Scholarlink

<p>This study aimed to use machine-learning-based ensemble species distribution models (SDMs) under projected climate change scenarios to predict future habitat changes for cold-water freshwater fish (CWFF) in South Korea. Five representative CWFF species, namely masu salmon (<i>Oncorhynchus masou</i>), lenok salmon (<i>Brachymystax lenok tsinlingensis</i>), spotted barbel (<i>Hemibarbus mylodon</i>), splendid dace (<i>Coreoleuciscus splendidus</i>), and torrent catfish (<i>Liobagrus mediadiposalis</i>), were analyzed using presence–absence data from the National Institute of Environmental Research as well as high-resolution data on water quality, topography, and climate. Future climate variables were derived from four IPCC Shared Socioeconomic Pathway (SSP) scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) and examined across three 30-year periods: the near (2011–2040), medium-term (2041 2070), and far (2071 2100) future. In addition, eight individual models, including random forest, artificial neural network, gradient boosting machine, and MaxEnt models, were tested, and the seven highest performing models were combined in an ensemble that achieved robust predictive performance with the area under the receiver operating characteristic curve being 0.924, and the true skill statistic having a value of 0.771. The results also revealed that the mean annual temperature, dissolved oxygen content, pH, total phosphorus, and turbidity had strong effects on the determination of CWFF habitat suitability. Across all scenarios, the CWFF habitats progressively contracted, with losses particularly severe under the high emission SSP58.5 scenario. The minimal emergence of new suitable habitats highlighted the risk of increasing fragmentation and habitat loss. By integrating multiple environmental factors through ensemble machine learning SDMs, the present study provides robust scientific evidence supporting spatial prioritization and policy development for CWFF conservation and climate adaptation.</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

Predicting habitat changes for cold-water freshwater fish under climate change scenarios using a machine-learning-based ensemble species distribution model

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Wheemoon Kim,
  • Kyeongtae Kim,
  • Wonkyong Song

摘要

This study aimed to use machine-learning-based ensemble species distribution models (SDMs) under projected climate change scenarios to predict future habitat changes for cold-water freshwater fish (CWFF) in South Korea. Five representative CWFF species, namely masu salmon (Oncorhynchus masou), lenok salmon (Brachymystax lenok tsinlingensis), spotted barbel (Hemibarbus mylodon), splendid dace (Coreoleuciscus splendidus), and torrent catfish (Liobagrus mediadiposalis), were analyzed using presence–absence data from the National Institute of Environmental Research as well as high-resolution data on water quality, topography, and climate. Future climate variables were derived from four IPCC Shared Socioeconomic Pathway (SSP) scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) and examined across three 30-year periods: the near (2011–2040), medium-term (2041 2070), and far (2071 2100) future. In addition, eight individual models, including random forest, artificial neural network, gradient boosting machine, and MaxEnt models, were tested, and the seven highest performing models were combined in an ensemble that achieved robust predictive performance with the area under the receiver operating characteristic curve being 0.924, and the true skill statistic having a value of 0.771. The results also revealed that the mean annual temperature, dissolved oxygen content, pH, total phosphorus, and turbidity had strong effects on the determination of CWFF habitat suitability. Across all scenarios, the CWFF habitats progressively contracted, with losses particularly severe under the high emission SSP58.5 scenario. The minimal emergence of new suitable habitats highlighted the risk of increasing fragmentation and habitat loss. By integrating multiple environmental factors through ensemble machine learning SDMs, the present study provides robust scientific evidence supporting spatial prioritization and policy development for CWFF conservation and climate adaptation.