應用被動式聲學監測台灣西海岸中華白海豚行為生態與棲地利用

Abstract

台灣西海岸極度瀕危的中華白海豚，族群量不足一百隻，其活動概況與棲地利用亟需研究以作為保育政策擬定的參考。本研究為首篇利用被動式水下聲學儀器來追蹤白海豚生態活動的論文，共採用兩種資料搜集法：2008年於海上船行調查時，收錄了79群白海豚的水下錄音資料(25.9小時)，據以分析其哨叫聲(whistle)的曲調與曲譜的組成。2009年7月至2012年9月間，復於新虎尾溪口放置水下聲學資料紀錄器，收集白海豚的高頻答聲(click)。論文內容分兩部分：(1)哨叫聲與答聲自動化偵測演算法的研發與驗證，(2)應用被動式聲學監測法來瞭解白海豚的行為生態與棲地利用。 在自動化哨叫聲偵測器的研發部分，首先增加局部極大值偵測器(local-max detector)中鯨豚哨叫聲檢測的即時頻寬(instantaneous frequency bandwidth)，透過此一步驟，即使在吵雜的環境中，哨叫聲的代表頻率(representative frequency)仍可被有效的萃取出來，偵測效能達到75-85%的正確率，僅有5%的誤報率。哨叫聲的代表頻率在萃取出來後，可用以量化諧音(harmonic)、重疊哨叫聲的出現頻度，以及哨叫聲音頻使用的複雜度。此外，透過聲學資料紀錄器和與之搭配的資料篩選器，進行白海豚高頻率答聲的偵測。篩選器中包含七項參數：聲壓偵測閾值、最小的答聲間距、答聲間距變化比例、最小與最大答聲數、聲壓值與聲源角度變異係數。經過驗證後，偵測效能可達60%的正確率，只有4%的誤報率。 在中華白海豚聲音行為與生態研究方面，本論文有以下三項結果：(1) 根據曲調分析和非監督式的自動化分類方法，共辨識出17種不同的哨叫聲類型。由各類哨叫聲的出現機率可知，白海豚最常發出的哨叫聲類型為平緩曲調，頻率範圍介於5-7 kHz。分析其哨叫聲與行為狀態的關聯，發現白海豚社交時的哨叫聲，在發聲頻度及哨叫聲曲譜的多樣性兩方面皆為最高。(2) 在白海豚棲地利用的時空變異方面，本研究發現白海豚的每日活動與日夜週期無明顯關聯，卻與潮汐週期顯著相關，且潮汐影響僅顯現在靠河口一側的海域，其中，退潮時段白海豚的接觸機率(encounter rate)顯著低於其他三個潮汐階段。此外，推測為捕食表層性魚群而呈現繞圈行為的群體(circling schools)，其出現機率與空間移動的變化模式也與潮汐時段的改變顯著相關。(3) 在白海豚季節分佈模式的變異方面，白海豚出現的頻度與環境中超音波脈衝聲數量的多寡(可以視為槍蝦活動的指標)皆在春夏等雨季與大雨過後的時段，呈現出從河口向外海偏移其主要分布梯度的趨勢。因此雨季中大量的河川逕流，可能改變了河口無脊椎動物和魚類的分布，進而影響白海豚的季節性分布梯度。 總結來說，被動式聲學監測可提供一個有效的平台來了解白海豚活動的時空變異。白海豚的活動主要受到牠們食餌獵物的活動模式影響，而這些獵物的活動又與棲地環境因子的變動高度相關。本論文所得白海豚活動時空變異的資料可做為研擬保育措施時的科學證據，據以落實減輕白海豚受人為活動干擾的決策。

Critically endangered, the Taiwanese population of the humpback dolphin (Sousa chinensis) consists of only less than 100 animals. In an attempt to provide essential baseline information for the conservation planning and policy-making regarding these animals, the present study investigated their activity patterns and habitat utilization. Two methods were adopted to collect data in two phases. In 2008, broadband recordings of 79 schools of sighted humpback dolphins were collected by a hydrophone during onboard surveys. Whistle contours and the composition of the call repertoires were subsequently analyzed to examine the acoustic behaviors of the humpback dolphin. Between July 2009 and September 2012, acoustic data loggers were deployed in XinHuwei River Estuary to record the ultrasonic clicks of the humpback dolphin. The findings are hereby presented in two parts centering respectively on (1) the development and validation of two automatic detection algorithms for extracting dolphin tonal sounds and biosonar sounds and (2) the use of passive acoustic monitoring in the study of the behaviors and habitat use of the humpback dolphin. To isolate dolphin tonal sounds from the broadband recordings, an algorithm was developed by integrating the procedure of detecting instantaneous frequency bandwidths of the tonal sounds in the local-max detector. The algorithm allowed effective detection of the representative frequencies of the dolphin whistles even in noisy conditions with a 75-85% detection rate, of which 5% are false alarms. The representative frequencies were then used to quantify the prominence of harmonics and overlapping calls, and the modulation of tonal sounds. On the other hand, dolphin biosonar sounds were recorded by acoustic data loggers and then detected by an off-line filter using 7 parameters: the pressure level detection threshold, minimum inter-click interval, ratio of inter-click interval change, minimum and maximum number of clicks, coefficient variance of pressure level and time difference. Based on manual validation, the off-line filter accurately extracted biosonar sounds 60% of the time with only a 4% false alarm rate. The acoustic behaviors and ecology of the humpback dolphin can be described in three ways as follows. (1) 17 whistle types were identified based on the results of a contour analysis and an unsupervised classification. In addition, the whistles produced by the humpback dolphin were predominantly of gentle frequency modulation between 5-7 kHz. Insofar as the connection between the whistle usage and behavioral states is concerned, the results show that the whistling rate and repertoire diversity of the socializing state were the highest among the four defined behavioral states (traveling, foraging, socializing, and milling states). (2)With regard to the temporal and spatial variations in the habitat use of the humpback dolphin, no significant diurnal variation was found in dolphin activities. Activity patterns of the humpback dolphins on the river side of the estuary were instead correlated with the tidal cycle. The encounter rate, moreover, during the ebb tide was significantly lower than that of the other tidal phases. The occurrence probability and spatial position of the circling schools, associated with the hunt for epipelagic fish, also changed with the tidal cycle. (3) Both seasonal variations in the humpback dolphin occurrence and in the ambient ultrasonic pulse levels (an index of snapping shrimp activity) showed an outward shift during rainy seasons, as well as the time periods after heavy rain. The large volume of river runoff during the rainy season could alter the distribution of invertebrates and fish near the river mouth, thus affecting the seasonal distribution gradient of the humpback dolphin. To sum up, passive acoustic monitoring could provide an effective platform for investigating the spatial and temporal variations in the activities of the humpback dolphin. The behavioral changes of the dolphins may be related to the activities of their prey, which may in turn be influenced by the dynamics of other environmental factors. The findings of the present study can be used as the basis for viable conservation efforts and policy-making regarding the endangered Taiwanese population of the humpback dolphin.