探討紅外線自動相機於臺灣中大型哺乳動物之偵測效率與影響因子

Abstract

隨著生態監測技術的進步，紅外線自動相機已成為全球野生動物長期監測的重要工具，在臺灣也被廣泛應用於哺乳動物的分布、行為與族群動態調查。為了解相機樣點設置的空間代表性、調查時長與環境因子對偵測效率的影響，本研究以農業部林業及自然保育署2017至2021年於臺灣本島所設置之261台紅外線自動相機為資料來源，系統性探討中大型哺乳動物的偵測機率及其影響因子。 研究首先比較物種分布模型 (Species Distribution Model, SDM) 與專家意見海拔區間兩種空間預測資訊對相機偵測結果的解釋力，結果顯示，無論採用SDM或海拔預測分組，多數保育類及一般類物種於預測範圍內的偵測比例均顯著高於範圍外，尤其保育類物種（如山羊、水鹿、石虎等）在預測範圍內外的差異最為明顯，顯示兩類預測工具皆具高度參考價值。然而，極難偵測物種（如黑熊、野兔等）即使長期部署於預測分布範圍內，其偵測機率仍偏低，反映資料稀少、樣點分布、物種習性等多重因素所造成的限制。 進一步以廣義線性模型 (Generalized Linear Model, GLM) 分析15種中大型哺乳動物之環境因子影響，結果發現海拔與距城市距離為影響最多物種偵測機率的關鍵因子。海拔對物種偵測機率的影響呈現明顯分化，分布於中高海拔的物種（如山羊、水鹿、黃鼠狼等）偵測機率隨海拔升高而提升，偏好低海拔的物種（如穿山甲、白鼻心、麝香貓等）則呈現相反趨勢。距城市距離亦顯著影響部分物種之偵測機率，反映不同物種對人為干擾的敏感度不一。 於調查時間分析方面，累積偵測曲線顯示多數常見或族群密度高之物種於調查初期即可快速被偵測到，達到80%累積偵測機率約需485天，90%則需806天。事件發生時間分析 (Time-to-Event Analysis, TTE) 進一步指出，族群密度高且分布廣的物種（如山羌、獼猴等）在多數樣點中，達到首次偵測機率50%的中位數天數（t₀.₅）僅需數天至數十天，而隱密性高或族群密度低之物種（如黑熊、石虎、野兔等）即使延長調查時間，累積偵測機率亦難以突破50%。 本研究建議，未來自動相機樣點規劃應強化對都市邊緣、農地等多元棲地環境的代表性，並針對稀有或保育類物種提高監測樣點密度與設置時間，提升樣點分布的全面性與監測結果的解釋力，作為優化臺灣野生動物長期監測與保育政策之依據。

With advances in ecological monitoring technology, infrared camera traps have become a key tool for long-term wildlife monitoring worldwide, and have been widely applied in Taiwan for investigating mammalian distribution, behavior, and population dynamics. In order to assess how camera trap site selection, survey duration, and environmental factors influence detection probability, detection probabilities and associated factors of medium-to-large mammals were systematically examined based on data from 261 infrared camera traps deployed by the Forestry and Nature Conservation Agency across Taiwan from 2017 to 2021. Species distribution models (SDMs) and expert-based altitudinal ranges were compared to assess their explanatory power for camera trap detection probability. The findings indicated that, regardless of whether SDMs or expert-defined altitude zones were used, most protected and common species exhibited significantly higher detection probability within predicted ranges than outside, with the differences particularly pronounced among protected species such as Formosan serow, sambar deer, and leopard cat. These results highlight the high reference value of both spatial prediction tools. However, for elusive species with very low detection rates (e.g., Formosan black bear, hare), detection probability remained low even when traps were deployed long-term within predicted ranges, reflecting constraints related to data scarcity, site coverage, and species ecology. Further analysis using generalized linear models (GLM) indicated that elevation and distance to the nearest city were the most influential environmental factors affecting detection probability for the 15 focal species. The effect of elevation showed strong differentiation among species: detection probability increased with elevation for high-mountain species (e.g., serow, sambar, weasel), but decreased for species preferring lower elevations (e.g., pangolin, masked palm civet, small Indian civet). Distance to city was also significant for some species, reflecting varied sensitivity to human disturbance. Regarding survey duration, cumulative detection probability curves revealed that most common or high-density species could be rapidly detected early in the survey, with 80% cumulative detection probability reached after approximately 485 days and 90% after 806 days. Time-to-event (TTE) analysis further indicated that high-density, widely distributed species (such as Reeves’s muntjac and Formosan macaque) required only several days to several tens of days for the median time to first detection, while protected or elusive species (e.g., black bear, leopard cat, hare) rarely surpassed a 50% cumulative detection probability even with extended survey periods. Based on these findings, it is recommended that future camera trap network designs in Taiwan should strengthen the representativeness of urban-edge and agricultural habitats, and moderately increase both the density and duration of monitoring for rare or protected species. Such improvements will enhance the spatial comprehensiveness of sampling and the explanatory power of monitoring results, providing a more robust scientific foundation for optimizing long-term wildlife monitoring and conservation policy in Taiwan.