干支「太过不及」的事件类型特异性验证:火太过年偏热不偏涝的混淆矩阵检验

Written by

in

Language

Evidence level: B

五运六气最可检验的承诺不只是”火太过之年有异常天气”,而是”火太过之年偏热而非偏涝、水太过之年偏寒而非偏旱”。用一个 5 类事件 × 5 类干支的混淆矩阵,就能把”笼统相关”和”类型特异”分开——前者可能是巧合,后者才是理论真正的指纹。

正文

《素问·气交变大论》那句”岁火太过,炎暑流行;岁水太过,寒气流行”,最容易被人偷换成一种低要求的说法:”火太过之年天气有点异常”。但原文其实给出的是类型特异的承诺:火太过对应的是”热/炎暑”,不是”涝”,也不是”风”。这个区别至关重要——因为”某年异常”可能是运气、也可能是任何别的强迫;只有”火太过年偏偏热、水太过年偏偏寒、且不跨类溢出”才是指向干支周期的独特指纹。要验证这枚指纹,需要的不只是相关性,而是一张事件类型 × 干支类型的混淆矩阵。

为什么笼统相关性不够。 假设火太过之年的极端事件确实多了一些,但热、寒、旱、涝、风五类混在一起算总数,只能证明”火太过年更热闹”,无法区分”岁火太过真的带来热”还是”火太过年恰好是多事之秋”。气候学里这种事并不罕见:中国东部千年旱涝等级网格数据集显示,极端旱与极端涝在特定世纪(如 1551–1600)会同时高频出现,干旱事件有自身的百年周期(PMID 37041161)。如果火太过年份恰好落在某个干旱多发的百年段,笼统的”极端事件增多”检验就会被旱灾的自身周期假性抬高——这就是必须做类型特异性检验的根本原因。

怎么建混淆矩阵。 标准做法分四步。第一步,把 500 年级的旱涝/温度等级序列转成逐年主导事件类型:取每年的极端等级,标成”热/寒/旱/涝”四类(风另立或并入异常场)。中国东部已有的千年网格化旱涝等级数据集提供了 2.5°×2.5° 的空间框架,文献旱涝等级与树轮代用记录融合、角距离权重插值并附采样误差估计(PMID 37041161)——它是”文献事件类型化”的现成底稿。第二步,把每一年归入干支”太过/不及”类:按天干配五行、阳干太过阴干不及,给每年贴上”火太过/水太过/木太过/金太过/土太过/火不及/……”的标签。第三步,对每一类”太过/不及”年统计五类极端事件的实际分布,得到 5(干支类型)× 5(事件类型)的混淆矩阵。第四步,用卡方检验看”类型 × 事件”关联是否显著,并用 FDR 或置换检验做多重比较校正——60 个干支 × 5 类事件有几百个格子,纯随机也能冒出十几个假”显著”。

“对角线显著”才是理论赢。 混淆矩阵最有价值的地方是它能回答特异性问题:火太过年的”热”检出率(敏感性)高不高?同时,火太过年的”涝”溢出率(假阳性)高不高?真正的运气规律应该表现为主对角线显著——火太过年集中在”热”格子、水太过年集中在”寒”格子——而不是某一行整体偏高。历史气候学已经示范过这种”事件类型化”的可行性:研究者把秦岭大鱼洞洞穴内的古代碑文记载(1520–1920 年间 7 次旱灾的准确时间与影响)与同一洞穴的高分辨率石笋氧同位素记录直接对齐,结果相互印证,证明了文字记载的干旱事件与代用记录同步吻合(PMID 26270656)。这提示:文献编码的事件类型与独立代用记录可以互相印证,为干支类型检验提供了交叉验证的模板。季风信号本身在温度与降水序列中均有明确反映——对长江中下游 300 年重建序列的分析发现,梅雨降水与冬季温度各自携带显著的东亚季风信号,且存在 2–7 年、30–40 年、50–100 年的多尺度周期(PMID 26107375)。也就是说,历史上确实存在”气候状态对应旱涝格局”的稳健信号——问题在于它是否被 60 年干支周期锁定,而不是被这些更长的季风/气候周期锁定。

怎么排除”撞周期”。 干支周期是 60 年准周期,而中国气候本身存在 50–70 年、100–120 年的显著周期(东亚季风的多尺度变率,PMID 37359908)。这两套周期在百年尺度上会部分重叠,一个不小心,检验到的就是气候自身的振荡节律而非干支。对策是两步:一是置换检验——随机打乱”年份—干支”对应关系数千次构建零分布,看真实观测的混淆矩阵对角线强度是否显著超过随机排列;二是排除百年强迫段——把中世纪暖期/小冰期等百年气候段作为协变量或分层变量,检验类型特异性在”控掉大周期”之后是否依然存在。同时,五运六气的临床对照研究也提示:理论确实携带可检测的信号,但效应量不大、需大样本——一项基于五运六气理论的北京地区 15 年回顾研究发现,结核发病率与当年及既往 1–3 年的气象因子显著相关、风速贡献最强(PMID 26525546)。这从疾病一侧说明:干支/运气与环境的关联”存在但弱”,类型特异性检验恰好是分辨”真信号”与”大周期假象”的试金石。

一句话总结:把”火太过年有没有异常”升级成”火太过年偏热、且不偏涝”的混淆矩阵检验,是把五运六气从”言之有物的玄学”推向”可证伪的气候学假设”的关键一步——对角线显著而溢出受控,干支周期就拿到真正的特异性证据;只有笼统相关、无对角线,那可能只是气候大周期的镜像。

证据与局限

已验证: 中国东部千年网格化旱涝等级数据集(文献+树轮融合、附采样误差)已建成(PMID 37041161);秦岭洞穴碑文记载的旱灾事件与石笋氧同位素记录同步吻合(PMID 26270656);长江中下游 300 年重建序列显示温度与梅雨降水均携带显著季风信号、存在多尺度周期(PMID 26107375);东亚季风存在 50–70 年、100–120 年多尺度周期(PMID 37359908);基于五运六气的北京结核发病率-气象回顾研究检出显著但弱的关联(PMID 26525546)。

部分验证: “历史文献可被编码为事件类型序列并与独立代用记录交叉验证”已被秦岭洞穴案例示范(PMID 26270656),但该案例只覆盖 7 次干旱事件、样本极小;千年旱涝数据集覆盖 960 年以来东部,西部地区(尤其西北)代用记录稀疏,类型匹配的空间完整性不足(PMID 37041161)。

待验证: ①全球尚无研究把”5 类事件类型化 + 干支太过不及归年 + 混淆矩阵 + FDR/置换检验”做成完整的 500 年级特异性检验;②”火太过年偏热不偏涝”的主对角线特异性是否在控掉 50–70 年/百年大周期后依然稳健,尚无直接数据;③文献事件编码的空间不均匀(东部密西部疏)与漏记偏倚如何进入混淆矩阵的格子计数,尚无可复现的统计方案。

相关阅读

证据等级: B
参考文献: PMID 37041161, PMID 26270656, PMID 26107375, PMID 37359908, PMID 26525546


免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

Event Type Specificity Verification of \”Excess and Deficiency\” in Heavenly Stems and Earthly Branches: Confusion Matrix Analysis of Fire Excess Showing Heat Without Flooding

Evidence level: B

The most verifiable promise of Five Movements and Six Qi is not merely “abnormal weather in years of Fire Excess,” but rather “years of Fire Excess exhibit heat rather than flooding, and years of Water Excess exhibit cold rather than drought.” A confusion matrix mapping 5 event types against 5 Gan-Zhi categories can distinguish between “general correlation” and “type specificity”—the former may be coincidental, while the latter constitutes the true theoretical fingerprint.

Main Text

The statement in Su Wen · Qi Jiao Da Bian Lun (“In years of excessive Fire, intense heat prevails; in years of excessive Water, intense cold prevails”) is often loosely rephrased with a low standard: “abnormal weather in years of excessive Fire.” However, the original text provides a commitment to type specificity: Excessive Fire corresponds to “heat/intense heat,” not “flooding,” nor “wind.” This distinction is crucial—because “abnormal weather in a certain year” could be due to luck or any other forcing factor; only the pattern where “Fire Excess years specifically show heat, Water Excess years specifically show cold, without cross-category spillover” constitutes a unique fingerprint pointing to the Gan-Zhi cycle. Validating this fingerprint requires more than simple correlation; it demands a confusion matrix of event types × Gan-Zhi types.

Why general correlation is insufficient. Assume that extreme events in years of Fire Excess are indeed more frequent, but if heat, cold, drought, flooding, and wind are aggregated into a single total count, it can only prove “years of Fire Excess are busier,” failing to distinguish whether “Fire Excess truly brings heat” or if “years of Fire Excess simply happen to be turbulent.” Such phenomena are not uncommon in climatology: the China East millennium drought-flood grade grid dataset shows that extreme drought and extreme flooding can co-occur with high frequency in specific centuries (e.g., 1551–1600), with drought events possessing their own centennial cycles (PMID 37041161). If a year of Fire Excess coincides with a century-long drought-prone phase, a general test for “increased extreme events” could be falsely elevated by the drought cycle itself—this is the fundamental reason why type-specific testing is necessary.

Constructing the confusion matrix. The standard procedure involves four steps. First, convert the 500-year drought/flood and temperature grade sequences into annual dominant event types: assign each year’s extreme grade to one of four categories (“heat/cold/drought/flood”), treating wind separately or merging it into an anomaly field. Existing millennium grid datasets for China East provide a 2.5°×2.5° spatial framework; literature integrates drought-flood grades with tree-ring proxies, applies angular distance weighting interpolation, and includes sampling error estimates (PMID 37041161)—serving as a ready draft for “literature event typification.” Second, classify each year into Gan-Zhi “Excess/Deficiency” categories: match Heavenly Stems to Five Phases, assigning “Yang Stem Excess/Yin Stem Deficiency,” and label each year accordingly (e.g., “Fire Excess/Water Excess/Wood Excess…”). Third, for each “Excess/Deficiency” category, statistically distribute the five extreme event types to obtain a 5 (Gan-Zhi type) × 5 (event type) confusion matrix. Fourth, use Chi-square tests to assess the significance of the “type × event” association and apply FDR or permutation testing for multiple comparison corrections—hundreds of cells across 60 Gan-Zhi types × 5 event types mean that pure randomness can generate dozens of false “significant” results.

“Significant Diagonal Lines” indicate theoretical success. The most valuable aspect of the confusion matrix is its ability to address specificity: Is the detection rate (sensitivity) for “heat” in Fire Excess years high? Simultaneously, is the spillover rate (false positive) for “flooding” in Fire Excess years high? A genuine pattern of luck should manifest as a significant main diagonal—Fire Excess years concentrated in the “heat” cell, Water Excess years concentrated in the “cold” cell—rather than an overall elevation in a single row. Historical climatology has already demonstrated the feasibility of this “event typification”: researchers aligned ancient inscription records from the Big Fish Cave in Qinling (accurate timing and impact of 7 drought events between 1520–1920) with high-resolution stalagmite oxygen isotope records from the same cave, finding mutual confirmation that textual drought events synchronize with proxy records (PMID 26270656). This suggests that coded event types from literature can cross-verify with independent proxy records, providing a template for Gan-Zhi type testing. Monsoon signals themselves are clearly reflected in both temperature and precipitation sequences—analysis of a 300-year reconstructed sequence for the middle and lower reaches of the Yangtze reveals that both the rainy season precipitation and winter temperature carry significant East Asian monsoon signals, with multi-scale cycles ranging from 2–7 years to 30–40 years and 50–100 years (PMID 26107375). In other words, robust signals for “climate states corresponding to drought-flood patterns” do exist historically—the question is whether they are locked by the 60-year Gan-Zhi cycle rather than these longer monsoon/climate cycles.

How to exclude “cycle collision.” The Gan-Zhi cycle is a quasi-period of 60 years, while Chinese climate itself exhibits significant cycles of 50–70 years and 100–120 years (multi-scale variability of the East Asian monsoon, PMID 37359908). These two sets of cycles partially overlap at the centennial scale; a careless test might detect the climate’s own oscillatory rhythm rather than Gan-Zhi. The countermeasure involves two steps: First, permutation testing—randomly shuffle the “year-Gan-Zhi” correspondence thousands of times to construct a null distribution, observing whether the diagonal intensity in the observed confusion matrix significantly exceeds random permutation; Second, exclude century-scale forcing segments—treat Medieval Warm Periods or Little Ice Ages as covariates or stratification variables, testing whether type specificity persists after controlling for large cycles. Additionally, clinical control studies based on Five Movements and Six Qi suggest that the theory carries detectable signals but with small effect sizes requiring large samples—a 15-year retrospective study in Beijing based on Five Movements and Six Qi theory found significant but weak correlations between tuberculosis incidence rates and meteorological factors of the current and previous 1–3 years, with wind speed contributing most strongly (PMID 26525546). This indicates from the disease perspective that the association between Gan-Zhi/Luck and environment “exists but is weak,” making type-specific testing a touchstone for distinguishing “true signals” from “large-cycle illusions.”

In summary, upgrading the question from “Are there anomalies in Fire Excess years?” to “Do Fire Excess years show heat without flooding?” via confusion matrix testing is a critical step moving Five Movements and Six Qi from “substantive metaphysics” to “falsifiable climatic hypotheses”—significant diagonals with controlled spillover yield true specificity evidence for the Gan-Zhi cycle; only general correlation without diagonals may merely be a mirror of large climate cycles.

Evidence and Limitations

Verified: The China East millennium grid drought-flood grade dataset (literature + tree-ring fusion with sampling errors) has been established (PMID 37041161); drought event records from Qinling cave inscriptions synchronize with stalagmite oxygen isotope records (PMID 26270656); the 300-year reconstructed sequence for the middle and lower Yangtze shows temperature and rainy season precipitation carrying significant monsoon signals with multi-scale cycles (PMID 26107375); the East Asian monsoon exhibits multi-scale cycles of 50–70 years and 100–120 years (PMID 37359908); a Beijing tuberculosis incidence-meteorology retrospective study based on Five Movements and Six Qi detected significant but weak associations (PMID 26525546).

Partially Verified: “Historical literature can be coded into event type sequences and cross-verified with independent proxy records” has been demonstrated by the Qinling cave case (PMID 26270656), but this case covers only 7 drought events with extremely small samples; the millennium drought-flood dataset covers East China since 960 AD, but proxy records in Western regions (especially Northwest) are sparse, resulting in insufficient spatial completeness for type matching (PMID 37041161).

To Be Verified: ① No global study has yet completed a full 500-year specific test combining “5 event type typification + Gan-Zhi Excess/Deficiency year assignment + confusion matrix + FDR/permutation testing”; ② Whether the main diagonal specificity of “Fire Excess years showing heat without flooding” remains robust after controlling for 50–70 year/centennial large cycles lacks direct data; ③ How spatial heterogeneity in literature event coding (dense East, sparse West) and omission biases enter confusion matrix cell counts lacks a reproducible statistical scheme.

Related Reading

Evidence Level: B
References: PMID 37041161, PMID 26270656, PMID 26107375, PMID 37359908, PMID 26525546


Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *