跳至正文

概率句被读成了因果句

目录

机制裁决第 145 篇 · 对称双向第 140 篇 · section H 社会/经济/集体 · 全库第 203 篇 本篇审的是「这场灾难是气候变化造成的」这句话。先读三句红线：

  1. 本篇不裁决任何气候政策、赔偿诉讼或减排责任的是非，不给任何政治立场背书，不评价任何在世的归因科学家的人品或动机——只审这句话从文件里读出来时承重承不承得住。凡涉及利益关系，只登记公开披露的事实，不推断因果。
  2. 「归因结论错了」与「归因结论被读错了」是两件不同的事。本篇的裁决主要落在后者：没有发现概率式事件归因的方法学地基是假的，发现的是它产出的概率句在传播中被换成了因果句，而几乎所有被换掉的限定词都印在造尺人自己的文件里。
  3. 全篇承重句均给出可点击来源；英文逐字引用一律标注取回方式。本篇引用的所有数字（概率比、风险倍数、温度变化、时间线、资金额）都给出出处，自算部分先复现官方数字校验算法、通过后才使用，算式写进正文，可复现。

零、一句话裁决

「这场灾难是气候变化造成的」这句话里，每个字都能在造尺人自己的文件里找到邻居，而整句话作为一句完成时的因果陈述，是造尺人自己明文说不能说的。

归因科学给出的标准产品不是因果句，是概率句：IPCC AR6 第 11 章对概率式方法的定义逐字就是「anthropogenic climate change made this event type twice as likely」（使该类事件可能性加倍）[一手逐字]；IPCC FAQ 11.3逐字「Scientists cannot answer directly whether a particular event was caused by climate change」[一手逐字]；开山之作 Stott 2004逐字「It is an ill-posed question whether the 2003 heatwave was caused, in a simple deterministic sense, by a modification of the external influences on climate」[一手逐字]；WWA 自己的记者指南逐字「climate change cannot cause an event (in a binary use of the term ’cause’)」[一手逐字]。五份文件，五个方向的同一句话：概率句是唯一的合法产品，因果句是读法问题。

同时，反方向的四句话同样不立：归因科学不是模型游戏（观测在第一步、模型阈值锚定观测、有预测且被十年后的观测回验）；快速归因不是没有质量控制的新闻稿（方法本身经过同行评审、个案事后送审、约四分之一入刊且「主要发现从未改变」是官方自陈）；「每场极端天气都是气候变化造成的」不成立（桑迪无人给出确定性归因、2010 俄罗斯热浪的权威结论是「主要自然变率」、WWA 自己做过「现在比过去冷十五倍」的寒潮）；「归因全是炒作」也不成立（2021 太平洋西北热浪的观测温度本身就破纪录 5°C，这是任何模型都抹不掉的事实）。

而本篇真正独占的一格在第六、七层：同一个事件，三种合法的事件定义给出横跨两个数量级的「稀有度」答案——WWA 自己把「1-in-1000 年」写进假设、独立研究估计「1-in-100,000 年」、另一篇综述把三档估计并列印在同一句话里；同时，把归因概率句直接搬进侵权诉状（「热穹顶不是自然天气事件，是被告直接且可预见的结果」）的诉讼文本，把 WWA 原文里那行「假设该事件不是由新的非线性过程造成」的限定词省略了。 概率句不是谎言，被当成因果句用时才是别的东西。

一、本篇测什么

1.1 被审对象

被审的是一句话及其用法，不是一门学科的对错。

主张的谱系从学术句到标题句：

  • 学术句：IPCC AR6 WG1 第 11 章 11.2.3 对概率式方法的官方定义逐字「anthropogenic climate change made this event type twice as likely」[一手逐字]；
  • 快速归因句：WWA 2021 年太平洋西北热浪分析逐字「was virtually impossible without human-caused climate change」「at least 150 times rarer without human-induced climate change」[一手逐字]；
  • 标题句：BBC「US-Canada heatwave ‘virtually impossible’ without warming」、Guardian「Pakistan floods ‘made up to 50% worse by global heating’」[一手逐字]。

1.2 结构胎记

概率跳 × 速度跳 × 尺子账 ＋ 反向红跳。

  • 概率跳：把「反事实模拟下的事件类概率比（PR/FAR）」读成「这场事件被气候变化造成」的因果句。
  • 速度跳：把「事件后数天至数周发布的快速归因（未经传统同行评审）」读成「已确立的科学结论」。
  • 尺子账：事件定义（区域、季节、阈值、历时）的选择自由度决定答案——归因陈述对定义敏感是造尺人自陈的（IPCC、NASEM、WWA 三方都写过）。
  • 反向红跳：对称审计反方向的四句话（模型游戏／不可证伪／每场都是／全是炒作）。

三跳共用的动作是同一个：把一个带完整限定语的概率陈述，剥掉限定语之后当成一个完成时的因果陈述使用。而所有限定语都印在造尺人自己的文件里。

1.3 落位：归因三件套的收口

本库已有两篇「归因」主题：

  • 空气污染篇（全库第 182 篇）：人群归因分数——「每年 X 万人因 PM2.5 死亡」，分母是人群；
  • 长新冠篇（全库第 170 篇）：个体症状归因——「我的症状是不是长新冠」，分母是个体。

本篇是第三件：单次极端天气事件的归因（event attribution）。三篇共用「反事实」这把尺，但分母不同：人群健康负担、个体症状、单次天气事件。分界在 1.5 节写死。

1.4 去重实测

用 python 逐词扫全库（不用 grep，本环境的 ugrep 对大文件会静默返回空，见 runbook）。全库 202 篇 / 12,154,136 字符，检索 极端天气／天气归因／气候归因／事件归因／归因于气候／WWA／World Weather Attribution／event attribution／probability ratio／fraction of attributable／storyline／rapid attribution／热浪／洪水／干旱／暴雨／extreme weather 等 25 组词：

  • 中文归因词（天气归因/气候归因/事件归因/归因于气候/气候变化归因）全部零命中；「极端天气」仅 1 处、「飓风」仅 1 处，都在数字孪生篇（DestinE「极端天气孪生」模型平台语境，与归因无关）。
  • WWA／World Weather Attribution／event attribution／probability ratio／fraction of attributable／storyline／rapid attribution 全部零命中；「热浪/洪水/干旱/暴雨」全库零命中。
  • 非零命中逐条打印语境后全部判为别义：「attribution」5 篇命中全是特征归因/属性归属的通用义（机制可解释性篇 13、UAP 篇 2、维生素 D 篇 2、费米悖论篇 1、二甲双胍篇 1）；「counterfactual」24 篇命中全是反事实思维的哲学/因果推断通用义；「IPCC AR6」3 篇命中（地球工程 14 处、临界点 5 处、碳核算 3 处）均为正常引用不构成覆盖。

结论：极端天气事件归因专篇全库零命中，处女地。

1.5 边界对四篇写死

  • 与空气污染篇（第 182 篇）：那篇审「人群归因分数」——把暴露-反应函数积分到人群得到的反事实死亡数（PAF/GBD），分母是人群；本篇审「事件归因」——把今日气候与反事实气候的概率分布相比得到的概率比，分母是事件类。同一把「反事实」尺的两个用途，不互相覆盖。
  • 与长新冠篇（第 170 篇）：那篇审「个体归因」——「我的症状是否可归因于感染」；本篇不审任何个体层面的归因。
  • 与地球工程篇（第 194 篇）：那篇审太阳辐射修正的模型-部署叙事；本篇不碰任何干预手段。
  • 与临界点篇（2026-07-03）：那篇审气候系统临界点与盖亚叙事；本篇管单次事件的归因陈述，不碰系统突变。
  • 与数字孪生篇（2026-07-26）：DestinE「极端天气孪生」是气候模型平台（4.4 公里分辨率预报），不是归因研究；本篇不审预报模型。

1.6 方法备案

  • 取证通路：IPCC AR6 WG1 第 11 章与 SPM PDF（curl 直取 200）、NASEM 2016 报告分章（nap.edu 加 --compressed）、Philip et al. 2020 协议（Copernicus 系直取 200）、WWA 官网各页（/analyses/、/methods/、/faqs/、/by-year/、具体分析页全部 200，仅首页偶发连接失败）、nature.com（实测可直连 200）、Copernicus 系期刊（esd/ascmo 直取 200）、Crossref API（DOI 校验）、Wayback id_ 原件（新闻站与 Wiley/PNAS 403 时）。
  • 承重统计一律 python3（ugrep 对大文件静默返回空）。
  • 任务书三处 URL 勘误（已由子代理实测纠正并登记）：nature02367 实为无关论文（祖母假说），Allen 2003 正确入口为 https://www.nature.com/articles/421891a；nclimate1457 实为访谈《Living in the doughnut》，Trenberth 2012 归因文实为 Climatic Change 的 10.1007/s10584-012-0441-5，Trenberth 2015 三人合著实为 10.1038/nclimate2657；nclimate3411 实为另一文，Shepherd 2016 实为 Current Climate Change Reports 的 10.1007/s40641-016-0033-y。另：Dole 2011 发表于 GRL 非 Nature Geoscience；Rahmstorf & Coumou 的莫斯科数字是 fivefold/80% 而非「约 3 倍」；RWE 案终局裁定为 2025-05-28 而非 2024。

二、核验标记与十二层速览

核验标记沿用本库四档：[一手逐字]＝原文逐字取自实际取回文件；[文献较稳]＝多源一致或权威评估；[需亲核]＝正文未全取回或版本存疑；[多源检索]＝方法学阴性检索。

层 名称 核心问题 裁决方向
三 守真锚 归因科学的地基硬不硬 一条不动
四 定义账 「概率比」的官方定义是什么 全部印在文件里
五 概率跳·语言账 概率句如何变成因果句 造尺人自己禁止
六 误读账 三组「原文 vs 标题」并排 限定词逐级脱落
七 速度跳·制度账 快速归因的制度设计 3–25 天、中位 9 天
八 评审账 先发布后评审的辩护与代价 官方自陈「从未改变」
九 尺子账·定义自由度 同一事件几种答案 跨两个数量级
十 边界账 桑迪与故事线 「答案不再关于实际事件」
十一 制度账 损失与损害基金 链已通、款未放
十二 消费账 诉讼、再保险、媒体、公众 限定词在消费端脱落
十三 反向红跳 四个反向说法 全部不立
十四 母裁决 一句话裁决 见末章

（注：目录编号按本库惯例从零起，层号与节号对应；上表「十三/十四」即正文「八/九」章。）

三、守真锚（这一侧一条都不动）

1. 全球变暖是真的，人类是主因，热浪变多变强是最高置信度结论。 IPCC AR6 WG1 第 11 章执行摘要逐字「It is an established fact that human-induced greenhouse gas emissions have led to an increased frequency and/or intensity of some weather and climate extremes since pre-industrial time, in particular for temperature extremes.」[一手逐字]；SPM A.3.1逐字「It is virtually certain that hot extremes (including heatwaves) have become more frequent and more intense across most land regions since the 1950s… with high confidence that human-induced climate change is the main driver of these changes. Some recent hot extremes observed over the past decade would have been extremely unlikely to occur without human influence on the climate system.」[一手逐字]。注意限定词：some（某些）热极端、main driver（主因，SPM 脚注 14 逐字「’main driver’ means responsible for more than 50% of the change」[一手逐字]）——最高置信度句也自带限定。

2. 事件归因是正经学科，且有官方制度文本。 AR6 以专章（第 11 章）系统处理极端事件归因（技术摘要逐字「A dedicated chapter in AR6 (Chapter 11) concludes…」[一手逐字]；「首次专章」的表述无据，AR5 已有第 10 章 D&A）；NASEM 2016是国家级评估；WWA 的方法协议 Philip et al. 2020（DOI 10.5194/ascmo-6-177-2020）发表在同行评审期刊并被 AR6 第 11 章引用。

3. 官方对「事件归因」的定义是概率式的。 AR6 第 11 章 11.1.4 逐字「the attribution of extreme weather events, or the investigation of changes in the frequency and/or magnitude of individual and local- and regional-scale extreme weather events due to various drivers」[一手逐字]；11.2.3 逐字「produces statements such as ‘anthropogenic climate change made this event type twice as likely’… This is done by estimating probability distributions of the index characterizing the event in today’s climate, as well as in a counterfactual climate」[一手逐字]；FAQ 11.3逐字「Scientists cannot answer directly whether a particular event was caused by climate change, as extremes do occur naturally… Instead, scientists quantify the relative importance of human and natural influences on the magnitude and/or probability of specific extreme weather events.」[一手逐字]。

4. 概率比与 FAR 有明确定义式。 NASEM 2016 Summary逐字「FAR = (p1 – p0)/p1 — or the risk ratio (RR)—RR = p1/p0」[一手逐字]；Philip 2020 协议逐字「PR = p1/p0, with p1 the probability of an event as strong as or stronger than the extreme event in the current climate and p0 the probability in a counterfactual climate without anthropogenic emissions」[一手逐字]；Stott 2004 Methods 给出同式。概念源头：Allen 2003（Nature 421, 891–892）逐字「We will never be able to say, at any confidence level, that human influence has contributed x% to an actual weather event. What we can say is that past greenhouse-gas emissions are likely… to have increased the risk of that event over its preindustrial value.」[一手逐字]——概率归因思想的奠基句，它同时是反方向的锚：创始人明说「永远无法说人类影响对实际事件贡献了 x%」。

5. 部分归因结论极稳，且有预测-回验闭环。 Stott 2004（人类归因研究第一篇）逐字「we estimate it is very likely (confidence level >90%) that human influence has at least doubled the risk of a heatwave exceeding this threshold magnitude」[一手逐字]；Christidis 2015十年后回验逐字「Human influence at least doubled the chances of the event according to the first formal event attribution study, which also made the ominous forecast that severe heatwaves could become commonplace by the 2040s. Using a previously employed temperature threshold… events that would occur twice a century in the early 2000s are now expected to occur twice a decade.」[一手逐字]——归因研究自带可检验预言，且被新观测回验。

6. 观测先于模型，模型阈值锚定观测。 Philip 2020 协议 Step 3 逐字「The first task is to analyse observational data.」[一手逐字]；Step 4 逐字「The threshold value is again obtained by means of setting the return period in the current conditions’ run equal to the value from observational data.」[一手逐字]；Step 4 验证条款逐字「Before they are used one must ascertain that the models are fit for purpose, i.e. that they can represent the particular extreme under study well enough.」[一手逐字]。

四、定义账：从 FAR 到「probability-based approach」

归因陈述的语言谱系有一条清晰的起源链：

  • Allen 2003（Nature 421, 891–892）：提出按风险变化分派责任的思想，逐字「If, at a given confidence level, past greenhouse-gas emissions have increased the risk of a flood tenfold, and that flood occurs, then we can attribute, at that confidence level, 90% of any damage to those past emissions.」[一手逐字]（90% = 1 − 1/10，即 FAR 的雏形；注意原文未用 FAR 术语）。
  • Stott 2004（Nature 432, 610–614）：把 ill-posed 与可回答的问题分开，逐字「It is an ill-posed question whether the 2003 heatwave was caused, in a simple deterministic sense, by a modification of the external influences on climate—for example, increasing concentrations of greenhouse gases in the atmosphere—because almost any such weather event might have occurred by chance in an unmodified climate. However, it is possible to estimate by how much human activities may have increased the risk of the occurrence of such a heatwave.」[一手逐字]；并给出责任阈值逐字「the figure of relevance in establishing possible liability for compensation has been FAR = 0.5, corresponding to a doubling of risk over natural conditions」[一手逐字]。
  • NASEM 2016：把 FAR/RR 公式写进国家评估（见守真锚 4）。
  • AR6 第 11 章：官方术语定为「probability-based approach」，结论模板逐字就是「made this event type twice as likely」（见守真锚 3）。Hulme 2014（作者提交稿）给出 FAR 的等价式逐字「FAR is simply 1-P0/P1. FAR may vary between 0 and 1 and expresses the fraction of a weather risk… that can be attributed to a specified influence.」[一手逐字]。
  • Trenberth 2012（Climatic Change 115:283–290）：对「因果问法」的釜底抽薪，逐字「The answer to the oft-asked question of whether an event is caused by climate change is that it is the wrong question. All weather events are affected by climate change because the environment in which they occur is warmer and moister than it used to be.」[一手逐字]。
  • Trenberth, Fasullo & Shepherd 2015（Nature Climate Change 5:725–730）：区分两种归因，逐字「The first relates the particular extreme event to the associated weather and weather patterns… they are really a description of the event, not a cause.」与「it is not possible to attribute a single climate extreme event, which by definition is unique and which has a large element of chance in its occurrence, to a specific cause」[一手逐字]；摘要逐字「Headlines attributing an event solely to either human-induced climate change or natural variability can be misleading when both are invariably in play.」[一手逐字]。

术语账的关键观察：从 Allen 的「风险倍数」到 AR6 的「probability-based approach」，造尺人一路都在说同一件事——产品是概率变化，不是事件归因于。而公众问句（Hulme 命名为「extreme weather blame question」[一手逐字]）要的是后者。两者之间的差价，就是概率跳。

五、概率跳·语言账：从「made X more likely」到「caused by」

5.1 WWA 句式谱系（2015–2024，十篇样本逐字）

WWA 分析页的结论句式有一个可追溯的演化：

年份 事件 结论句（逐字） 句法形态
2015 欧洲热浪 「virtually certain that the heatwave… was more likely to happen now than in the past due to climate change」+「twice as often over a large part of Europe」 more likely due to
2017 Harvey 降雨 「human-caused climate change made the record rainfall… roughly three times more likely and 15 percent more intense」 made X more likely AND more intense
2020 西伯利亚热浪 「made at least 600 times more likely」+「would have effectively been impossible without human-induced climate change」+「even with climate change, the prolonged heat was a very rare event expected to occur less than once every 130 years」 倍数＋impossible＋自带频次限定
2021 太平洋西北热浪 「virtually impossible without human-caused climate change」「about a 1 in 1000 year event in today’s climate」「at least 150 times rarer without human-induced climate change」「about 2°C hotter」 impossible＋1-in-N＋rarer 方向
2021 西欧洪水 「increased by a factor between 1.2 and 9」+「pushes the limits of what current methods of extreme event attribution are designed for」 区间式＋自限
2022 巴基斯坦洪水 「about 75% more intense」「about 50% more intense」+「rendering it infeasible to quantify the overall role of human-induced climate change」 强度差＋无法量化自限
2022 英国 40°C 「made the event at least 10 times more likely」+「1 in 1000 years in the current climate」+「Note that return periods… are therefore highly uncertain」 at least 下限＋限定
2023 北美/欧洲/中国热浪 「would have been virtually impossible」「2.5 °C warmer in Southern Europe, 2 °C warmer in North America and about 1 °C in China」 impossible＋°C 差
2024 飓风海伦妮 「about 10% heavier due to climate change」「made about 40% and 70% more likely」「climate change was responsible for an increase of about 150% in the number of such storms」 标题即「key driver of catastrophic impacts」
2026 加拿大野火 「made a factor of about 3.5 more likely with 1.4°C of global warming」+「we cannot attribute a change in 24-month dry conditions to human-induced climate change」 倍数与否定句并存

（全部为 curl 直取 200 后逐字摘录，抓取件存 tmp/extreme-weather-attribution/raw/。）

谱系的观察：概率差（PR）与强度差（ΔI）自 2017 年起被压缩进「made X more likely/intense」的单一因果动词；「impossible」句式 2020 年出现并逐年加强；2024 年标题从「made」升级到「key driver of catastrophic impacts」——句式演化的方向与限定词脱落的方向同向。同时每一篇都自带限定句（「cannot attribute the trends to specific causes」「infeasible to quantify」「highly uncertain」「cannot say whether this was a so-called ‘freak’ event」[一手逐字]），这些限定几乎从不出现在媒体标题里。

5.2 「at least」的脱落：造尺人亲口承认的传播机制

Philip et al. 2020 协议 9.8 节「Language」逐字：「a phrase like ‘at least’ was found to be dropped in the majority of popular accounts. Secondly, quoting only the lower bound de-emphasizes the most likely result and therefore communicates too conservative an estimate」[一手逐字]——协议自己承认「至少」在大多数大众转述中被丢弃。这是概率跳的语言学机制：下限被读成定值，区间被读成点。「at least doubled」（Stott 2004）、「at least 150 times rarer」（PNW 2021）、「at least 10 times more likely」（英国 2022）——同一句法，同一命运。

5.3 三组误读并排（学术原句 vs 媒体标题）

案例一：Harvey 2017。 学术原句（WWA 分析）逐字「made the record rainfall that fell over Houston during Hurricane Harvey roughly three times more likely and 15 percent more intense」[一手逐字]；另一篇 Emanuel 2017（PNAS 114:12681）逐字「then in 2017 the annual probability would be 6%, a sixfold increase since the late 20th century」[一手逐字]——「3 倍」是 WWA 的数字、「6 倍」是 Emanuel 的数字，媒体转述把两者混为一谈。媒体侧：Guardian 标题（Wayback 原件）「Global warming made Hurricane Harvey deadly rains three times more likely」；AP「Studies: Warming made Harvey’s deluge 3 times more likely」（同一报道正文里有科学家澄清「man-made climate change didn’t trigger Harvey」——澄清句没进标题）；PIK/Rahmstorf 2017-08-28 反例逐字「climate change is not the cause of tropical storm Harvey… Asking for a single cause of a specific weather event usually doesn’t make sense」[多源检索]。

案例二：2021 太平洋西北热浪。 学术原句（WWA）逐字「as rare as 1 in a 1000 years would have been at least 150 times rarer without human-induced climate change」[一手逐字]——方向是「无气候变化时更稀有 150 倍」；BBC 标题「US-Canada heatwave ‘virtually impossible’ without warming」（正文保留 Otto 原话「at least 150 times rarer in the past」——保真样本）；AP则译成「climate change increased chances of the extreme heat occurring by at least 150 times」（数学等价但语义方向翻转：从「反事实稀有度」变成「正因果倍数」）；MIT Tech Review「Yes, blame climate change」「made the unprecedented weather event 150 times more likely」。同一个 150，三种方向。

案例三：2022 巴基斯坦洪水。 学术原句（WWA）逐字「the 5-day maximum rainfall over the provinces Sindh and Balochistan is now about 75% more intense」「the 60-day rain across the basin is now about 50% more intense」+ 自限逐字「This discrepancy suggests that long-term variability, or processes that our evaluation may not capture, can play an important role, rendering it infeasible to quantify the overall role of human-induced climate change」[一手逐字]；Guardian 标题「Pakistan floods ‘made up to 50% worse by global heating’」——「强度 +50%」变成「损害加重 50%」（worse），且 WWA 同篇的「无法量化总体作用」未进标题；BBC同事件报道则保留了「the team were unable to quantify the scale of the impact」[一手逐字]——同一事件、两家媒体、一保真一去限定。

5.4 批评侧的语言账

Hulme 2014（Progress in Physical Geography，作者提交稿）逐字：「the meaning of climate change in popular western discourse has changed from being a descriptive index of a change in climate… to becoming an independent causative agent (as in ‘climate change caused this event to happen’)… climate change has been granted power to change worlds」[一手逐字]；并把公众问句命名为「’extreme weather blame’ question」[一手逐字]。Olsson 2022（Earth’s Future，Crossref 摘要级）逐字「probabilistic event attribution is particularly sensitive to these predicaments, rendering the choice of method value laden and hence political」[摘要逐字]。Russill 2023（Social Media + Society，开放获取全文）逐字「catastrophe is now attributed to climate change publicly, without the earlier expectation to wait on peer review or scientific consensus before acting」[一手逐字]。

而最锋利的内部证据是 WWA 自己的记者指南（Reporting extreme weather and climate change: A guide for journalists，web_extract 全文逐字）：「There are three common mistakes made by news organisations… ignoring climate change as a cause… attributing the event to climate change without providing any evidence for that claim, and attributing an extreme weather event to climate change as the sole cause」「climate change cannot cause an event (in a binary use of the term ’cause’) because all weather events have multiple causes」「This happens partly because the question of whether climate change caused an event, while seemingly reasonable, is poorly posed」[一手逐字]——造尺人给自己的消费者写的使用说明书，第一条就是禁止做本篇要审的那件事。

六、误读账：三组并排之后的结构观察

把 5.3 的三组并排放在一起，可以抽出四个可复现的结构规律：

  1. 限定词按固定顺序脱落：区间（1.5–5）→ 下限（at least 3）→ 定值（3）→ 因果动词（made/blamed）。Guardian 的 Harvey 标题保留了「made…more likely」但删了「roughly」与区间；AP 的 PNW 报道把「rarer without」翻成「more likely」——方向翻转比删除更隐蔽。
  2. 媒体有保真与去限定的稳定分工：BBC 两例（PNW、巴基斯坦）都保留了造尺人的限定句（「rarer in the past」「unable to quantify」），Guardian/AP/MIT-TR 两例都去掉了。这不是媒体质量的单变量问题，而是标题体裁的固定成本——标题没有放限定词的位置。
  3. 「worse」是比「more intense」更远的一步：巴基斯坦案例里「50% more intense」（强度差）→「50% worse」（损害差）是两步跳：先跳过「事件定义绑定」（+50% 只属于 5 日定义），再跳过「无法量化总体作用」的自限。WWA 原文逐字「Some of these models suggest climate change could have increased the rainfall intensity up to 50% for the 5-day event definition」[一手逐字]——「up to」也被 Guardian 删了。
  4. 诉讼文本是限定的终极脱落点（详见十二章）：Multnomah 县诉状逐字「150 times less likely」「one time in 1000 years. Escalating carbon pollution has increased the likelihood that it will reoccur every 5 to 10 years」「The heat catastrophe was caused by carbon pollution… In sum, but for carbon atmospheric pollution, the 2021 Pacific Northwest heat dome would have not occurred」[一手逐字]——WWA 页面上那行「All numbers below assume that the heatwave was a very low probability event that was not caused by new nonlinearities」[一手逐字]没有出现在诉状里。

七、速度跳·制度账：两周内的科学结论

7.1 组织史

WWA About 页逐字「In 2014 Dr. Friederike Otto, Dr. Geert Jan van Oldenborgh and the non-profit Climate Central, serving as secretariat as well as helping to secure initial funding, founded World Weather Attribution (WWA)」[一手逐字]；「The results are made public as soon as they are available, often days or weeks after the event」[一手逐字]。注意口径矛盾：官网（2014）与创始人 Carbon Brief 客座文（2015）自述成立年份不同；by-year 归档首条为 2014-12-11、首项事件研究为 2015 夏欧洲热浪。本篇采用「2014 年组建、2015 年首发研究」的表述，并把矛盾登记为需亲核。

7.2 发布滞后实测（主笔自算，可复现）

从 WWA by-year 归档（145 条，2014–2026）取 6 个「事件截止日 → 官方发布日」样本：

事件 事件截止 发布日 滞后（天）
Harvey 2017 2017-08-28 2017-08-31 3
地中海热浪 2023-04 2023-04-28 2023-05-05 7
PNW 热浪 2021-06 2021-06-29 2021-07-07 8
英国 40°C 2022-07 2022-07-19 2022-07-28 9
巴基斯坦洪水 2022 2022-08-25 2022-09-14 20
西伯利亚热浪 2020 2020-06-20 2020-07-15 25

滞后 3–25 天，中位 9 天。「两周内发布」是外界对 WWA 的概括口径，WWA 自己的原句是「often days or weeks after the event」（About/FAQ）[一手逐字]；最快个案是 Yale E360 记录的 2015 年 Desmond 风暴 5 天（web_extract 取回，逐字「Otto and her colleague bashed out their report by the 10th」[一手逐字]）。

7.3 速度的代价清单（权威评估）

NASEM 2016 第 3 章（快速归因专节）逐字：「The media, the public, and decision makers increasingly ask for results from event attribution studies during or directly following an extreme event. To meet this need, some groups are developing rapid and/or operational event attribution systems to provide attribution assessments on faster timescales than the typical research mode timescale, which can often take years」[一手逐字]——「传统研究模式往往以年计」出自美国科学院；同章列出快速归因的方法学代价逐字「reliance on a primarily observationally based approach and possibly on station data that have not yet been quality controlled; inability to assess the robustness of model-based results through reliance on single models」[一手逐字]，并要求逐字「it is important to follow up on any rapid attribution with studies that are not subject to such limitations」[一手逐字]。

7.4 事后评审的制度设计（自陈）

WWA FAQ逐字「While rapid attribution studies are published before peer review in order to release the results soon after events have taken place, about a quarter of WWA studies are published in peer-reviewed journals」「The main findings of our rapid studies have never changed after being submitted, reviewed and accepted by an academic journal」[一手逐字]；Crossref 验证三篇事后论文：协议 Philip 2020（ASCMO）、巴基斯坦 2023-03-17（滞后约 6 个月）、Harvey 2017-12-01（滞后约 3 个月）；Otto 2020 回应（The Conversation，作者署名）逐字「this has also made these studies somewhat controversial, because they do not go through peer-review before publication」「rapid attribution studies use peer-reviewed methods, make data publicly available, and regularly submit to peer-review journals after the fact」[一手逐字]；E360 记录评审者反弹逐字「about half of them [reviewers] said it can’t be right because it’s so fast」[一手逐字]。

7.5 十年制度弧线：年 → 周 → 日

NASEM 2025 跟进报告 Highlights（2026-07）逐字「The number of EEA studies has grown substantially over the past decade… enabling attribution studies to be completed within days of an event」「EEA has evolved from an emerging research approach into an operational practice」并建议逐字「operational programs should conduct periodic peer review of routine, rapid EEA studies」[一手逐字]；2026 年运营化落地：Copernicus C3S2_451 服务官方页逐字「Production of factsheets within five business days of an event」「An Extreme Event Attribution Office – A distributed office across four European meteorological services」、海报逐字「Rapid attribution studies, within 5 working days for heat/cold and precipitation」「expected to produce 1-2 studies per month」「occurring anywhere in the world」[一手逐字]——速度承诺从 WWA 的「数天或数周」压缩到「5 个工作日」，且全球覆盖、每月 1–2 篇的常态化产出。同时期的批判性方法学声音：Voelker & Goessling 2026（EGU 摘要）逐字「Our results underscore the need for careful interpretation of current probabilistic extreme-event attribution, particularly when extrapolating into the future」[一手逐字]。

八、尺子账·事件定义自由度：同一事件，几种答案

8.1 官方三处自认

  • AR6 第 11 章 11.2.3 逐字「The outcome of event attribution is dependent on the definition of the event (Leach et al., 2020), as well as the framing」「Attribution statements are also dependent on the spatial and temporal extent of event definitions… large-scale averages generally yield higher attributable changes in magnitude or probability due to the smoothing out of noise」「Large-scale, longer-duration events tend to have notably larger attributable risk ratios」[一手逐字]——官方明说：尺子放大（区域更大/历时更长）→ 归因概率比系统性变大。
  • NASEM 2016 第 2 章 逐字「Statements about attribution are sensitive to the way the questions are posed… when defining an event, choices must be made about defining the duration… and the geographic area」「A robust attribution analysis would show that results are qualitatively similar across a range of event definitions… without strong sensitivity to the exact definition」[一手逐字]——稳健性判据：定性相似、定量允许不同。并举例「Black and colleagues (2015) examine the January 2014 heat events… using definitions of heat wave with durations ranging from 1 to 5 days」[一手逐字]——历时（duration）也是尺子。
  • Philip 2020 协议 Step 2 逐字「The return period of the event can be very sensitive to this choice. A definition that optimizes the location and duration on the meteorological extreme event logically results in very high values for the return period. … The choice whether to optimize the event definition for meteorological rarity or to describe the impacts should be documented.」[一手逐字]；Step 1 自陈选择偏倚逐字「This leads to a trend bias towards more extreme events in any collection of event attribution studies. This trend is exacerbated if events are selected on an expected positive outcome」[一手逐字]——文献整体偏向正结果，是协议自己写的。

8.2 太平洋西北热浪 2021：同一个事件，三个量级

  • WWA 快速分析（2021-07-07）：定义＝区域框 45–52°N, 119–123°W（逐字「based on the expected and reported human impacts rather than on the meteorological extremity」[一手逐字]）内的年最大日最高温 TXx；结论逐字「about a 1 in 1000 year event in today’s climate」「at least 150 times rarer without human-induced climate change」「about 2°C hotter than it would have been」——且页面自陈假设逐字「All numbers below assume that the heatwave was a very low probability event that was not caused by new nonlinearities」[一手逐字]。
  • 期刊版 Philip et al. 2022（ESD）逐字「this step requires some expert judgement」「this general result is not strongly tied to the exact return period」[一手逐字]；并把 1-in-1000 换算到全球尺度逐字「the entire area affected by heat is larger – about 1500 km × 1500 km… there are about 60 independent areas… This implies that the return period… somewhere over land is about 60 times smaller than the estimated 1000-year interval」[一手逐字]——同一个事件，把空间定义从「局部框」换成「全球任一地点」，答案再除 60。
  • 独立研究 McKinnon & Simpson 2022（GRL）逐字「temperature anomalies that exceed 4.5σ occur with an approximate frequency of one in a hundred thousand years」「the probabilities are shockingly low」[一手逐字]——1-in-100,000。
  • White et al. 2023（Nature Communications）综述逐字「estimates of the current return period of this event vary from 1 in 10,000–100,000 years, 1 in 1000 years, to 1 in 200 years」[一手逐字]——同一事件三档估计，横跨两个数量级，印在同一句话里；同文又逐字「It is clear, however, that anthropogenic warming of the planet contributed to the severity of this event」[一手逐字]——「稀有度」答案发散与「人为贡献方向一致」两条轴线必须分开讲。

8.3 2010 俄罗斯热浪：同一个事件，两个公开相反的结论

  • Dole et al. 2011（GRL 38, L06702，NOAA 作者副本）逐字「Analysis of forced model simulations indicates that neither human influences nor other slowly evolving ocean boundary conditions contributed substantially to the magnitude of this heat wave… We conclude that the intense 2010 Russian heat wave was mainly due to natural internal atmospheric variability」[一手逐字]。
  • Rahmstorf & Coumou 2011（PNAS，Wayback 原件）逐字「For July temperature in Moscow, we estimate that the local warming trend has increased the number of records expected in the past decade fivefold, which implies an approximate 80% probability that the 2010 July heat record would not have occurred without climate warming」「Our results thus explicitly contradict those of Dole et al.」[一手逐字]。
  • NASEM 2016 仲裁逐字「This apparent contradiction in conclusions can be reconciled by understanding that these two studies aim at answering the attribution question in two different ways」[一手逐字]——同一个事件、两个同行评审结论、官方判定「回答的是两个不同的问题」：Dole 问「平均温度异常是否超出来自趋势的解释范围」（月平均、区域平均定义），Rahmstorf & Coumou 问「破纪录事件的发生概率是否被改变」（纪录定义）。定义不同，答案相反。

8.4 2022 巴基斯坦洪水：双定义自证

WWA 分析采用两个并列定义逐字「(1) The annual maximum of the mean 60-day precipitation during June-September over the Indus river basin… (2) the annual maximum of the mean 5-day precipitation… over the worst hit provinces Sindh and Balochistan」[一手逐字]；期刊版 Zachariah et al. 2023（Environ. Res.: Climate）逐字「best estimates of a change in intensity of up to 30% for the large region and up to 50% for the small region」「For the 60 day event… determining even the sign of the change remains impossible based on this assessment」「at present, there is overall low confidence in existing attribution findings for this region」[一手逐字]——同一场洪水：60 日定义 +30% 且符号不定，5 日定义 +50%。而媒体标题取的是 50%（Guardian「made up to 50% worse」）。

8.5 方法论对照：storyline 与条件归因

Shepherd 2016（Current Climate Change Reports）逐字「the question can be framed in various ways, and the answer depends very much on the framing」「If the quantitative estimates of altered risk are sensitive to the spatiotemporal footprint of the event, as they almost certainly will be, then the quantification provided by the climate model may not be relevant at the spatiotemporal scale of the extreme weather event itself」[一手逐字]——对尺子账最锋利的方法论陈述；Shepherd et al. 2018（Climatic Change）storyline 正式定义逐字「a physically self-consistent unfolding of past events… No a priori probability of the storyline is assessed」[一手逐字]；Trenberth et al. 2015条件归因逐字「regard the circulation regime or weather event as a conditional state (whose change in likelihood is not assessed), and ask whether the impact of the particular event was affected by known changes in the climate system’s thermodynamic state」[一手逐字]——概率归因不是唯一合法框架，但它至今是媒体与诉讼消费的那个。

九、边界账：桑迪——「答案不再直接关于实际事件」

2012 年飓风桑迪是「未得出结论」的教科书：不是数据缺失，是方法学上的定义与条件化选择。

  • NASEM 2016 第 2 章逐字「event attribution is usually a matter of changing probabilities rather than a deterministic yes or no. For example, a scientific researcher might repose the question ‘Was Hurricane Sandy caused by climate change?’ as ‘How much did human influence on climate increase the odds of a tropical or post-tropical storm with winds greater than 65 knots making landfall in northern New Jersey?’」「probabilities are usually estimated using a definition of an event that differs from the specific event… As a result, the answers obtained are no longer directly about the actual event」[一手逐字]——确定性提问必须改写成概率化事件类定义，且答案不再直接关于实际事件。这是「概率跳」的官方镜像句：不是媒体跳了，是方法本身要求跳。
  • NASEM 2016 第 3 章逐字「For extremely rare meteorological events (e.g., Hurricane Sandy), the combination of rarity and spatial scale makes an unconditional attribution approach challenging from a modeling perspective」「this conditioning needs to be clearly communicated because it strongly constrains the types of questions that the attribution study can address」[一手逐字]。
  • NASEM 2016 第 4 章逐字「definitive attribution of a specific event in a deterministic manner is generally not possible… most events could have happened in the counterfactual world (so the probability of necessary causation is less than 100%)」[一手逐字]——NASEM 对「确定性归因」的总判决。

十、制度账：损失与损害基金——归因结论的官方消费者

归因科学进入国际制度的第一站是「损失与损害」（loss and damage）：

  • 巴黎协定第 8 条（2015-12-12）逐字「Parties recognize the importance of averting, minimizing and addressing loss and damage associated with the adverse effects of climate change, including extreme weather events and slow onset events」[一手逐字]——「极端天气事件」进入条约正文；但第 8 条通篇无「attribution」字样。
  • COP27 决议 2/CP.27（2022-11-20）第 3 段逐字「decide… to establish a fund for responding to loss and damage whose mandate includes a focus on addressing loss and damage」[一手逐字]；序言把紧迫性建立在 AR6 之上逐字「as illustrated by the findings in relevant recent scientific reports, including the contributions of Working Groups I and II to the Sixth Assessment Report」[一手逐字]——制度文本以科学评估为启动依据，但基金条款本身不写归因方法。
  • COP28 决议 1/CP.28（2023-12-11）逐字「Approve the Governing Instrument of the Fund」+ 世界银行托管逐字「to operationalize the Fund as a World Bank hosted financial intermediary fund for an interim period of four years」+ 首轮承诺逐字「amounting to the equivalent of USD 792 million」[一手逐字]；免责条款逐字「funding arrangements… are based on cooperation and facilitation and do not involve liability or compensation」[一手逐字]——基金是合作机制，不是赔偿机制。
  • COP29 决议 5/CP.29（2024-11-22）逐字「amount to the equivalent of USD 731.15 million」+ 首任执行主任逐字「Ibrahima Cheikh Diong」[一手逐字]（注意两个金额口径不同：792M 是含基金外安排的供资总额、731.15M 是给基金本身的承诺，不可相加）。
  • COP30（2025-11 贝伦）：FRLD 官方新闻稿逐字「the Fund for responding to Loss and Damage (FRLD) has launched its first call for funding requests for the Barbados Implementation Modalities (BIM)… The BIM has been allocated an initial USD 250 million」[一手逐字]——首个资金窗口开放；但截至 2026 年 7 月第 9 次董事会（马尼拉），首批 4 个申请（合计 7,740 万美元）被推迟至 2026 年 12 月审批，基金尚未产生首笔放款（FRLD 官方发布页，[多源检索]）——「制度链已通、钱还没放」。

十一、消费账：诉讼、再保险、媒体与公众

11.1 诉讼：概率句被翻译成侵权语言

  • Saúl Luciano Lliuya v. RWE（OLG Hamm 5 U 15/17，2025-05-28 终局裁定，判决书全文取回）逐字（德文）「Die Klage ist zwar mit ihren zuletzt gestellten Hauptanträgen zulässig und dem Grunde nach schlüssig, aber unbegründet」[一手逐字]——可受理、事实层面表面成立（schlüssig），但因风险评估不成立：法院指定专家给出 30 年内洪水概率 1%，未进入因果关系审查（RWE 官方新闻稿逐字「the court-appointed experts put the probability of flood risk… at ‘1 per cent over the next 30 years’. From the court’s point of view, there is therefore no reason to investigate the existence of a causal link」[一手逐字]）——归因链（排放份额 → 升温 → 冰川融化 → 洪水威胁）未被实体否定；France 24 转述法院原则逐字「companies ‘may be obligated to take preventive measures’ to counter their emissions」[文献较稳]。
  • Multnomah County v. ExxonMobil 等（俄勒冈州巡回法院，2023-06-22 诉状 203 页全文取回）逐字「The heat dome that cost so much life and loss was not a natural weather event… Rather, the heat dome was a direct and foreseeable consequence of the Defendants’ decision to sell as many fossil fuel products over the last six decades as they could」+ 引 Philip et al. 2022 逐字「the occurrence of a heat wave of the intensity experienced in the study area would have been virtually impossible without anthropogenic climate change」+「150 times less likely to have occurred in the absence of ACC」「one time in 1000 years. Escalating carbon pollution has increased the likelihood that it will reoccur every 5 to 10 years」「In sum, but for carbon atmospheric pollution, the 2021 Pacific Northwest heat dome would have not occurred」[一手逐字]——概率句被逐字搬进诉状并翻译成「but for」因果，而 WWA 原文的假设限定（「not caused by new nonlinearities」）未出现。
  • ICJ 气候变化咨询意见（2025-07-23，官方摘要逐字）「Extreme weather events, such as hurricanes, droughts and heatwaves, are becoming more frequent and intense」+「Article 8 of the Paris Agreement… does not, however, address issues of liability or compensation… Accordingly, the Court does not find in Article 8 of the Paris Agreement any clearly expressed lex specialis that would exclude the application of the general rules on State responsibility」[一手逐字]；并在因果关系上区分两段（事件能否归因于气候变化＝科学问题；损害能否归因于特定国家＝法律问题，判决正文 para. 437，[需亲核] 经二手转述）——国际法院把「科学归因」与「法律归因」切开，这正是本篇的命题在制度层的回声。

11.2 再保险：两个口径

  • Swiss Re sigma 1/2025逐字「observed climate change effects are also playing a role in compounding losses for some perils and regions」+ 新闻稿逐字「Much of this increasing loss burden results from value concentration in urban areas, economic growth, and increasing rebuilding costs」[一手逐字]——气候变化是「叠加因子」，主驱动是社会暴露。
  • Munich Re 2024 年度回顾（标题即「Climate change is showing its claws」）逐字「The impact of man-made climate change on weather disasters has been proven many times over by research」「This was the case for Helene and Milton, where World Weather Attribution studies have shown that both hurricanes were significantly more severe and brought much more extreme rainfall than in a hypothetical world without climate change」[一手逐字]——直接点名引用 WWA。两家再保险巨头，一个用「compounding」、一个用「proven」——归因产品的产业消费也有口径差。

11.3 公众认知

Yale CCAM Fall 2024（n=1,013，官方报告，[需亲核] 经搜索快照双源核对）逐字「66% of Americans think global warming is affecting weather in the United States」；极端高温 77%、野火 74%、干旱 73%、洪水 72%、飓风 71%；「56% of Americans think extreme weather poses either a ‘high’ (16%) or ‘moderate’ (40%) risk to their community」[多源检索]；Yale 归因专项地图自评逐字「Public beliefs lag scientific assessments of the effects of global warming on extreme heat and rising sea levels」「Judgments about extreme weather appear better explained by pre-existing ideologies and beliefs about climate change than by the attribution science」[一手逐字]——公众的「归因」判断受意识形态影响大于受归因科学影响。

十二、反向红跳（对称审计）

反方一：「归因全是模型游戏」——不立。 观测先于模型（Philip 2020 Step 3「The first task is to analyse observational data」[一手逐字]）；模型阈值锚定观测（Step 4「setting the return period in the current conditions’ run equal to the value from observational data」[一手逐字]）；观测记录本身是独立证据（2021 PNW 多处破纪录 5°C 以上、加拿大全国纪录 49.6°C 破 4.6°C，White 2023 与 WWA 双源逐字[一手逐字]）；模型-观测一致性是 IPCC 评估对象（AR6 第 11 章逐字「Climate models can reproduce the sign (direction) of changes in temperature extremes observed globally and in most regions, although the magnitude of the trends may differ (high confidence)」[一手逐字]）。

反方二：「归因不可证伪」——不立。 归因研究自带可检验预言且被回验（Christidis 2015 回验 Stott 2004 的 2040s 预言，见守真锚 5）；事后独立检验存在（McKinnon & Simpson 2022 用独立方法给出不同数量级并公开）；Philip 2020 逐字「A trend may not yet have emerged from the noise in the observations even though the probability of the event has changed. So the absence of a significant trend would not lead by itself to a non-attribution statement」[一手逐字]——不可证伪的指控与方法文本相反。

反方三：「每场极端天气都是气候变化造成的」——不立。 桑迪无人给出确定性归因（第九章）；2010 俄罗斯热浪权威结论是「主要自然变率」（Dole 2011 逐字，见 8.3）；WWA 自己的寒潮分析（North America cold Dec 2017–Jan 2018）逐字「Cold outbreaks like this are getting warmer (less frequent) due to global warming… Temperatures like these are now about fifteen times rarer」[一手逐字]——概率降低的事件同样被归因；2021 年 2 月德州寒潮（WWA 无此分析，[多源检索] 阴性结果）由独立研究 Doss-Gollin et al. 2021（[摘要逐字]）给出逐字「local temperatures and the inferred demand for heating per capita… were more severe during a storm in December 1989 than during February 2021」[摘要逐字]——观测记录里 1989 年那场更冷。

反方四：「归因全是炒作」——不立。 信心分级是公开制度（SPM 脚注 4逐字「virtually certain 99–100% probability; very likely 90–100%; likely 66–100%…」+ 信心五档「very low, low, medium, high and very high」[一手逐字]）；事件类型信心分布是公开评估（AR6 第 11 章逐字「confidence in attribution statements for large-scale heat and lengthy extreme precipitation events have higher confidence than shorter and more localized events」「Severe convective storms (tornadoes, hail, rainfall, wind, lightning) Low confidence」[一手逐字]；NASEM 2016逐字「Extreme heat and cold events are the simplest events on which to perform attribution… Tornadoes… no studies have been performed」[一手逐字]——「没有归因到」≠「归因不可能」，是方法论空位）；WWA 自我设限句成建制（PNW 页「we cannot say whether this was a so-called ‘freak’ event」「we only give a lower bound of the estimate」[一手逐字]）；早期研究缺模型验证的批评被官方评估文本纳入并回应（AR6 第 11 章逐字「The lack of model evaluation, in particular in early event attribution studies, has led to criticism of the emerging field of attribution science as a whole」[一手逐字]）——连批评史都写进了 IPCC。

十三、母裁决

概率式事件归因是真的科学：观测在第一步，模型阈值锚定观测，结论自带置信度分级，预言会被十年后的观测回验——这些都不是炒作。 而「这场灾难是气候变化造成的」不是它的产品：它的产品是「这类事件的可能性被改变了多少倍、强度被改变了多少度、在什么定义下、带多大区间、以多快的速度发布、在什么假设下成立」。五份文件共同画出的边界是：概率句是合法产品，因果句是读法问题；「caused by」是造尺人自己明文禁止的用法，而它恰恰是传播链末端唯一活下来的句法。 同一天、同一个事件，WWA 页面写「at least 150 times rarer without」、诉状写「but for」、标题写「blame」——三句话之间的差价，就是这篇报告测的东西。

灵魂句：这场灾难从来没有被归因于气候变化——被归因的是一整类像它的灾难；而那句完成时的因果句，是读者自己写的，用的却是造尺人亲手递出的数字。 数字是真的，限定也是真的；只是它们没有一起抵达终点。

附：不确定点与待核对

  1. Stott 2004 正文：镜像 PDF 为双栏排版，pdftotext 文本层左右栏交错；摘要关键句（ill-posed／at least doubled）已从 nature.com 官方页核对一致，正文其余逐字引用以官方页摘要与镜像重组为准，标 [一手逐字] 的句均已两路对读。
  2. WWA 成立年份：官网 About/FAQ 与 Russill 2023 作 2014，创始人 Carbon Brief 客座文作 2015。本篇采用「2014 年组建、2015 年首发研究」，未裁决哪个对。
  3. 「两周内发布」：是外界概括口径，非 WWA 原句；WWA 原句为「often days or weeks after the event」。实测 3–25 天、中位 9 天（6 样本）。
  4. 「1-in-1000 年」的假设链：WWA/Philip 的 150 倍与 +2°C 均建立在「当前气候 1-in-1000 年」假设上（页面与论文自陈），McKinnon & Simpson 独立估计为 1-in-100,000；White 2023 三档并陈。引用任何一个数字都必须带这条链。
  5. Rahmstorf & Coumou 的「3 倍」：任务书初拟的「概率增加约 3 倍」在原文中是理论推演句（分布平移 1σ 时热+冷极端总数增 3 倍以上）；莫斯科事件的数字是 fivefold/80%。本篇采用后者。
  6. Dole 2011 摘录：来自 NOAA 作者副本全文（200）与 Wiley 摘要双路；结论句两路一致。
  7. ICJ 判决正文 para. 427/436–438：目前为二手转述（哥伦比亚气候法博客逐字引用），官方摘要已取回（200），全文意见本体 403 未取回；正文引用仅用官方摘要逐字句，para. 编号句标 [需亲核]。
  8. Yale CCAM Fall 2024：官方 PDF 未单独下载，承重句经官方页与 GMU 镜像两处搜索快照交叉核对；标 [多源检索]。
  9. Doss-Gollin 2021：IOP 被 Radware 反爬拦截，承重句经 Semantic Scholar API 官方摘要取得，标 [摘要逐字]。
  10. Christidis 2015：nature.com 页面摘要级取回（200），正文未取回；仅引用摘要内逐字句。
  11. Multnomah 诉状：203 页全文取回（200），引用句均已逐字核对；被告答辩状未取回。
  12. WWA 无德州 2021 寒潮分析：by-year 全归档 + 站内检索均无（方法学阴性，[多源检索]），该事件归因的承重源改用独立研究。
  13. FRLD 首笔放款：截至 2026-07 第 9 次董事会（马尼拉）首批 4 申请推迟至 2026-12 审批——「基金尚未放款」按官方发布页登记，未来状态会变。
  14. 两个基金金额：792M（COP28 供资安排总额，含基金外）与 731.15M（COP29 基金承诺）口径不同，本篇未相加。

来源清单

（编号来源，按章序；全部链接在正文中已给出。抓取件与逐字核对脚本存 tmp/extreme-weather-attribution/，可复现。）

  1. IPCC AR6 WG1 第 11 章（PDF 17.1MB，curl 200）— https://www.ipcc.ch/report/ar6/wg1/downloads/report/IPCC_AR6_WGI_Chapter11.pdf
  2. IPCC AR6 WG1 决策者摘要 SPM（PDF，curl 200）— https://www.ipcc.ch/report/ar6/wg1/downloads/report/IPCC_AR6_WGI_SPM.pdf
  3. NASEM 2016《Attribution of Extreme Weather Events in the Context of Climate Change》（分章 1–9 全取，–compressed 200）— https://nap.nationalacademies.org/catalog/21852/
  4. Philip et al. 2020 协议（ASCMO 6:177–203，PDF 3.9MB curl 200）— https://doi.org/10.5194/ascmo-6-177-2020
  5. Allen 2003（Nature 421:891–892，官方页 + climateprediction.net 作者副本 Wayback）— https://www.nature.com/articles/421891a
  6. Stott, Stone & Allen 2004（Nature 432:610–614，官方页 + LMD 镜像）— https://www.nature.com/articles/nature03089
  7. Trenberth 2012（Climatic Change 115:283–290，CC-BY 开放 PDF）— https://doi.org/10.1007/s10584-012-0441-5
  8. Trenberth, Fasullo & Shepherd 2015（Nature Climate Change 5:725–730，CentAUR 开放版）— https://doi.org/10.1038/nclimate2657
  9. Hulme 2014（Progress in Physical Geography 38:499–530，作者提交稿）— https://www.mikehulme.org/wp-content/uploads/2014/03/14_03-PiPG-Weather-attribution_submitted.pdf
  10. Olsson et al. 2022（Earth’s Future 10:e2021EF002258，Crossref 摘要级）— https://doi.org/10.1029/2021EF002258
  11. Russill 2023（Social Media + Society，开放获取全文）— https://doi.org/10.1177/20563051231177897
  12. Shepherd 2016（Current Climate Change Reports 2:28–38，CentAUR 开放版）— https://doi.org/10.1007/s40641-016-0033-y
  13. Shepherd et al. 2018（Climatic Change 151:555–571，LSE 开放版）— https://doi.org/10.1007/s10584-018-2317-9
  14. Christidis, Jones & Stott 2015（Nature Climate Change 5:1003–1005）— https://www.nature.com/articles/nclimate2468
  15. Dole et al. 2011（GRL 38:L06702，NOAA 作者副本）— https://doi.org/10.1029/2010GL046582
  16. Rahmstorf & Coumou 2011（PNAS 108:17905–17909，Wayback 原件）— https://doi.org/10.1073/pnas.1101766108
  17. McKinnon & Simpson 2022（GRL 49:e2022GL100380，Wayback/作者副本）— https://doi.org/10.1029/2022GL100380
  18. White et al. 2023（Nature Communications 14:3601）— https://www.nature.com/articles/s41467-023-36289-3
  19. Philip et al. 2022（ESD 13:1689–1713，Copernicus 直取）— https://doi.org/10.5194/esd-13-1689-2022
  20. Zachariah et al. 2023（Environmental Research: Climate 2:025001）— https://doi.org/10.1088/2752-5295/acbfd5
  21. Emanuel 2017（PNAS 114:12681–12686，PMC 全文）— https://pmc.ncbi.nlm.nih.gov/articles/PMC5715789/
  22. Doss-Gollin et al. 2021（ERL 16:064046，Semantic Scholar 摘要级）— https://doi.org/10.1088/1748-9326/ac0278
  23. WWA 官网全套：/about/、/methods/、/faqs/、/by-year/、/analyses/、记者指南、2015–2026 各分析页（curl 200，仅首页偶发 000）— https://www.worldweatherattribution.org/
  24. NASEM 2025 跟进报告 Highlights（2026-07）— https://nap.nationalacademies.org/resource/28590/Report_Highlights_Attribution_Extreme_Weather.pdf
  25. Copernicus C3S2_451 运营化归因服务官方页与海报 — https://climate.copernicus.eu/operational-extreme-event-monitoring-and-attribution-service
  26. Voelker & Goessling 2026（EGU26-19199 摘要）— https://meetingorganizer.copernicus.org/EGU26/EGU26-19199.html
  27. 巴黎协定（UNFCCC 官方全文）— https://unfccc.int/sites/default/files/english_paris_agreement.pdf
  28. COP27 决议 2/CP.27（FCCC/CP/2022/10/Add.1，_adv 官方变体）— https://unfccc.int/sites/default/files/resource/cp2022_10a01_adv.pdf
  29. COP28 决议 1/CP.28（FCCC/CP/2023/11/Add.1）— https://unfccc.int/documents/637067
  30. COP29 决议 5/CP.29（InforMEA 官方镜像）— https://www.informea.org/sites/default/files/decisions/2025-11/cp2024_11a01E_0.pdf
  31. FRLD 官方新闻稿与发布页 — https://www.frld.org/press-releases
  32. OLG Hamm 判决书 5 U 15/17（2025-05-28，北威州司法门户 PDF）— https://nrwe.justiz.nrw.de/olgs/hamm/j2025/5_U_15_17_Urteil_20250528.html
  33. Multnomah County v. ExxonMobil 诉状（203 页）— https://multco-web7-psh-files-usw2.s3-us-west-2.amazonaws.com/s3fs-public/Complaint_0.pdf
  34. ICJ 气候变化咨询意见官方摘要（2025-07-23）— https://icj-web.leman.un-icc.cloud/sites/default/files/case-related/187/187-20250723-sum-01-00-en.pdf
  35. Swiss Re sigma 1/2025 报告与新闻稿 — https://www.swissre.com/institute/research/sigma-research/sigma-2025-01-natural-catastrophes-trend
  36. Munich Re 2024 年度自然灾害回顾 — https://www.munichre.com/en/company/media-relations/media-information-and-corporate-news/media-information/2025/natural-disaster-figures-2024.html
  37. BBC 2021-07-08 PNW 热浪报道 — https://www.bbc.com/news/science-environment-57751918
  38. Guardian 2022-09-15 巴基斯坦洪水报道 — https://www.theguardian.com/environment/2022/sep/15/pakistan-floods-made-up-to-50-worse-by-global-heating
  39. Reuters 2021-07-08 与 2022-09-15 报道 — https://www.reuters.com/business/environment/heat-wave-pacific-northwest-could-soon-repeat-due-climate-change-research-2021-07-07/
  40. Yale CCAM Fall 2024 与归因专项地图 — https://climatecommunication.yale.edu/app/uploads/2025/02/climate-change-american-mind-beliefs-attitudes-fall-2024.pdf
  41. Otto 2020 The Conversation 回应 — https://theconversation.com/yes-climate-change-can-affect-extreme-weather-but-there-is-still-a-lot-to-learn-136003
  42. Yale E360 快速归因特写 — https://e360.yale.edu/features/pinning-wild-weather-on-climate-change-scientists-are-upping-their-game
  43. Carbon Brief 客座文（WWA 五年复盘）— https://www.carbonbrief.org/guest-post-lessons-learned-from-five-years-of-extreme-weather-rapid-attribution/
  44. WWA 巴基斯坦期刊版（Zachariah 2023，同 20）与 Harvey 期刊版 van Oldenborgh 2017（ERL 12:124008）— https://doi.org/10.1088/1748-9326/aa9ef2
标签: