developing management actions (such as ranger patrol strategies) from raw observational data. We further demonstrate the value of combining multiple lines of evidence (statistical models and interview responses) for more robust inference in the face of uncertainty.

中文翻译：

---

游骑兵和建模师合作建立和评估非洲大象偷猎的空间模型

摘要 在全球范围内，数以万计的野生动物护林员在广阔的区域巡逻，并记录大象尸体和圈套等偷猎活动的证据。此类数据具有为保护提供信息的巨大潜力，但巡逻在空间和时间上是非随机的，因此原始巡逻数据的结论可能有偏差。在这里，我们根据津巴布韦赞比西河谷的护林员巡逻队发现的尸体（201 具尸体，2000-2017 年）对大象偷猎的空间模式进行建模，使用不同的方法论场景来纠正巡逻偏差。我们遵循参与式建模框架，通过与从业人员（护林员和经理）的访谈来帮助构建和评估这些模型。我们发现偏差校正场景中的偷猎模式在它们之间和未校正场景中是不同的。从业者在每种情况下询问预测的可信度，从而帮助从巡逻偏见解释的那些中辨别出真正的偷猎模式。我们发现靠近水是偷猎的最强驱动因素，这可能反映了偷猎者和大象的行为。我们的结果表明，在根据原始观察数据制定管理行动（例如护林员巡逻策略）之前，必须考虑观察者的偏见。我们进一步证明了在面对不确定性时结合多条证据（统计模型和访谈回复）进行更稳健推理的价值。可能反映了偷猎者和大象的行为。我们的结果表明，在根据原始观察数据制定管理行动（例如护林员巡逻策略）之前，必须考虑观察者的偏见。我们进一步证明了在面对不确定性时结合多条证据（统计模型和访谈回复）进行更稳健推理的价值。可能反映了偷猎者和大象的行为。我们的结果表明，在根据原始观察数据制定管理行动（例如护林员巡逻策略）之前，必须考虑观察者的偏见。我们进一步证明了在面对不确定性时结合多条证据（统计模型和访谈回复）进行更稳健推理的价值。