极端指数「去趋势后干支效应」vs「含趋势直接对比」:非平稳暖化趋势如何制造虚假符合

Written by

in

Language

Evidence level: B

六十甲子周期恰好落在现代暖化趋势的时间尺度上——直接拿含趋势的极端指数做干支分组对比,很可能把”暖化恰好撞上火太过年”误读成”火太过真的带来热”。去趋势与否,是检验五运六气最容易被忽略、也最致命的一道关卡。

正文

检验五运六气”太过不及”与极端气候的符合度,最直观的做法是:把每年按天干归入”火太过/水太过/木太过……”分组,计算各组极端指数(热夜日数、暖昼日数、霜冻日数、强降水日数等)的平均偏离,看是否符合《内经》预期。但这里藏着一个被大多数检验忽略的统计陷阱:现代气象序列不是平稳的。中国自 1961 年以来地表平均气温以每十年约 0.16–0.30℃ 的速率显著上升,极端高温事件频次持续增加(PMID 38937621)——这意味着任何跨 60 年的对比,都叠加了一条强劲的、单向的暖化趋势。而干支纪年恰好是一个 60 年周期:在 1961–2020 这样一个 60 年窗口里,”火太过”和”水太过”年份的分布会与趋势的升降段部分对齐。如果”火太过”年份恰好集中在暖化加速的后半段,直接对比就会得到”火太过年偏热”的假符合——哪怕干支本身不携带任何信息。

为什么趋势会制造假象。 非平稳序列的谱结构会被趋势污染:一条单调趋势在频域里表现为极低频的强功率,其能量会向相邻频段泄漏,抬高周期检测的底噪,甚至与真实周期产生谐波混叠,制造出不存在的周期性。吴召华等人系统讨论过这个问题:趋势的定义本身不是唯一的,线性去趋势与自适应趋势(如经验模态分解 EMD 提取的本征趋势)结果可以差异很大;选择不同的趋势定义,直接决定了”去掉趋势后还剩什么”(PMID 17846430)。换句话说,“含趋势对比”检验的是”干支 + 暖化”的混合信号,”去趋势对比”检验的才是”纯干支”信号——两者结论不同完全可预期,问题在于该信哪一个。

去趋势方法不是一刀切。 对极端指数序列,可选的去趋势路线至少三种:一是线性/多项式去趋势,简单但对非线性暖化(20 世纪 80 年代后期加速)拟合不足;二是 EMD 自适应分解,把慢变趋势当作最后一个本征模态函数剔除(PMID 17846430);三是小波分解,把高频离群与长期趋势在尺度上分离后再重构残差(PMID 38163127)。后者在非平稳植被指数序列的趋势-周期分离中已有成熟应用(PMID 33030692)。关键在于:不同去趋势方法可能给出方向相反的干支效应结论。所以检验要做”方法谱系”——线性、EMD、小波各跑一遍,看干支效应量的符号与显著性是否稳健。若只有含趋势通道显著、三种去趋势通道全部消失,基本可以判定:所谓干支效应只是暖化趋势的镜像。

双通道 + 合成序列标定。 一套严谨的设计应是”同一批站、双通道”:通道 A 用原始极端指数直接做干支分组对比(含趋势),通道 B 用去趋势距平做同样的对比,再对每个通道计算效应量(如火太过年热夜日数偏离气候态的标准差数)、年份协变量回归(把公历年份作为协变量,从效应量中再扣除线性时间趋势)以及置换检验(随机打乱”年份—干支”对应关系上万次,看观测效应量落在零分布的位置)。但光有真实数据还不够——因为真实数据里我们不知道”真效应”是多少,无法评估检验方法本身对趋势的敏感度。所以必须加一层合成序列标定:生成已知”趋势强度 × 干支分组效应”组合的人工序列(趋势强度取观测值的 0 到 2 倍、干支效应设为零或设定值),跑同样的双通道流程,画出”去趋势后检验功效 vs 趋势强度”曲线。这一步能直接回答:在观测到的暖化强度下,去趋势通道到底还有多少把握检出真效应、又有多少概率把纯趋势误报成干支信号。

运气效应与暖化方向的交互是核心。 干支理论内部其实自带”方向不对称”的预期:火太过预期偏热(与暖化同向)、水太过预期偏寒(与暖化反向)、木太过预期偏风(与暖化方向无关)。这意味着:同向组(火)天然容易被趋势抬高而得到虚假符合,反向组(水)天然容易被趋势抵消而得到虚假不符——暖化趋势会系统性地把”符合度热力图”朝”火符合、水不符”的方向扭曲。这个扭曲一旦被识别,反而提供了一个内部校准的机会:如果去趋势后”火与水”的效应量差大幅缩小甚至反转,说明原结论主要是趋势的产物;如果去趋势后方向仍然成立且量级稳健,干支信号才值得继续追。东亚季风自身还存在 50–70 年与 100–120 年的多尺度变率,与干支 60 年周期频带重叠(PMID 37359908,见前文),这更要求把慢速振荡与暖化趋势一并处理——否则检到的可能是季风的自然节律,而不是干支。

一句话总结:“含趋势对比”检验的是”干支+暖化”的混合假设,”去趋势对比”检验的才是”纯干支”假设。 用线性/EMD/小波三种去趋势跑”方法谱系”,配以合成序列标定检验功效,再把火(同向)与水(反向)的效应量差作为内部校准尺——这三件套做完,才能回答五运六气最诚实的问题:暖化背景下,干支效应还剩多少是它自己的。

证据与局限

已验证: 中国极端温度指数呈现显著非平稳趋势,复合温度极端事件与暖化/城市化强关联(PMID 38937621);非平稳序列的趋势定义与去趋势方法影响分析结论,EMD 等自适应方法可提取本征趋势(PMID 17846430);小波分解能分离长期趋势与高频波动(PMID 38163127);去趋势偏互相关(DPCCA)可量化非平稳信号间的真实关联(PMID 25634341);非平稳序列的趋势-周期分离已有成熟应用(PMID 33030692)。

部分验证: 小波分析中统计检验的零假设选择会显著影响显著性结论(PMID 24284892),提示”去趋势后残余效应”的显著性判定依赖检验框架;但尚无任何研究把”干支分组效应 × 去趋势方法谱系 × 合成序列标定”组合成完整的虚假符合/虚假不符判定流程。

待验证: ① 同一批站长序列上,含趋势通道 vs 线性/EMD/小波去趋势通道的干支效应量差异(符号是否反转、显著性是否消失);② 在观测暖化强度下,合成序列标定出去趋势通道对真效应的检验功效曲线——趋势强度多大时功效崩溃;③ “火同向/水反向”效应量差在去趋势前后的变化——它是否为内部校准提供稳定信号。

相关阅读

证据等级: B
参考文献: PMID 17846430, PMID 25634341, PMID 24284892, PMID 38163127, PMID 33030692, PMID 38937621


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。

De-trended vs. Trend-inclusive Direct Comparison: How Non-stationary Warming Trends Create False Significance

Evidence level: B

The 60-year Gan-Zhi cycle coincides precisely with the timescale of modern warming trends. Directly comparing extreme indices containing these trends across Gan-Zhi groups risks misinterpreting “warming coinciding with Fire Excess years” as “Fire Excess truly causing heat.” Whether to detrend or not represents the most overlooked yet critical checkpoint in validating Five Movements and Six Qi.

Main Text

The most intuitive approach to validating the concordance between Five Movements and Six Qi “excess/deficiency” and extreme climates is to group each year by Heavenly Stem into categories such as “Fire Excess,” “Water Excess,” “Wood Excess,” etc., calculate the average deviation of extreme indices (number of hot nights, warm days, frost days, heavy precipitation days, etc.) within each group, and assess whether they align with the expectations in the Inner Canon. However, this approach overlooks a statistical trap: modern meteorological sequences are non-stationary. Since 1961, China’s mean surface temperature has risen significantly at a rate of approximately 0.16–0.30°C per decade, while the frequency of extreme heat events has continued to increase (PMID 38937621). This implies that any comparison spanning 60 years is superimposed with a strong, unidirectional warming trend. Since the Gan-Zhi calendar operates on a 60-year cycle, the distribution of “Fire Excess” and “Water Excess” years within a 60-year window (e.g., 1961–2020) may partially align with the rising and falling phases of the trend. If “Fire Excess” years cluster specifically in the latter half of the accelerating warming phase, a direct comparison will yield a false concordance showing “Fire Excess years are hotter,” even if the Gan-Zhi system carries no actual signal.

Why trends create illusions. Non-stationary sequences have spectral structures contaminated by trends: a monotonic trend manifests as strong low-frequency power in the frequency domain, whose energy leaks into adjacent bands, raising the baseline noise for cycle detection and potentially causing harmonic aliasing with true cycles, thereby fabricating non-existent periodicity. Wu Zhaohua and others have systematically discussed this issue: the definition of a trend itself is not unique; linear detrending versus adaptive trends (such as intrinsic trends extracted via Empirical Mode Decomposition, EMD) can yield significantly different results; choosing different trend definitions directly determines “what remains after removing the trend” (PMID 17846430). In other words, “comparison with trend” validates a mixed signal of “Gan-Zhi + Warming,” while “detrended comparison” validates the “pure Gan-Zhi” signal—different conclusions are entirely expected, but the question is which one to trust.

Detrending methods are not one-size-fits-all. For extreme index sequences, at least three detrending routes are available: (1) Linear or polynomial detrending, which is simple but insufficient for fitting non-linear warming acceleration observed in the late 1980s; (2) EMD adaptive decomposition, which removes the slow-varying trend as the last Intrinsic Mode Function (PMID 17846430); (3) Wavelet decomposition, which separates high-frequency outliers from long-term trends across scales before reconstructing the residuals (PMID 38163127). The latter has already established mature applications in trend-cycle separation for non-stationary vegetation index sequences (PMID 33030692). The critical point is that different detrending methods may yield opposite directions for Gan-Zhi effects. Therefore, validation must employ a “method lineage”—running linear, EMD, and wavelet detrending separately to check if the sign and significance of Gan-Zhi effect magnitudes remain robust. If only the “with-trend” channel shows significance while all three detrending channels vanish, it can be reasonably concluded that the so-called Gan-Zhi effect is merely a mirror image of the warming trend.

Dual-channel + Synthetic Sequence Calibration. A rigorous design should be “same stations, dual channels”: Channel A uses raw extreme indices for direct Gan-Zhi grouping (with trend), while Channel B uses detrended anomalies for the same comparison. Subsequently, calculate effect magnitudes (e.g., standard deviation of hot night deviations from climatology for Fire Excess years), perform year covariate regression (including public year as a covariate to subtract linear temporal trends from effect magnitudes), and conduct permutation tests (randomly shuffling the “year-Gan-Zhi” correspondence thousands of times to see where the observed effect magnitude falls within the zero distribution). However, real data alone is insufficient because we do not know the “true effect” in real data, making it impossible to evaluate the sensitivity of the testing method itself to trends. Therefore, an additional layer of synthetic sequence calibration is required: generate artificial sequences with known combinations of “trend intensity × Gan-Zhi grouping effects” (setting trend intensity at 0 to 2 times the observed value and setting Gan-Zhi effects to zero or fixed values), run the same dual-channel workflow, and plot a curve of “detrended detection power vs. trend intensity.” This step directly answers: under the observed warming intensity, how much confidence does the detrended channel have to detect true effects, and what is the probability of falsely attributing pure trends to Gan-Zhi signals.

Interaction between Luck Cycle Phase Drift and Warming Direction is Core. Internally, the Gan-Zhi theory already carries expectations of “directional asymmetry”: Fire Excess expects higher temperatures (aligned with warming), Water Excess expects lower temperatures (opposed to warming), and Wood Excess expects windier conditions (unrelated to warming direction). This means: same-direction groups (Fire) are naturally prone to being elevated by trends, yielding false concordance, while opposite-direction groups (Water) are naturally prone to having their signals canceled by trends, yielding false discordance. The warming trend systematically distorts the “concordance heatmap” toward “Fire Concordance, Water Discordance.” Identifying this distortion actually provides an internal calibration opportunity: if the difference in effect magnitudes between “Fire” and “Water” shrinks significantly or reverses after detrending, the original conclusion was primarily a product of the trend; if the direction remains valid and the magnitude is robust after detrending, the Gan-Zhi signal warrants further pursuit. The East Asian monsoon itself exhibits multi-scale variability with periods of 50–70 years and 100–120 years, overlapping with the Gan-Zhi 60-year cycle band (PMID 37359908, see previous text), which further necessitates handling both slow oscillations and warming trends together—otherwise, what is detected may be the natural rhythm of the monsoon rather than Gan-Zhi.

In summary: “Comparison with trend” validates a mixed hypothesis of “Gan-Zhi + Warming,” while “detrended comparison” validates the “pure Gan-Zhi” hypothesis. Running a “method lineage” with linear/EMD/wavelet detrending, paired with synthetic sequence calibration to assess detection power, and using the difference in effect magnitudes between Fire (same-direction) and Water (opposite-direction) as an internal calibration ruler—these three components allow us to answer the most honest question regarding Five Movements and Six Qi: under the context of warming, how much of the Gan-Zhi effect remains its own.

Evidence and Limitations

Verified: Chinese extreme temperature indices show significant non-stationary trends; composite temperature extremes are strongly associated with warming and urbanization (PMID 38937621); trend definitions and detrending methods in non-stationary sequences influence analytical conclusions, where adaptive methods like EMD can extract intrinsic trends (PMID 17846430); wavelet decomposition can separate long-term trends from high-frequency fluctuations (PMID 38163127); Detrended Partial Canonical Correlation Analysis (DPCCA) can quantify true associations between non-stationary signals (PMID 25634341); trend-cycle separation for non-stationary sequences has established mature applications (PMID 33030692).

Partially Verified: The choice of null hypothesis in statistical tests within wavelet analysis significantly affects significance conclusions (PMID 24284892), suggesting that the significance determination of “residual effects after detrending” depends on the testing framework; however, no study yet combines “Gan-Zhi grouping effects × detrending method lineage × synthetic sequence calibration” into a complete false concordance/false discordance determination workflow.

To Be Verified: ① The difference in Gan-Zhi effect magnitudes between the “with-trend” channel and linear/EMD/wavelet detrending channels on the same station series (whether signs reverse or significance disappears); ② Under observed warming intensity, the detection power curve for the detrended channel derived from synthetic sequence calibration—at what trend intensity does power collapse; ③ The change in the difference between “Fire same-direction/Water opposite-direction” effect magnitudes before and after detrending—whether it provides a stable signal for internal calibration.

Related Reading

Evidence Level: B
References: PMID 17846430, PMID 25634341, PMID 24284892, PMID 38163127, PMID 33030692, PMID 38937621


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 *