比較系統性與非系統性公民科學資料於鳥類物種豐富度預測之表現差異

Comparing the Effectiveness of Species Richness Estimation Models by Using Structured and Unstructured Citizen Science Data in Taiwan

沈芳伃(Fang-Yu Shen)

指導教授 ： 丁宗蘇

共同指導教授 ： 蔡若詩(Jo-Szu Tsai)

國立臺灣大學/生物資源暨農學院/森林環境暨資源學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202002665

摘要

物種豐富度常做為物種多樣性評估指標。近年由於公民科學興起，可望成為 蒐集生物多樣性資料的一項方法。公民科學主要分為兩類:系統性公民科學與非系 統性公民科學。系統性公民科學比非系統性公民科學更具有標準化的調查方法，但 志工培訓與參與度維持的成本也較高，資料缺失發生頻率相對較高。非系統性公民 科學沒有一致的標準調查方法，且志工參與條件較低，大量的觀測資料有機會彌補 系統性公民科學的資料缺失。基於非系統性公民科學在調查上的彈性，物種偵測率 與努力量的變異(例如:調查持續時間)都很容造成資料偏差。僅管預測物種豐富 度可以減少不完美物種偵測率所造成的偏差，但在非系統性公民科學中，不同物種 豐富度預測方法的表現仍不清楚。另外，在非系統性公民科學，較缺乏探討時間調 查努力量與物種豐富度之間的非線性關係。本研究使用誤差值(bias)，以台灣繁殖 鳥類大調查(BBS)樣區之原始物種豐富度為比較基準，計算與該樣區鄰近範圍 eBird 紀錄清單在標準化時間調查努力量下評估物種豐富度預測表現。我選擇包含在每 個獨立的 2×2 km BBS 樣區內所有 eBird 紀錄清單，並計算三種物種豐富度預測方 法中誤差值最小的預測方法。為探討物種豐富度經預測後在標準化時間調查努力 量上的表現，我於四個非線性方程式中探討時間調查努力量與物種豐富度表現最 好的方程式。本研究發現，Chao1 物種豐富度預測方法有最低的誤差值。而冪函數 方程式為解釋時間調查努力量與物種豐富度關係的最佳非線性方程式。在 60 分鐘 基準之冪函數方程式上，從原始物種豐富度經過 Chao1 物種豐富度預測後，誤差 值更接近於零(從-0.34 至-0.14)。代表 eBird 物種豐富度經預測後相對於 BBS 紀錄 物種數從 66%提升至 86%。結果指出，單獨使用原始物種豐富度來做物種豐富度 指標時，不完美偵測率可能導致資料誤差。經過物種豐富度預測後會增加物種豐富 度指標的準確度。在非系統性公民科學中，調查方法與物種偵測率影響偵測物種數 量。另外，低時間調查努力量容易產生較高比例的單隻種(singleton)，影響物種豐 富度預測的準確性，可能限制非系統性公民科學資料的使用性。本研究建議，非系 統性公民科學的物種豐富度需經過預測才能降低不完美偵測率所造成的資料偏差。 另外，使用 Chao1 物種豐富度方法執行預測時，需評估樣本的單隻種比例所產生 之預測誤差。

關鍵字

群聚多樣性 ； 物種偵測率 ； Chao 物種豐富度預測 ； 調查努力量 ； 調查誤差 ； 監測

並列摘要

Ecologists have long recognized species richness as an essential indicator of biodiversity and ecosystem functioning. More recently, citizen science has emerged as a means for collecting species richness data. There are two main categories of citizen science: structured and unstructured citizen science. These two categories employ different investigations methods, as structured citizen science tends to be more rigorous, but requires volunteers with more training and determination, resulting in high frequency of missing observations. In contrast, unstructured citizen science is less formal and easier to participate, and may be considered to make up for missing observations. However, unstructured citizen science tends to suffer from biases due to imperfect species detection probability and variable effort (e.g., survey duration). Although species richness estimation methods have been applied to many datasets in order to account for imperfect detection probability, the ability of these estimators to control for biases and the non-linear relationship between duration and species richness in unstructured citizen science data remain unclear. This study was aimed to investigate the effectiveness of species richness estimation applied to eBird dataset by comparing it to observed species richness of Breeding Bird Survey Taiwan (BBS) sites at a standardized duration. For this comparison, I selected eBird checklists that fell within a 2×2 km square buffer placed around BBS sites across Taiwan. Bias was used to evaluating the effectiveness of species richness estimates from the eBird dataset. I presented three species richness estimation methods based upon the eBird dataset that have been commonly reported in the ecological literature. To measure the reduction value of bias with before and after species richness estimation at a standardized duration, four non-linear functions were first used to examine the relationship between duration and species richness. The result showed that the Chao1 estimator was the least biased estimation method. The power function was the best selected parsimonious of non-linear function to explain the relationship between duration and species richness. Based on the power function, the eBird dataset can produce species richness estimates comparable to those generated using the BBS dataset raised from 66% to 86% after applying the Chao1 estimator on the eBird dataset. These results suggested that measuring species richness by raw species count alone would be biased, and species richness estimation takes imperfect detection probability into account, which improved the accuracy of measuring species richness. Survey protocols and species detection probability significantly influenced the species detected in unstructured citizen science data. Problems with biased results derived from high occurrence of singleton species, especially in low-effort surveys, limit the quality and potential uses of unstructured citizen science data. Overall, to accurately present species richness in a given area, I suggest species richness should be estimated, and the effect of number of singletons should be evaluated before applying Chao1 estimation from unstructured citizen science data.

並列關鍵字

community richness ； species detection probability ； Chao estimator ； sampling effort ； sampling bias ； monitoring

參考文獻

Arrhenius, O. (1921). Species and area. Journal of Ecology, 9(1), 95-99.

Bean, W. T., Stafford, R., Brashares, J. S. (2012). The effects of small sample size and sample bias on threshold selection and accuracy assessment of species distribution models. Ecography, 35(3), 250-258.

Bird, T. J., Bates, A. E., Lefcheck, J. S., Hill, N. A., Thomson, R. J., Edgar, G. J., Stuart-Smith, R. D., Wotherspoon, S., Krkosek, M., Stuart-Smith, J. F. (2014). Statistical solutions for error and bias in global citizen science datasets. Biological Conservation, 173, 144-154.

Boakes, E. H., McGowan, P. J., Fuller, R. A., Chang-qing, D., Clark, N. E., O'Connor, K., Mace, G. M. (2010). Distorted views of biodiversity: spatial and temporal bias in species occurrence data. PLoS Biology, 8(6), e1000385.

Bonter, D. N., Cooper, C. B. (2012). Data validation in citizen science: a case study from Project FeederWatch. Frontiers in Ecology and the Environment, 10(6), 305-307.