Investigating the influence of data quality on ecological niche models for alien plant invaders

Ecological niche modeling is a method designed to describe and predict the geographic distribution of an organism. This procedure aims to quantify the species-environment relationship by describing the association between the organism's occurrence records and the environmental characteristics at these points. More simply, these models attempt to capture the ecological niche that a particular organism occupies. A popular application of ecological niche models is to predict the potential distribution of invasive alien species in their introduced range. From a biodiversity conservation perspective, a pro-active approach to the management of invasions would be to predict the potential distribution of the species so that areas susceptible to invasion can be identified. The performance of ecological niche models and the accuracy of the potential range predictions depend on the quality of the data that is used to calibrate and evaluate the models. Three different types of input data can be used to calibrate models when producing potential distribution predictions in the introduced range of an invasive alien species. Models can be calibrated with native range occurrence records, introduced range occurrence records or a combination of records from both ranges. However, native range occurrence records might suffer from geographical bias as a result of biased sampling or incomplete sampling. When occurrence records are geographically biased, the underlying environmental gradients in which a species can persist are unlikely to be fully sampled, which could result in an underestimation of the potential distribution of the species in the introduced range. I investigated the impact of geographical bias in native range occurrence records on the performance of ecological niche models for 19 invasive plant species by simulating two geographical bias scenarios (six different treatments) in the native range occurrence records of the species. The geographical bias simulated in this study was sufficient to result in significant environmental bias across treatments, but despite this I did not find a significant effect on model performance. However, this finding was perhaps influenced by the quality of the testing dataset and therefore one should be wary of the possible effects of geographical bias when calibrating models with native range occurrence records or combinations there of. Secondly, models can be calibrated with records obtained from the introduced range of a species. However, when calibrating models with records from the introduced range, uncertainties in terms of the equilibrium status and introduction history could influence data quality and thus model performance. A species that has recently been introduced to a new region is unlikely to be in equilibrium with the environment as insufficient time will have elapsed to allow it to disperse to suitable areas, therefore the occurrence records available would be unlikely to capture its full environmental niche and therefore underestimate the species' potential distribution. I compared model performance for seven invasive alien plant species with different simulated introduction histories when calibrated with native range records, introduced range records or a combination of records from both ranges. A single introduction, multiple introduction and well established scenario was simulated from the introduced range records available for a species. Model performance was not significantly different when compared between models that were calibrated with datasets representing these three types of input data under a simulated single introduction or multiple introduction scenario, indicating that these datasets probably described enough of the species environmental niche to be able to make accurate predictions. However, model performance was significantly different for models calibrated with introduced range records and a combination of records from both ranges under the well established scenario. Further research is recommended to fully understand the effects of introduction history on the niche of the species.生态位建模（Ecological Niche Modeling）是一类用于描述并预测某一生物地理分布的方法。该流程旨在通过量化物种-环境关系，刻画物种出现记录与对应点位环境特征之间的关联。简言之，此类模型旨在捕捉某一特定生物所占据的生态位。 生态位模型的一项热门应用，是预测外来入侵物种在其入侵扩散范围内的潜在分布。从生物多样性保护的视角来看，对外来入侵开展主动管理的前置策略，便是预测物种的潜在分布范围，以此识别易受入侵的区域。 生态位模型的性能与潜在分布预测的准确性，取决于用于校准与评估模型的数据质量。在外来入侵物种的入侵范围内生成潜在分布预测时，可采用三类不同的输入数据来校准模型：可使用原生分布范围的出现记录、入侵分布范围的出现记录，或是同时结合两类分布范围的记录。 然而，原生分布范围的出现记录可能因采样偏差或采样不全而存在地理偏倚。当出现记录存在地理偏倚时，物种得以存续的潜在环境梯度往往无法被完整采样，这可能导致低估该物种在入侵范围内的潜在分布。 本研究通过在物种的原生分布范围出现记录中模拟两类地理偏倚场景（共六种不同处理方式），探究了原生分布范围出现记录的地理偏倚对19种入侵植物物种的生态位模型性能的影响。本研究模拟的地理偏倚足以在不同处理间造成显著的环境偏倚，但尽管如此，并未发现其对模型性能存在显著影响。不过，这一结论或许受限于测试数据集的质量，因此在使用原生分布范围出现记录或其组合数据校准模型时，应当警惕地理偏倚可能带来的影响。 其次，可使用从物种入侵分布范围获取的记录来校准模型。但当使用入侵分布范围的记录校准模型时，物种入侵平衡状态与入侵历史的不确定性可能影响数据质量，进而影响模型性能。新近被引入新区域的物种往往尚未与环境达到平衡，因为其尚未有足够时间扩散至所有适宜生境，因此现有的出现记录大概率无法完整涵盖其全部生态位，从而低估该物种的潜在分布。 本研究针对7种外来入侵植物物种，比较了分别使用原生分布范围记录、入侵分布范围记录，或是结合两类分布范围的记录进行校准时的模型性能，同时模拟了三种不同的入侵历史场景：单次引入、多次引入以及种群已完全建立。在模拟单次引入或多次引入的场景下，使用这三类输入数据集校准的模型，其性能并无显著差异，表明这些数据集已足够捕捉物种的环境生态位，能够生成准确的预测结果。但在种群已完全建立的场景下，使用入侵分布范围记录与结合两类记录校准的模型，其性能存在显著差异。建议开展进一步研究，以充分阐明入侵历史对物种种群生态位的影响。