Spatiotemporal volunteer data in development of spatiotemporal roadkill model and analysis of roadkill hotspot zoning

The 36th Conference on Surveying and Geomatics, 第36屆測量及空間資訊研討會
Yu-Pin Lin1, Wei-Chih Lin2, Johnathen Anthony3, Wan-Yu Lien4, Te-En Lin5, Guo-Hao Huang4
林裕彬1 林韋志2 安子傑3 連宛渝4 林德恩5 黃國豪4
1.國立台灣大學生物環境系統工程學系 特聘教授兼主任
2.環輿科技股份有限公司資深系統工程師
3.國立台灣大學生物環境系統工程學系 博士生
4.國立台灣大學生物環境系統工程學系 博士後研究員
5.特有生物研究保育中心 助理研究員

發表日期: 
2017/8/30

Due to increasing urbanization and regional development, wildlife habitats have been experiencing changes and even losses, which further lead to the death of slow moving animals (e.g. herpetofauna) when they attempt to cross the road in search for food and water. Situations like the abovementioned, where animals are hit and killed by cars or other vehicles are called roadkill or road mortality. Based on the professionally-collected database (TaiBIF), we employed the species distribution Models (SDMs), including logistic regression, Poisson regression, general additive model, support vector machine, and maximum entropy approaches to investigate the uncertainties originating from the structures and the parameters of SDMs. In addition, we also used bootstraping approach to generate the random distribution of 100 datasets for training the five SDMs and evaluating both the data uncertainty and model uncertainty. That is, for each species, we generated 500 possible spatial distribution of its habitat suitability. Then, we analyzed the seasonality pattern of the roadkill data and developed a spatio-temporal roadkill probability distribution model. Based on the SDMs derived from five models, we created the local spatial uncertainty map for each species. We combined the local spatial uncertainty maps and the roadkill data with seasonality pattern and environmental variables of the road to estimate the possible spatial-temporal distributions of roadkill data, and then create the possible roadkill probability distribution in different seasons. The results of the estimated spatial-temporal roadkill model show that there is a significant difference in the probability map of roadkill between seasons. Therefore, when estimating geographical roadkill distributions, it is necessary to factor in the time and seasonal differences in roadkill data in addition to species distributions.

Keyword: species distribution model, roadkill, spatio-temporal roadkill probability distribution model, uncertainty, Volunteered Geographic Information (VGI)

由於都市及區域開發的情況增加,使得移動速度緩慢的動物 (如兩棲爬蟲類動物) 於尋找食物或水源時,常因穿越馬路導致死亡。本研究以專家調查資料 (TaiBIF) 為基礎,首先利用僅具空間特性的物種分布模式,包括羅吉斯迴歸、卜瓦松迴歸、支持向量機、廣義加性模型與最大熵法,探討環境空間變異對物種分布的影響,並利用拔靴法隨機產生100組樣本,來評估物種分布以及物種觀測資料的局部及空間不確定性 (每個物種共產生500組棲地適合度空間分布)。然後針對路殺資料時間趨勢進行分析,探討路殺資料是否具有時間特性,並建立具有時間-空間特性之路殺機率分布模式。利用整合五種物種分布模式結果而得之局部及空間不確定性分布,結合路殺的季節資料與環境變數資料,來推估13個物種潛在路殺分佈,並繪製不同季節的物種路殺分布圖。路殺時間-空間特性模式的推估結果顯示物種於不同季節的路殺比率有顯著差異。因此推估路殺發生的點位時,除了考慮物種棲地適合度的空間分布外,也需要考量路殺的時間特性。

關鍵詞:物種分布模式、路殺、具有時間-空間特性之路殺機率分布模式、不確定性、自願性地理資訊

回應本文