YuCROWN × EarthLinC · 獨立研究站Independent research station

真實觀測與資料抽樣 · 馬瑙斯 2018–2019Real observations & data resampling · Manaus 2018–2019

城市變吵時，
牠怎麼溝通？When the city gets loud, how do they talk? Signals in a Noisy City

黑白花狨在城市森林裡用長叫聲呼喚，也在枝幹上留下氣味。這裡把研究者在馬瑙斯記錄的每段 5 分鐘行為片段，和當時量到的噪音放在一起——讓你親手翻看觀測紀錄，而不是看一個預測模型。Pied tamarins call across urban forest and leave scent on branches. Here, every 5-minute behaviour bout recorded in Manaus sits beside the noise level measured at the time — so you can browse the real observations yourself, not a predictive model.

氣味表 · 片段Scent table · bouts
787
50 段有標記 · 5 個群體50 bouts with marking · 5 groups
叫聲表 · 片段Call table · bouts
1862
466 段有叫聲 · 共 4164 次 · 8 群體466 bouts with calls · 4164 calls · 8 groups
噪音實測範圍（1 分鐘 C 加權 LCeq）Measured noise (1-min C-weighted LCeq)
40.2–84.7 dB · 氣味表 49.2–84.7· scent table 49.2–84.7
787／1862 是觀測片段，不是動物數量；兩表分開，不可相加。are observation bouts, not numbers of animals; the two tables are separate and never added.
黑白花狨在樹枝上發出長叫聲並留下氣味標記，遠處城市噪音 / A pied tamarin long-calls and scent-marks on a branch while city noise hums in the distance 城市噪音 city noise 氣味標記 scent mark 長叫聲 long call 示意圖 · 非實測位置 illustration · not measured positions
下方可聽研究資料庫的實地錄音；聲景示意圖為程式繪製。Below you can hear a field recording from a research database; the soundscape illustration is code-drawn.

資料版本Data version

noise-observations-v1

分析範圍Scope

清理後的觀測片段描述與經驗抽樣。不是訓練好的 world model，不預測未來、過橋、物種恢復或因果效應；也不是今天的即時監測。Description and empirical resampling of cleaned observation bouts. Not a trained world model; no forecasts of the future, crossing, recovery or causal effects; not live monitoring.

尚未解決Unresolved

  • 原文 9 群體／5 片森林，副本讀到 8 群體／6 個地點標籤，待核對。Paper: 9 groups / 5 forests; this copy reads 8 groups / 6 site labels — to be checked.
  • 座標像投影座標、CRS 未知：不畫精確地圖、不算距離。Coordinates look projected, CRS unknown: no exact map, no distances.
  • Metadata 保留舊投稿資訊，不能宣稱與最末原始 XLSX 逐位元組相同。Metadata keeps older submission info; not claimed byte-identical to the final original XLSX.

先聽見牠Listen first

一段 Parque do Mindú（馬瑙斯）的實地錄音，片段經人工標注為黑白花狨。先聽原錄音，再加入合成背景噪聲，比較叫聲是否變得難以分辨。A field recording from Parque do Mindú (Manaus); the clips are manually annotated as pied tamarin. Listen to the original, then add synthetic background noise and compare.

播放器需要 JavaScript。可直接開啟The player needs JavaScript. Open the 原始錄音檔（約 99 MB）original recording (~99 MB).

片段（人工標注 Pied tamarin）Clip (manually annotated Pied tamarin)

合成背景噪聲示意，非馬瑙斯現場錄音Synthetic background noise, not a Manaus field recording

你覺得比較難聽見了嗎？這是人類聽覺的混音比較，不測量黑白花狨的聽力或行為反應；裝置音量不等於野外分貝。Did it get harder to hear? This is a mixing comparison for human ears; it does not measure pied tamarin hearing or behaviour, and device volume is not field decibels.

  • 叫聲是原實地錄音（包含原有背景），不是乾淨分離聲軌，不保證只有猴子。本站不生成猴叫，也不宣稱重現黑白花狨的聽覺。The call is the original field recording (with its own background), not a clean isolated track, and may contain more than tamarins. No tamarin calls are generated, and tamarin hearing is not simulated.
  • 錄音來自另一項聲音監測資料庫，未與本站 2018–2019 觀測表逐事件配對；播放結果不是資料模型或因果證據。The recording comes from a separate acoustic-monitoring database and is not matched event-by-event to this station’s 2018–2019 tables; listening is not a data model or causal evidence.
  • 錄音來源：Colonna 等研究團隊公開的 Sauim 資料庫；地點依檔案說明為 Parque do Mindú，Manaus；單檔錄音者與日期未獨立核實（檔名含 02.02.19，勿當作核實日期）。Recording source: the public Sauim database by Colonna and colleagues; location per the file notes is Parque do Mindú, Manaus; recordist and date of this file are not independently verified (the filename contains 02.02.19 — not a verified date).
來源與許可Source & licence
  • Sauim @ 4049e79 · records/README.md · 人工標注檔manual annotations · LICENSE
  • 參考論文（聲音偵測研究，非氣味表原論文）：Reference paper (acoustic detection study, not the scent-table paper): doi:10.1016/j.ecoinf.2026.104003
  • 儲存庫根目錄 LICENSE 為 MIT；無單獨的音訊授權檔。本頁依儲存庫所附許可串流來源音訊並署名，不宣稱 CC0，也不代表另獲錄音者授權。The repository root LICENSE is MIT; there is no separate audio licence file. This page streams the source audio with attribution under the repository’s licence; it does not claim CC0 or separate permission from the recordist.
MIT License

Copyright (c) 2025 Dr. Juan Gabriel Colonna

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

資料實驗室Data lab

三個工作台：看真實觀測、從紀錄重新抽樣、核對每一步清理。所有數字都來自頁內嵌的匯總頻數。Three benches: look at real observations, resample from the records, and check every cleaning step. Every number comes from the frequency summary embedded in this page.

看觀測 · 各噪音箱的真實片段Observe · real bouts in each noise bin

互動圖需要 JavaScript。核心數字見上方摘要與「核對資料」表格。The interactive chart needs JavaScript. Key numbers are in the summary above and the Check data tables.

指標Measure
噪音區間（10 dB 固定箱）Noise range (fixed 10 dB bins)
→

  • n<20 標為樣本少——這是展示門檻，不是統計顯著門檻。n<20 is flagged as a small sample — a display threshold, not a significance threshold.
  • n=0 顯示「無記錄」，不補零。n=0 shows “no records”; never filled with zero.
  • 「全部群體」按觀察片段計權，不是各群體同權。“All groups” is weighted by bouts, not equally by group.
各箱裡的群體組成差很多（例如某些群體幾乎只出現在 50–60 dB），群體差異可能左右整體看起來的關係。本頁刻意不畫單調趨勢線或因果預測線。Group composition differs a lot between bins (some groups appear almost only in 50–60 dB), so group differences can drive the apparent overall pattern. No monotonic trend or causal prediction line is drawn — on purpose.

從記錄重新抽樣 · 經驗抽樣，不是新實驗Resample observations · empirical draws, not new experiments

抽樣工作台需要 JavaScript。The resampling bench needs JavaScript.

指標Measure
抽樣次數Number of draws
  • 每次抽樣以頻數為權重、有放回地抽出一個已觀察到的片段值。Each draw picks one observed bout value, with replacement, weighted by frequency.
  • 抽 100 次不會讓原始樣本變多；原始片段數 n 不變。Drawing 100 times does not enlarge the original sample; n stays the same.
  • 這不是新動物實驗、不是新的連續軌跡；不提供生物信賴區間、顯著性或未來機率。Not a new animal experiment or a new continuous track; no biological confidence intervals, significance or future probabilities.
  • 噪音箱是野外量到的環境噪音，不是設備音量。氣味與叫聲分開抽樣，不能組成「同時發生」的事件。Noise bins are measured ambient noise, not a device volume. Scent and calls are sampled separately and never combined into joint events.

核對資料 · 從原始列到分析片段Check data · from raw rows to analysed bouts

指標Measure 原始列Raw rows 必要欄位完整Complete 排除衝突鍵列Conflict rows excluded 分析片段Analysed bouts 陽性片段Positive bouts 反應值總和Outcome sum 群體Groups 日期Dates LCeq dB
scent_mark （0/1）(0/1)8017914787505054949.2–84.7
long_call （次數）(count)19101876141862466416487240.2–84.7

缺失與重複規則Missing & duplicate rules

  • 每列是一段 5 分鐘行為片段；噪音是 1 分鐘 C 加權 LCeq。Each row is a 5-minute behaviour bout; noise is a 1-minute C-weighted LCeq.
  • long_call = 長叫聲次數；scent_mark = 是否發生（0/1），不是氣味標記次數。long_call = number of long calls; scent_mark = whether marking occurred (0/1), not a count.
  • 只要求本指標與 LCeq 有值；未使用的座標缺失不刪列。Only this measure and LCeq must be present; rows are not dropped for missing unused coordinates.
  • 以「日期＋時間＋地點＋群體」辨識重複事件鍵；衝突鍵的所有列保守排除，不裁決哪一列正確。Duplicate event key = date + time + site + group; all rows of a conflicting key are conservatively excluded, without judging which is right.
  • 兩表分別分析、不相加；兩表有 787 個共同事件鍵，也不可相加。The two tables are analysed separately and never summed; their 787 shared event keys must not be added either.

群體、地理與版本問題Group, geography & version issues

  • 原文報告 9 群體、5 片森林；副本讀到 8 群體、6 個地點標籤，待核對。The paper reports 9 groups in 5 forests; this copy reads 8 groups and 6 site labels — pending verification.
  • 座標看起來是投影座標但 CRS 未知，本頁不畫精確地圖、不計算距離。Coordinates appear projected but the CRS is unknown; no exact map, no distances.
  • Metadata 保留舊投稿資訊；不能宣稱與最末原始 XLSX 逐位元組相同。Metadata retains older submission info; not claimed byte-identical to the final original XLSX.
  • 來源是使用者提供的 Dryad 資料 Numbers 副本，轉換後唯讀分析。Source is a user-supplied Numbers copy of the Dryad data, converted for read-only analysis.
  • 研究在 2018–2019 年的馬瑙斯進行，不冒充今日即時監測。Fieldwork was in Manaus, 2018–2019 — not live monitoring today.

論文方法與本頁的差別Paper methods vs this page

  • 論文以 GLMM 等模型分析；本頁只做分箱描述與經驗抽樣，不估係數、不給 p 值或信賴區間。The paper uses models such as GLMMs; this page only does binned description and empirical resampling — no coefficients, p-values or confidence intervals.
  • 本頁的清理樣本不是原作者的最終模型樣本，不宣稱完整重現論文 GLMM。The cleaned sample here is not the authors’ final model sample; no claim of fully reproducing the paper’s GLMM.
  • 10 dB 分箱是本頁的展示選擇；端箱只部分被實測值覆蓋，不外推到整個箱邊界。10 dB bins are a display choice here; end bins are only partly covered by measured values and are not extrapolated to the full bin edges.
  • 不轉載論文中的受版權保護圖表。No copyrighted figures from the paper are reproduced.

欄位字典（內置匯總）Field dictionary (embedded summary)

groups
群體名稱，索引由 0 起。Group names, indexed from 0.
binEdges
[40,50,60,70,80,90]，箱 i = [edgei, edgei+1)。[40,50,60,70,80,90]; bin i = [edgei, edgei+1).
tables.*[ ]
[群體索引, 噪音箱索引, 觀測到的反應值, 頻數]。[group index, noise bin index, observed outcome value, frequency].
n
同指標頻數總和。Sum of frequencies.
比例 / 均值
sum(value×freq)／n（氣味＝發生比例；叫聲＝每片段平均長叫聲）。sum(value×freq)/n (scent = occurrence share; calls = mean long calls per bout).
positive
value>0 的頻數總和（有叫聲／有標記的片段）。Sum of freq where value>0 (bouts with calls / marking).
quality
清理各步驟列數、群體數、日期數與噪音範圍。Row counts per cleaning step, groups, dates, noise range.

下載（本機產生，不經網路）Downloads (generated locally, no network)

JSON 為頁內嵌資料原文；資料授權 CC0（Dryad 10.5061/dryad.np5hqbztm，2021 版）。The JSON is the embedded data verbatim; data licence CC0 (Dryad 10.5061/dryad.np5hqbztm, 2021 version).

顯示真實頻數表（核驗圖表用）Show the real frequency tables (to verify charts)

三個常被問錯的問題Three questions people get wrong

噪音更大，就一定更多叫聲嗎？Louder noise means more calls?

不能這樣推斷Can’t infer that

箱與箱之間混著不同群體、地點與日期；全部群體的匯總按片段計權，某個大群體就能主導一個箱。端箱樣本也少。描述性分箱沒有控制這些因素，所以本頁不畫趨勢線。Bins mix different groups, sites and dates; the all-groups summary is bout-weighted, so one large group can dominate a bin, and end bins are small. Binned description controls for none of this — so no trend line is drawn.

氣味和聲音是完全替代嗎？Do scent and sound fully replace each other?

論文未支持完全替代The paper did not support full substitution

本頁兩表分開分析、分開抽樣，無法也不會組成「同一片段裡兩者同時發生」的聯合事件；共同事件鍵也不相加。Here the two tables are analysed and resampled separately; no joint “both in the same bout” events are built, and shared event keys are never added.

這份資料能預測牠會不會過橋嗎？Can this data predict bridge crossing?

不能No

資料裡沒有過橋行為，這也不是 world model。過橋的問題請看「牠會過橋嗎」站。There is no crossing behaviour in this data, and it is not a world model. For crossing, visit the “Will they cross?” station.

來源 · 觀測表Sources · observation tables

  1. Dryad 原始資料集Dryad original datasetDOI 10.5061/dryad.np5hqbztm · CC0 · 2021 版本；本頁使用使用者提供的 Numbers 副本2021 version; this page uses a user-supplied Numbers copy
  2. 對應論文（Ethology Ecology & Evolution）Corresponding paper (Ethology Ecology & Evolution)DOI 10.1080/03949370.2023.2248591 · 2023 線上／2024 卷期online 2023 / issue 2024
  3. 論文 PDF（Rufford Foundation）Paper PDF (Rufford Foundation)僅連結，不轉載圖表Linked only; figures not reproduced

來源 · 錄音（先聽見牠）Sources · recording (Listen first)

  1. Sauim 聲音資料庫（Colonna 等）Sauim acoustic database (Colonna et al.)commit 4049e79 · MIT (Copyright (c) 2025 Dr. Juan Gabriel Colonna) · 非 CC0not CC0
  2. Mindu_Saguinus_bicolor_02.02.19-000.wavParque do Mindú 完整實地錄音 · 約 99 MB · 片段依人工標注Full field recording, Parque do Mindú · ~99 MB · clips from manual annotations
  3. 聲音偵測研究論文（Ecological Informatics）Acoustic detection paper (Ecological Informatics)DOI 10.1016/j.ecoinf.2026.104003
介紹主站Intro station你好，黑白花狨認識這種城市森林裡的小靈長類Meet the small primate of the urban forest 過橋站Crossing station牠會過橋嗎？Will they cross?樹冠橋與道路切割的問題Canopy bridges and roads that cut the forest