当前位置: X-MOL 学术Ecol Modell › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Fine-tuning niche models matters in invasion ecology. A lesson from the land planarian Obama nungara.
Ecological Modelling ( IF 3.1 ) Pub Date : 2021-08-05 , DOI: 10.1016/j.ecolmodel.2021.109686
Yoan Fourcade 1
Affiliation  

Despite the apparent simplicity of species distribution modelling approaches, the reliability of their predictions depends on the application of a number of good practices regarding the input data and the parametrisation of algorithms. In the context of invasion biology, inadequate modelling procedures may lead to erroneous conclusions regarding the potential spread of introduced species. However, clear guidelines for implementing these recommendations are often lacking, confusing or simply unknown by non-modeller end-users. Here, taking as an example the introduced land planarian Obama nungara, I fitted MaxEnt models applying six recommended processing steps with respect to sampling bias, predictor choice, training area, evaluation and hyperparameter tuning, separately or implemented together. I compared the resulting outputs to a model fitted with all default settings. All models differed from one another and from the default model, highlighting the importance of considering all these parameters when fitting species distribution models. However, the model that incorporated all fine-tuning methods was by far the most dissimilar, predicting much larger suitable areas globally, including in Africa where O. nungara has not been found so far. A closer examination suggested that it is likely a result of lower overfitting. This is a demonstration that modelling settings matter a lot, to the point that fined-tuned or default models may lead to considerably different conclusions when applied to invasive species.



中文翻译:

微调生态位模型在入侵生态中很重要。陆地涡虫奥巴马 nungara 的教训。

尽管物种分布建模方法明显简单,但其预测的可靠性取决于有关输入数据和算法参数化的许多良好实践的应用。在入侵生物学的背景下,不适当的建模程序可能会导致关于引入物种的潜在传播的错误结论。然而,非建模者最终用户通常缺乏、混淆或根本不知道实施这些建议的明确指南。在这里,以引入的陆地涡虫奥巴马 nungara 为例,我安装了 MaxEnt 模型,应用六个推荐的处理步骤,分别涉及采样偏差、预测变量选择、训练区域、评估和超参数调整,单独或一起实施。我将结果输出与配备所有默认设置的模型进行了比较。所有模型彼此不同,也与默认模型不同,突出了在拟合物种分布模型时考虑所有这些参数的重要性。然而,结合所有微调方法的模型是迄今为止最不同的,预测全球范围内更大的合适区域,包括在非洲,O. nungara至今未发现。更仔细的检查表明,这可能是过拟合程度较低的结果。这表明建模设置非常重要,以至于在应用于入侵物种时,微调或默认模型可能会导致截然不同的结论。

更新日期:2021-08-05
down
wechat
bug