比較系統性調查與公民科學資料於鳥類物種分布模式之表現差異

Abstract

物種分布模式為現今保育研究中的重要工具。近年來公民科學資料蓬勃發展，集結公眾之力蒐集物種分布資訊，相較傳統調查能耗費較低成本蒐集更大量的資料，但許多研究者對公民科學的資料品質及其可應用性存有疑慮。本文比較系統性調查資料以及公民科學資料應用於最大熵物種分布模式之差異，並評估eBird公民科學資料未來應用在物種分布模式之可行性，以提供未來公民科學資料用於分布預測之建議。系統性調查資料所提供的資料量較少，但是空間解析度較高且資料品質較為一致；eBird資料量大，但是紀錄資料的努力量以及不同紀錄點位的空間解析度差異很大，易使模式表現產生偏差。為比較資料量或空間解析度造成不同資料集間的模式表現差異，本研究除了比較系統性調查資料與eBird整體資料的模式表現，另外也篩選一筆空間解析度較高的eBird資料，比較提高紀錄點位空間解析度的模式表現差異；以及在與系統性調查資料相同樣本數的基準下，將eBird資料重新取樣後一併比較。本文以金門為研究範圍，使用2018年上述四組不同資料集，建構66種鳥種的最大熵模型，並比較不同資料集模式的AUC、預測出現機率值與預測出現網格數的差異。除此之外，也蒐集一筆獨立調查資料，以模式預測結果與實際野外調查物種是否有出現，計算Kappa值、精確率與正確率以比較不同資料集的模式預測表現。另外也將鳥種依據水鳥/陸鳥、普遍/不普遍、留/候鳥分組，比較不同鳥種分組之預測表現差異。結果顯示系統性調查資料在AUC、精確率與正確率的表現較eBird資料佳，而eBird資料在預測出現機率值與預測出現網格數、Kappa值較系統性調查資料高；鳥類分組的比較，在棲地專一性高的鳥種分組AUC表現較佳，預測出現範圍較小。eBird能提供大量資料及其紀錄地點廣布在研究區域內，是eBird資料表現較佳的優勢，而系統性調查資料因其調查樣區空間解析度高使得其預測的表現較佳。本研究結果顯示eBird資料應用於物種分布模式有不錯的預測表現，將不精確的空間紀錄篩選過後的eBird資料，會使模式預測表現更佳。因此，本文建議未來使用eBird資料運用於物種分布模式時，在資料前處理時能先剔除不夠精確的地點以及不確定性高的資料，能讓模式有較佳的表現。

Species distribution model (SDM) has become an important tool in conservation biology. In recent decades, the amount of citizen science data has accumulated rapidly. Comparing to systematic data collected from traditional surveys, citizen science data are contributed by many volunteers and could collect huge data in an efficient and budget way. However, systematic data tend to have higher data quality and finer spatial resolution than citizen science data. This study compared the performance of Maximum Entropy models (MaxEnt) between the systematic data and eBird data in Kinmen collected in 2018. The distribution of 66 bird species was predicted with MaxEnt model by using 4 types of datasets, including 1 systematic survey dataset and 3 eBird datasets. eBird culled data, which selected complete eBird checklists with finer spatial resolution obtained with shorter survy distances to test whether the spatial resolution of input data would impact model performance. In order to balance the sample size between eBird data and systematic data, eBird data were randomly sampled with the same data size of systematic data. In addition, this study collected independent field survey data to validate the model performance and compare model performance with three bird groups. The results showed that the systematic data performed better than eBird data in the AUC, precision, and accuracy of predicted distribution, while the eBird data showed better performance in predicted presence probability mean, predicted presence grids, and Kappa statistics due to its larger datasets and greater investigated area. The model performance of habitat specialists was better than that of habitat generalists and the predicted presence range of habitat generalists was larger than that of habitat specialists. The model performance of eBird data was good enough and can be even better when adapting the data with finer spatial resolution. The study suggests that one should remove inaccurate or spatial imprecise data before applying eBird data in species distribution modelling in the future.