深度文章系列 06 • 資料來源 • 評分演算法 • CSIS 方法論
Article Series 06 • Data Sources • Scoring Algorithm • CSIS Methodology
本站使用兩個主要資料來源，互相補充、交叉驗證。
src/fetch_ais_data.py 每 2 小時自動從台灣港務局 API 取得台灣海峽及周邊海域的即時 AIS 資料，監測範圍約 20–28°N、112–128°E。資料包含所有廣播 AIS 訊號的船隻位置、速度、航向、目的地等。資料透過 SOCKS5 代理取得以確保穩定性。
src/fetch_gfw_data.py 定期從 Global Fishing Watch（全球漁業觀察）API 取得合成孔徑雷達（SAR）衛星影像分析結果。SAR 可以在全天候條件下偵測水面上的金屬物體，並標記未廣播 AIS 的「暗船」（Dark Vessel）。兩個資料來源的交叉比對是發現 AIS 走暗行為的核心方法。
This site uses two primary data sources that complement and cross-validate each other.
src/fetch_ais_data.py, real-time AIS data is automatically retrieved from the Taiwan Port Bureau API every 2 hours, covering the Taiwan Strait and surrounding waters (~20–28°N, 112–128°E). Data includes position, speed, heading, destination, and vessel identity for all vessels broadcasting AIS. Retrieved via SOCKS5 proxy for reliability.
src/fetch_gfw_data.py, SAR satellite imagery analysis results are regularly retrieved from the Global Fishing Watch API. SAR detects metallic objects on the water surface in all weather conditions and flags "dark vessels" not broadcasting AIS. Cross-referencing both data sources is the core method for detecting AIS dark behavior.
本站透過 GitHub Actions 自動化整個資料處理流程，確保資料持續更新且過程可稽核。
The site automates the entire data processing workflow through GitHub Actions, ensuring continuous updates and an auditable process.
監測範圍內有大量的浮標、漁網標誌器等設備也會發出 AIS 訊號，但它們顯然不是我們關注的對象。在昂貴的分析工作開始之前，這些訊號會被優先排除：
9 開頭 → 助航設施（AtoN）/ 潛水浮標898 開頭 → 漁網標誌器% → 漁網浮標BUOY → 浮標12.5V → 漁網浮標The monitoring area contains many buoys, fishing net markers, and other equipment that also broadcast AIS signals, but they're obviously not vessels of interest. These are excluded before expensive analysis begins:
9 → Aids to Navigation (AtoN) / dive buoys898 → Fishing net markers% → Fishing net beaconsBUOY → Buoys12.5V → Fishing net beacons通過排除規則的船隻，將被套用以下 8 項指標進行評分。評分基礎來自 CSIS（戰略與國際研究中心）的「Signals in the Swarm」報告所提出的分析框架。
Vessels that pass the exclusion rules are evaluated against the following 8 criteria. The scoring framework is derived from CSIS (Center for Strategic and International Studies) "Signals in the Swarm" report.
| # | 指標 | Criterion | 偵測方法 | Detection Method | 分數 | Score |
|---|---|---|---|---|---|---|
| 1a | 電纜鄰近 | Cable Proximity | 航跡點距海底電纜 5 公里以內 | Track points within 5km of submarine cable | +2 | |
| 1b | 電纜低速滯留 | Cable Loitering | 在電纜附近低速（<8節）滯留超過 3 小時 | Low speed (<8kn) near cable for >3 hours | +3 | |
| 2 | 之字形航行 | Zigzag Pattern | ≥3 次 45° 以上的航向轉變 | ≥3 turns of ≥45° heading change | +1 | |
| 3 | 200 米等深線 | 200m Depth Contour | ≥30% 航跡時間在大陸棚邊緣附近 | ≥30% of track time near continental shelf edge | +1 | |
| 4 | AIS 異常 | AIS Anomalies | 船名更換≥2次、走暗>18小時、類型更換、身分事件 | Name changes ≥2, dark gaps >18hr, type changes, identity events | +1 / +3 | |
| 5 | 非前十旗幟國 | Non-Top-10 Flag | MMSI 前三碼非前十大旗幟國 | MMSI MID not in top-10 flag state set | +1 | |
| 6 | 聯合國制裁 | UN Sanctions | IMO 或船名與制裁名單匹配 | IMO or name match against sanctions list | +8 | |
| 7 | AIS 欺騙 | AIS Spoofing | 不可能的物理運動 / 方框模式 / 圓形模式 | Impossible physics / box pattern / circle pattern | +4 each | |
| 8 | ITU MARS 不符 | ITU MARS Mismatch | 船名、IMO 或呼號與 ITU 船站登記不符 | Ship name, IMO, or call sign differs from ITU registry | +3 | |
| 9 | STS 旁靠 | STS Transfer | 涉及船對船旁靠事件 | Involved in ship-to-ship rendezvous | +2 / +5 |
不同類型的船隻，相同行為的威脅程度不同。行為性指標（指標 1–3, 5）會乘以下列係數；高威脅指標（指標 4, 6–9）不乘以係數。
The same behavior carries different threat levels for different vessel types. Behavioral criteria (1–3, 5) are multiplied by the following coefficients; high-threat indicators (4, 6–9) are not multiplied.
| 船舶類型 | Vessel Type | 係數 | Multiplier | 理由 | Rationale |
|---|---|---|---|---|---|
| 貨船、油輪、LNG | Cargo, Tanker, LNG | ×1.0 | 重型船錨、高噸位 → 實質電纜損壞風險最高 | Heavy anchors, high tonnage → highest real cable damage risk | |
| 中國海警、海巡、海救、科研 | Chinese Gov't vessels (CG, MSA, Rescue, Research) | ×0.5 | 中國公務/特殊利益船隻，行為模式不同 | China public-service / special-interest state vessels | |
| 漁船 | Fishing | ×0.2 | 體積小、常規作業 → 電纜威脅低 | Small, routine operations → low cable threat | |
| 其他 / 未知 | Other / Unknown | ×0.5 | 不確定，保守估計 | Uncertain, conservative estimate |
最終分數計算公式：round(行為分數 × 船型係數) + 高威脅指標分數
Final score formula: round(behavioral_score × type_multiplier) + high_threat_indicators
本站的評分框架參考並延伸自美國智庫 戰略與國際研究中心（CSIS）發表的「Signals in the Swarm」研究報告。該報告系統性地分析了中國漁船群在台灣周邊海域的威脅行為模式，識別出電纜鄰近、走暗、之字形航行等核心指標。
本站在 CSIS 框架的基礎上，增加了 AIS 欺騙偵測、ITU MARS 登記查核、STS 旁靠偵測等指標，並整合 GFW SAR 衛星資料，以強化針對台灣海域的偵測能力。
This site's scoring framework references and extends the "Signals in the Swarm" research report published by the Center for Strategic and International Studies (CSIS). The report systematically analyzes behavioral patterns of Chinese fishing fleets in Taiwan's surrounding waters, identifying core indicators including cable proximity, AIS dark events, and zigzag navigation.
Building on the CSIS framework, this site adds AIS spoofing detection, ITU MARS registry verification, STS rendezvous detection, and integrates GFW SAR satellite data to enhance detection capabilities specific to Taiwan's waters.
透明度要求我們誠實地說明這套方法的局限性：
Transparency requires us to honestly state the limitations of this approach:
除了「任一航跡點是否在海纜 5 公里內」這個較粗的判斷，本站還會對船舶最近一筆位置做更精細的地理研判，把「位置」本身納入評分。
系統以內政部公告的領海基線計算該位置屬於哪個法域（內水／領海 12 浬／鄰接區 24 浬／經濟海域／公海），並計算其與最近海底電纜的距離緩衝帶（≤1km／≤5km／≤10km）。在此基礎上，加入兩項有上限的行為加分：
這兩項屬於「行為分」，一樣會乘上船型加權（商船權重高、漁船權重低）。它的用意是精修而非取代既有的 5 公里粗網——把「此刻就壓在電纜上、又落在我國管轄海域」這個最符合灰色地帶電纜威脅的情境，給予額外但有限的權重。EEZ 在此採「距基線 ≤200 浬」的簡化定義，僅供風險研判，並非法律上的劃界。
Beyond the coarser test of "is any track point within 5 km of a cable," the site also makes a finer geographic read of a vessel's most recent position, folding location itself into the score.
Using Taiwan's official territorial baseline, the system classifies which legal zone that position falls in (internal waters / territorial sea 12 nm / contiguous zone 24 nm / EEZ / high seas) and computes its distance band to the nearest submarine cable (≤1km / ≤5km / ≤10km). On top of that, two bounded behavioral bonuses are added:
Both are "behavioral" points and are multiplied by the same vessel-type weighting (cargo high, fishing low). The intent is to refine, not replace, the existing 5 km net — giving extra but limited weight to the scenario that best matches a gray-zone cable threat: "sitting on a cable right now, inside our jurisdiction." EEZ here uses a simplified "≤200 nm from baseline" definition for risk analysis only, not a legal delimitation.
本站刻意建立在公開可得的資料與公開的程式碼之上，這是一個方法論選擇，而非權宜之計。
OSINT 的價值不在於取代官方情報，而在於提供一個任何人都能查核的共同事實基礎。當討論台灣周邊的灰色地帶活動時，一個公開、可重現、誠實標示限制的工具，比一個「黑箱結論」更值得信任。
This site is deliberately built on publicly available data and public code — a methodological choice, not a convenience.
The value of OSINT is not to replace official intelligence but to provide a shared factual basis anyone can audit. When discussing gray-zone activity around Taiwan, an open, reproducible tool that states its limits honestly deserves more trust than a black-box conclusion.
The concepts on this page can be verified against these public, authoritative sources:
This page is for research and education; detection signals are leads to investigate, not legal or factual findings. See the methodology.