台灣特有種・冰河孑遺
完整授權說明見 CREDITS.txt。地圖分布點的座標經資料提供者模糊化，僅示意大致區域。
Full license details are in CREDITS.txt. Distribution coordinates were generalized by the data providers and show approximate areas only.
「依海拔資料」模式：每種山椒魚目前住在海拔帶 [L, U]，所在山系最高峰為 P。 氣溫約每升高 100 m 下降 0.6 °C（氣溫垂直遞減率），所以升溫 ΔT，適合的海拔帶往上移 s = ΔT ÷ 0.6 × 100 m。 棲地剩下 [L+s, min(U+s, P)]，族群大小與剩下的海拔帶寬度成正比；整個帶被推過山頂即消失。
紅皮書等級（國家地理 2026/9）：南湖 極危；觀霧、台灣、楚南氏 瀕危；阿里山 易危。
升溫情境：IPCC AR6 最高排放情境（SSP5-8.5）2081–2100 年全球平均比工業化前高約 4.4 °C； 現在已高約 1.1 °C，所以從 2025 年起到 2100 年「再升溫約 3.3 °C」，之後假設停住、時間繼續。（未考慮高山升溫可能比全球平均快。） 結語提到的低排放情境（SSP1-2.6）：本世紀末約比工業化前 +1.8 °C，相對現在約再升 0.7 °C。
長期機制（方向有文獻支持，數值為示意）：①遷移落後——下緣隨升溫失去，往上的新棲地每年最多擴散 2 m； ②長期壓力——高溫持續造成微棲地變乾、食物（小型無脊椎動物）減少、植被改變，族群逐年下降，棲地被壓縮得越多的族群下降越快（基準每年 2.5 %）； ③小族群風險——族群很小時，每代有 2 % 機率因隨機事件消失（滅絕漩渦）； ④其他威脅（依國家紅皮書等級校準）——棲地破壞、登山與山屋干擾、林道、污染等。借用 IUCN 滅絕機率標準的時間尺度： 極危約 30 年內 50%、瀕危約 50 年內 20%、易危 100 年內 10%，換算成族群逐年下降，與暖化效果疊加。 這代表「假設紅皮書反映的風險延續、威脅沒有減少」，不是預測；山椒魚實際上可能是依分布範圍等其他標準被評定的。所以消失的年份與順序每次播放都會略有不同。
音高：海拔每 100 m 對應 1 個半音，旋律往上爬的幅度與碰到山頂的位置都由海拔換算。
限制：這是極簡化的示意模型，不是科學預測。沒有考慮山頂面積越高越小、石縫與溪邊微棲地的避難效果、 生理耐熱極限、遷移能力、棲地破碎與登山、山屋、林道等人為干擾；各山系山頂取值為概略，台灣山椒魚尤其不確定。
分布、海拔、趾數、紅皮書等級：《國家地理雜誌》中文版 2026 年 9 月號（第 298 期）〈冰河留給臺灣的瓶中信：山椒魚〉及資訊圖〈臺灣的山椒魚〉（分布圖：朱有田教授）； 分布範圍越小越脆弱（④）的權重依該文的分布描述與國家紅皮書等級設定，數值為示意。地圖上的分布：依上述分布圖描繪（含阿里山與台灣山椒魚的雜交帶，兩色交錯）。
前言介紹來源：報導者（台灣為山椒魚分布最南界、1919 年楚南仁博發現）、中央社（小鯢科祖先與恐龍同期）、 公視（2025 年中文名更正、趾數）、Li, Fu & Lei 2011, PLoS ONE（山椒魚屬自日本西南部擴散至台灣的時間推估）。
「示意」模式：為了音樂結構讓五族依序消失，數值是設定的，不代表實際脆弱程度（只有排序參考海拔）。
Elevation-data mode: each species currently lives in an elevation band [L, U]; the highest peak in its range is P. Air temperature drops about 0.6 °C per 100 m of elevation (the lapse rate), so warming of ΔT moves the suitable band up by s = ΔT ÷ 0.6 × 100 m. The remaining habitat is [L+s, min(U+s, P)]; population size is proportional to the width of the remaining band, and a species disappears when the whole band is pushed past the summit.
National Red List status (National Geographic, Sept 2026): Nanhu Critically Endangered; Guanwu, Taiwan and Sonan's Endangered; Alishan Vulnerable.
Warming scenario: under the IPCC AR6 highest-emission scenario (SSP5-8.5), global mean temperature in 2081–2100 is about 4.4 °C above pre-industrial. It is already about 1.1 °C warmer today, so from 2025 to 2100 we add about 3.3 °C, then assume warming stops while time goes on. (Mountains may warm faster than the global average; this is not included.) The low-emission scenario mentioned at the end (SSP1-2.6): about +1.8 °C above pre-industrial by the end of the century, roughly +0.7 °C above today.
Long-term mechanisms (the direction is supported by the literature; the numbers are illustrative): ① Lagging dispersal — the lower edge is lost as it warms, while new habitat higher up is colonized by at most 2 m per year; ② Chronic stress — prolonged heat dries out microhabitats, reduces food (small invertebrates) and changes vegetation, so populations decline year by year, faster for species whose habitat is more squeezed (baseline 2.5% per year); ③ Small-population risk — very small populations have a 2% chance per generation of disappearing through chance events (the extinction vortex); ④ Other threats (calibrated to the national Red List) — habitat loss, disturbance from hiking and mountain huts, forest roads, pollution and more. Borrowing the time scales of the IUCN extinction-probability criterion: Critically Endangered ≈ 50% within about 30 years, Endangered ≈ 20% within about 50 years, Vulnerable 10% within 100 years, converted into a yearly decline and combined with warming. This represents “the risk reflected in the Red List continues and threats are not reduced” — not a prediction; these salamanders may actually have been assessed under other criteria such as range size. Extinction years and order therefore vary from one performance to the next.
Pitch: every 100 m of elevation corresponds to one semitone, so how far a melody climbs and where it hits the summit both come from elevation.
Limitations: this is a highly simplified, illustrative model, not a scientific prediction. It ignores the shrinking area near summits, refuges in rock crevices and along streams, physiological heat tolerance, dispersal ability, habitat fragmentation, and human disturbance such as hiking, mountain huts and forest roads. Summit values are approximate, especially for the Taiwan salamander.
Distribution, elevation, toe counts and Red List status: National Geographic (Taiwan edition), September 2026 (No. 298), “A message in a bottle from the ice age: salamanders” and the infographic “Taiwan's salamanders” (range map by Prof. Yu-Ten Ju). The weighting ④ — smaller, more fragmented ranges are more vulnerable — follows that article's range descriptions and national Red List categories; the numbers are illustrative. Map: traced from that range map (the Alishan–Taiwan salamander hybrid zone is shown with both colors interleaved).
Introduction sources: The Reporter (Taiwan as the southernmost limit; Sonan's discovery in 1919), CNA (hynobiid ancestors date to the age of dinosaurs), PTS (2025 correction of Chinese common names; toe counts), Li, Fu & Lei 2011, PLoS ONE (timing of Hynobius dispersal from southwestern Japan to Taiwan).
Illustrative mode: the five species disappear one after another for the sake of musical structure; the numbers are set by hand and do not reflect actual vulnerability (only the order follows elevation).