Bourke Scott D, Bennington Steph M, Turner Sam, Monks Joanne M
Department of Zoology University of Otago, Ōtākou Whakaihu Waka Dunedin New Zealand.
Department of Marine Science University of Otago, Ōtākou Whakaihu Waka Dunedin New Zealand.
Ecol Evol. 2025 Sep 11;15(9):e72155. doi: 10.1002/ece3.72155. eCollection 2025 Sep.
The potential utility of Species Distribution Models (SDMs) in conservation is apparent. One application for rare or highly cryptic taxa is using model predictions to increase the efficiency of sampling effort. Though this method is potentially powerful, the accuracy of model predictions is rarely tested in the field. Further, uncertainty remains about whether validation statistics reflect true model performance, particularly for species of high conservation concern. We assessed the usefulness of SDMs for predicting the distribution of six species of lizards in the Mackenzie Basin (Te Manahuna), Aotearoa New Zealand (NZ). We built MaxEnt models using readily available occurrence data and a publicly available suite of environmental predictors. We validated model performance using both data partitioning and independent occurrence records collected in the 2022/23 austral summer. Cross-validation suggested that the top models for each species generated reasonably accurate predictions; however, for common species, predictive accuracy decreased notably when validating with independent data. Models for rare species performed more variably when validated with independent data; however, these models were overfit and based on few data, making it difficult to have confidence in the resulting abstractions. We suggest that limitations in historical occurrence data, current knowledge of species ecology and low resolution of predictor data likely restrict the relevance of predictive modelling for NZ lizard species. Whilst attractive to species managers and easy to generate, predictive models should be subject to ground-truthing with temporally relevant data prior to being used to inform sampling effort.
物种分布模型（SDMs）在保护工作中的潜在效用是显而易见的。对于珍稀或高度隐秘的分类群，一种应用是利用模型预测来提高采样工作的效率。尽管这种方法可能很强大，但模型预测的准确性很少在实地进行检验。此外，关于验证统计数据是否反映了模型的真实性能仍存在不确定性，特别是对于具有高度保护价值的物种。我们评估了物种分布模型对于预测新西兰麦肯齐盆地（特马纳胡纳）六种蜥蜴分布的有用性。我们使用现成的出现数据和一套公开可用的环境预测因子构建了最大熵模型。我们使用数据划分和在2022/23南半球夏季收集的独立出现记录来验证模型性能。交叉验证表明，每个物种的顶级模型产生了相当准确的预测；然而，对于常见物种，使用独立数据进行验证时预测准确性显著下降。稀有物种的模型在使用独立数据进行验证时表现出更大的变异性；然而，这些模型过度拟合且基于的数据较少，因此难以对所得的抽象结果有信心。我们认为，历史出现数据的局限性、当前对物种生态学的了解以及预测因子数据的低分辨率可能限制了预测模型对新西兰蜥蜴物种的相关性。虽然预测模型对物种管理者有吸引力且易于生成，但在用于指导采样工作之前，应使用与时间相关的数据进行实地验证。
