2025-11-10 サセックス大学
サセックス大学の研究チームは、AIを用いてインドネシアにおけるコウモリの狩猟・取引・消費の分布を特定する手法を開発し、学術誌『People and Nature』で発表した。研究では「ヒト―自然インターフェース・マッピング」と呼ばれる新たなモデルを構築し、GoogleやSNS上の英語・インドネシア語投稿をAIが自動検索して、コウモリ利用に関する記録を収集。人口密度や保護区への近接性などの要因と照合し、データの乏しい地域でも活動の多発地を予測した。結果、ジャワ島、スラウェシ島、スマトラ島が主要なホットスポットと判明した。狩猟は取引より広域に発生し、動物が遠距離輸送されている可能性も示唆された。多言語検索の導入で精度が向上し、低コストでスケーラブルな保全手段として有効性が確認された。研究者は、オンラインデータを活用したAI分析が、現地調査の難しい地域で野生動物保全を支援する新たな道を拓くと強調している。
＜関連情報＞
- https://www.sussex.ac.uk/research/full-news-list?id=69452
- https://besjournals.onlinelibrary.wiley.com/doi/10.1002/pan3.70139
オンラインプラットフォームの記録を用いたインドネシアにおけるコウモリの狩猟、取引、消費のマッピング Mapping bat hunting, trade and consumption in Indonesia using records from online platforms
Sara Bronwen Hunter, Julie Weeds, Fiona Mathews
People and Nature Published: 16 September 2025
DOI:https://doi.org/10.1002/pan3.70139
Abstract
- Hunting, trade and consumption (exploitation) of bats for food, medicine or other uses is widespread and threatens many species worldwide. However, collecting exploitation data in the field is logistically challenging and resource-intensive, resulting in gaps in our knowledge of the extent and intensity of exploitation in many areas. Online platforms, such as social media, provide a new data source from which to better understand spatial patterns of human–wildlife interactions. Nonetheless, online records can be biased and often vary in spatial and taxonomic resolution.
- This study aimed to investigate the effectiveness of using human–nature interface mapping, whereby statistical approaches from species distribution modelling are used in building spatially explicit threat maps with online data. We predicted the probability of bat exploitation occurrence across Indonesia, using 475 records obtained from automated searches in English and Indonesian.
- Overall, MaxEnt models showed high performance, with an average AUC of 0.89. The use of bias layers to select background data did not consistently improve model performance when this was assessed using cross-validation, but it did slightly improve performance when this was assessed using a held-out dataset of threat occurrence points from academic literature. Predictions from models of trade and consumption occurrence had relatively low similarity (Pearson’s correlation = 0.586) with predictions from models of hunting occurrence, and hunting was predicted to occur over a more extensive area.
- This study demonstrates the utility of using presence-only modelling to produce spatially explicit predictions of human–nature interactions. These models can be combined with locally relevant information and ground-truthing to inform the threat status of bat populations in Indonesia and prioritise conservation interventions.
