聖羅倫斯灣北大西洋露脊鯨水下音響定位研究

Abstract

本研究基於 2024 年海洋哺乳動物的偵測、分類、定位和密度估計研討會（The Detection, Classification, Localization and Density Estimation of marine mammals workshop, DCLDE）提供的聖羅倫斯灣北大西洋露脊鯨聲學數據集，分析了兩種聲源定位演算法：到達時間差法（Time Difference of Arrival, TDOA）與到達方向法（Direction of Arrival , DOA）。 在研究方法部分，TDOA 法利用接收訊號的時間差，結合地圖網格統計，計算各網格點的理論到達時間差與實際接收時間差的差異，再透過高斯擴散來推估聲源位置。而 DOA 法則整合了聲納浮標中全向性（omni-channel）與指向性（DIFAR channel）的資料，計算「方位圖（Azigram）」獲得時間與方位角之關係後，透過核密度估計計算方位角之機率。將多組機率結果疊加後，機率最高處即被認定為聲源位置。 為驗證TDOA定位方法的可行性，先進行了聲源模擬，並以此聲源進行定位驗證。模擬聲源之定位結果顯示，最佳定位誤差可達 42.56 公尺，整體定位誤差約為 189 公尺。在確認模擬實驗的有效性後，本研究將 TDOA 與 DOA 方法應用於 DCLDE的數據集進行比較分析。結果顯示，兩種方法均能有效推估北大西洋露脊鯨的可能移動趨勢，並且此方法具備初步辨識群次數量的能力。 本研究證明被動聲學監測技術是一種高效且具成本效益的工具，可用於輔助鯨豚目視調查與長期生態追蹤。

This study is based on an acoustic dataset of North Atlantic right whales in the Gulf of St. Lawrence, provided by 2024 The Detection, Classification, Localization and Density Estimation of marine mammals workshop. Two sound source localization algorithms were analyzed: the Time Difference of Arrival (TDOA) method and the Direction of Arrival (DOA) method. In the TDOA approach, time differences between received signals were compared with theoretical values across a map grid. These differences were used to estimate the sound source location by applying a Gaussian distribution to highlight the most probable area. The DOA approach utilized both omni-channel and DIFAR channel data collected from the sonobuoy. By constructing an “Azigram” to examine the relationship between time and bearing, kernel density estimation was applied to calculate the probability distribution of bearing angles. The probability results from multiple sonobuoys were combined, and the location with the highest overall likelihood was identified as the estimated sound source position. To validate the TDOA method, a simulated sound source was created, and localization accuracy was assessed. The best-case localization error reached 42.56 meters, with an overall average of approximately 189 meters. After confirming the effectiveness of the simulation, both TDOA and DOA methods were applied to the actual DCLDE dataset for comparative analysis. The results showed that both methods could effectively infer the likely movement patterns of North Atlantic right whales and demonstrated potential in providing preliminary insights into group number estimation. This study highlights passive acoustic monitoring as an effective and cost-efficient tool for supporting visual surveys and long-term ecological monitoring of marine mammals.