Scholarlink

<p>Predicting the potential geographical distribution of alien invasive species is essential for effective prevention and management of biological invasions. Existing invasive species prediction (ISP) models typically utilize original data of environmental factors directly as inputs, which may introduce uncertainty into predictive outcomes due to inherent random errors in the data. However, such random errors are often overlooked in ISP modeling, potentially compromising the reliability of spatial distribution predictions. This study examines the impact of random errors in environmental factors on uncertainty in modeling and predicting species distributions. We first applied low-pass filtering to original environmental data to reduce random errors, while also introducing varying proportions of random noise to amplify error effects. Using presence-absence data of invasive species, we constructed multiple ISP models based on different machine learning algorithms. The case study in the Yangtze River Economic Belt, China, demonstrated that low-pass filtering effectively reduces random errors in environmental data of invasive species <i>Erigeron annuus</i>, thereby diminishing ISP uncertainty. Conversely, as the proportion of random errors increases from 5% to 20%, ISP uncertainty progressively escalates. Both the choice of machine learning models and the magnitude of random errors significantly influence ISP modeling outcomes. Furthermore, in the analysis of ISP interpretability uncertainty using the SHAP method that quantifies each environmental factor’s importance, it was observed that an increase in random errors causes deviations in environmental factor importance values and even changes in the dominant environmental factor. This research provides valuable insights for achieving more accurate and reliable predictions of potential geographical distributions of invasive species.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Uncertainties of Predicting the Potential Geographical Distribution of Species Invasion by Considering Random Errors in Environmental Factors

  • 期刊论文

  • 发表日期：

    2026年5月8日

查看全文
  • Jian Xie,
  • Haifeng Zhang,
  • Wentao Yang

摘要

Predicting the potential geographical distribution of alien invasive species is essential for effective prevention and management of biological invasions. Existing invasive species prediction (ISP) models typically utilize original data of environmental factors directly as inputs, which may introduce uncertainty into predictive outcomes due to inherent random errors in the data. However, such random errors are often overlooked in ISP modeling, potentially compromising the reliability of spatial distribution predictions. This study examines the impact of random errors in environmental factors on uncertainty in modeling and predicting species distributions. We first applied low-pass filtering to original environmental data to reduce random errors, while also introducing varying proportions of random noise to amplify error effects. Using presence-absence data of invasive species, we constructed multiple ISP models based on different machine learning algorithms. The case study in the Yangtze River Economic Belt, China, demonstrated that low-pass filtering effectively reduces random errors in environmental data of invasive species Erigeron annuus, thereby diminishing ISP uncertainty. Conversely, as the proportion of random errors increases from 5% to 20%, ISP uncertainty progressively escalates. Both the choice of machine learning models and the magnitude of random errors significantly influence ISP modeling outcomes. Furthermore, in the analysis of ISP interpretability uncertainty using the SHAP method that quantifies each environmental factor’s importance, it was observed that an increase in random errors causes deviations in environmental factor importance values and even changes in the dominant environmental factor. This research provides valuable insights for achieving more accurate and reliable predictions of potential geographical distributions of invasive species.