Global Extreme Weather and Climate Change Dashboard

Detection and Time of Emergence

Every verdict on this site answers one question: does the observational record show a detected change? The IPCC's emergence assessments answer a different question. Readers often treat the two as one. This page explains what each question asks, how they differ, and where this site and AR6 disagree about the same phenomenon.

The short version. Detection asks whether a signal has already appeared in the record. Emergence asks when a signal rises above the noise, and the literature calculates that in several different ways. A third question asks how long a record would need to run before a projected signal could appear at all. For some phenomena the answer runs to centuries. A "no detected change" verdict here answers the first question only.

The IPCC definitions, word for word

AR6 WG1 defines all four terms in Annex VII, the Glossary. They appear in full below, because the differences between them turn on the exact wording:

Detection

Detection of change is defined as the process of demonstrating that climate or a system affected by climate has changed in some defined statistical sense, without providing a reason for that change. An identified change is detected in observations if its likelihood of occurrence by chance due to internal variability alone is determined to be small, for example, <10%.

Attribution

Attribution is defined as the process of evaluating the relative contributions of multiple causal factors to a change or event with an assessment of confidence.

Emergence (of the climate signal)

Emergence of a climate change signal or trend refers to when a change in climate (the 'signal') becomes larger than the amplitude of natural or internal variations (defining the 'noise'). This concept is often expressed as a 'signal-to-noise' ratio and emergence occurs at a defined threshold of this ratio (e.g., S/N > 1 or 2). Emergence can refer to changes relative to a historical or modern baseline (usually at least 20 years long) and can also be expressed in terms of time (time of emergence) or in terms of a global warming level.

Time of emergence (ToE)

Time when a specific anthropogenic signal related to climate change is statistically detected to emerge from the background noise of natural climate variability in a reference period, for a specific region.

Three points follow.

• Detection makes no claim about cause. The glossary says so directly — "without providing a reason for that change". Attribution names the causes. It requires model evidence, and this site does not do it.
• Time of emergence builds on detection. The glossary defines it as the time when a signal "is statistically detected to emerge from the background noise of natural climate variability in a reference period". That shared wording makes the two look like one test looked at from two directions. The section below works through how far that holds.
• Both definitions include a reference period. The glossary names "a historical or modern baseline (usually at least 20 years long)", and a different baseline gives a different answer for the same series. AR6 Chapter 12 works to a pre-industrial baseline by default, which differs from the glossary's modern-baseline case. The chapter states which baseline it uses each time.

What that means for a reader here: a "no detected change" verdict says that this index, over this window, shows no change clearing the two conditions set out below. It does not say the phenomenon holds steady, it gives no date for a signal, and it does not settle what AR6 would conclude about emergence.

Looking backwards, are they the same test?

AR6 puts emergence at the date a signal first exceeds a natural-variability reference. A date in the past therefore means the signal exceeds that reference today. Detection and emergence share that much. Four differences in AR6's definitions separate them, and each one appears below.

• Cause. AR6 defines a time of emergence around "a specific anthropogenic signal related to climate change". It defines detection "without providing a reason for that change". A past time of emergence therefore names human influence. A detected change here names nothing at all.
• The statistic. Detection asks how often internal variability alone would produce a record like this one by chance, and AR6's worked example puts the bar below 10%. Emergence divides the signal by the noise and asks whether the ratio clears a threshold, "e.g., S/N > 1 or 2".
• The baseline. AR6 Chapter 12 measures emergence against a pre-industrial period by default, and no instrument ever measured that baseline. The verdicts here rest on the observational record, over a window chosen for that variable's data quality. Move the baseline and the emergence date moves with it.
• Two different outputs. Detection returns a verdict over a window. Emergence returns a date: the year the signal first exceeded the reference.

The two also disagree about the same series, in both directions.

• Detection without emergence. A long, quiet record can show a trend that clears the significance bar while the total change stays smaller than the year-to-year variation. The signal-to-noise ratio sits under 1, so nothing has emerged, and the significance test still reads positive. The magnitude condition below covers that case, and it turns away some statistically real trends.
• Emergence without detection. Measured against a pre-industrial baseline, a shift can already have emerged while the trend inside the observed window misses the significance bar. The emergence literature reports a date and this site reports no detected change, from the same underlying climate.

Three things separate a past time of emergence from a detected change: a named cause, a baseline no instrument recorded, and a statistic different from the one behind every verdict on this site.

What "detected" means on this site

This dashboard applies the IPCC definition: a change counts as detected when its likelihood of arising by chance from internal variability alone falls below 10%. The test applies Mann-Kendall to the observed series, with the sample size corrected for lag-1 autocorrelation in the detrended residuals, since persistence in a record inflates the apparent certainty of a trend. Three further conditions apply:

• A 30-year floor. A series shorter than the WMO's climate-normal period gets no verdict.
• A magnitude condition. The likelihood test alone does not settle it here: the total change across the window must also reach 25% of the 90% range within that same record. A trend can pass a significance test and remain small against the variability a reader would notice.
• A correction for testing many variables at once. Of the 72 headline variables here, 57 hold enough record to test, and a Benjamini-Hochberg false-discovery-rate correction ranks those p-values together. 27 clear the corrected bar. See the Methodology page for the full treatment.

All of that concerns detection, not attribution. Measuring how much of a change traces to human influence requires model-based analysis. This project does not do that, and no page here claims otherwise.

Emergence covers several methods, not one test

AR6 WG1 Chapter 12 opens its emergence section by naming two definitions in a single sentence: a signal emerges when it "exceeds some critical threshold (usually taken to be a measure of natural variability)", or when "the probability distribution of an indicator becomes significantly different to that over a reference period". The chapter then states that "a homogeneous interpretation of multiple studies is hampered by heterogeneous methodologies used to calculate emergence", assesses across "multiple methods as provided by the literature", and only then adopts a house convention — unless specified otherwise, a signal-to-noise ratio above 1 against a pre-industrial baseline and interannual variability.

The convention involves two further choices, and the chapter states both. The choice of reference period changes the answer: measured against the pre-industrial period, emergence estimates the size of an anthropogenic change; measured against a recent past, it estimates change against recent conditions. The variability timescale matters too — interannual noise and decadal noise give different answers for the same series. Of heavy precipitation the chapter says plainly that results "depend on the method used for the calculation of the ToE".

The third question: how long would it take?

Crompton, Pielke Jr. & McAneney (2011) put this question to normalized United States tropical cyclone losses. They took the projected change in storm activity from an independent modelling study, translated it into the units of the loss record, and asked how long that record would have to run before such a trend cleared the variability already in it. The answer ran from 120 to 550 years across models, and 260 years for the 18-model ensemble.

The signal comes from a projection made independently of the record, and the record supplies only the noise. An effect size taken from the same data makes the calculation circular, and a circular calculation can always explain away a null result. The statistics come from standard methods. Weatherhead et al. (1998) gives the years-to-detection for a linear trend against autocorrelated noise, and this site already computes it internally for every variable.

This calculation needs two things, and only one variable here has both. A projected signal has to come from an assessment made independently of the observations, and it has to arrive in the variable's own units. AR6 Tables 12.3–12.11 assess confidence in the direction of change, not magnitudes, so they cannot supply one. Figure SPM.6 quantifies three hazards, and two of them do not convert: its drought figures apply to drying regions while the drought indices here run over global land, and its frequency multipliers describe 10- and 50-year return levels rather than the annual indices tracked here.

An emergence timescale for TXx

AR6 WG1 Chapter 11 assesses warming of land-mean TXx as similar to land-mean warming, which runs about 45% higher than global warming. TXx on this site tracks that same quantity in that same unit — the annual hottest daily maximum, averaged across global land stations — so the ratio transfers with no conversion step. Global warming supplies the rate: SR1.5 assesses anthropogenic warming as increasing at 0.2 °C per decade, likely between 0.1 °C and 0.3 °C, which gives a projected TXx trend of 0.29 °C per decade at the central rate.

The noise and the year-to-year persistence come from this site's TXx record; only the trend magnitude comes from AR6. A faster warming rate steepens the projected trend, so it needs a shorter record: the 0.3 °C per decade case needs fewer years than the 0.1 °C case. At the central rate the requirement falls inside the record this site already holds, and TXx does show a detected increase. The calculation and the verdict agree, which is the check to run before trusting either.

The remaining 24 headline variables have no assessed projection to test against. No assessment quantifies one for them, which says what the assessments cover, not a gap in this site's reading of the literature. The table below names each one and what the assessments leave unsaid.

Where detection and emergence differ

• They ask different questions. A null detection says a signal has not appeared in this record. A long emergence timescale says how long a signal of an assessed size would take to appear. Neither implies the other.
• They measure from different baselines. This site shows change within the observational record, across a window chosen for that variable's data quality. AR6's convention measures against a pre-industrial baseline, which no modern instrument observed directly.
• They use different statistics. A signal-to-noise ratio and a rank-based significance test can disagree about the same series.
• They apply to different units. A geophysical index and a loss record behave differently. In a loss series, exposure and wealth drive most of the variance, not weather. The Loss Normalization page comes first for that reason.
• Confidence and significance differ. AR6's colours express expert assessment across many studies and lines of evidence. The verdicts here come from one computation on one dataset. Treating a white cell and a "no detected change" as the same finding misreads both.

AR6 Table 12.12: what has emerged, and what has not

Chapter 12's Table 12.12 assesses 33 climatic impact-drivers across three windows. The table below reproduces it. The colour, not the text, shows the assessment. In AR6's words, "white cells indicate where evidence is lacking or the signal is not present, leading to overall low confidence of an emerging signal." Source: AR6 WG1 Chapter 12, Table 12.12, page 1856.

9 of 33 drivers show an assessed emergence in the historical period: Mean air temperature; Extreme heat; Cold spell; Permafrost; Lake, river and sea ice; Mean ocean temperature; Ocean salinity; Dissolved oxygen; Atmospheric CO2 at surface. Of the 18 drivers that name a discrete extreme event rather than a mean or a background state, 2 have emerged — extreme heat and cold spell — and this site shows both. A further 15 event drivers stay white in all three windows, with no emergence assessed even by 2100 under RCP8.5.

The event-versus-state distinction in that paragraph belongs to this site, not to AR6. Table 12.12 groups its rows by physical type, and draws no such line.

The same drivers, measured here

Table 12.12 holds a driver for 19 of this site's 25 headline variables. The rest measure something AR6's driver list does not treat as a category: US significant tornado count (F/EF2+); Mean global CAPE; Days with widespread or extreme instability; Global dry matter combusted; Global burned area; % of land in D2+ (severe or worse) meteorological drought — 6 variables in all. Wildfire shows the mismatch most clearly. AR6's driver covers fire weather, the conditions that favour fire. The two wildfire series here measure fire on the ground. The Detection & Attribution page treats that distinction directly.

11 variables measuring these drivers show a detected change here: Heat wave index (GSN stations), Heat wave magnitude (HWMId-style, GSN stations), TXx (annual max daily high temperature), WSDI (warm spell duration index), TN90p (warm nights), CSDI (cold spell duration index), TN10p (cold nights), % of pour-point cells in D2+ (severe or worse) low flow, % of land in D2+ (severe or worse) soil-moisture deficit, Storm count by peak intensity category, % of hurricanes that are Cat 3+. A detected change here and an emergence assessment in AR6 answer different questions. Neither column tests the other.

Where the two disagree: drought

Chapter 12 assigns "low confidence in the emergence of drought frequency in observations, for any type of drought, in all regions", and Table 12.12 leaves hydrological drought, agricultural and ecological drought, and aridity white in all three windows. The series measured here give a different answer.

• % of land in D2+ (severe or worse) meteorological drought: detected increase
• % of land in D2+ (severe or worse) soil-moisture deficit: detected increase
• % of pour-point cells in D2+ (severe or worse) low flow: detected increase

Each of those measures shows a detected change here, while AR6 assesses low confidence in emergence for every drought driver. Several differences separate the two: a different index, a three-month window against AR6's longer ones, reanalysis precipitation against gauge-based records, and a significance test against a signal-to-noise criterion. The Drought page sets out the full treatment, including this site's departure from the paper its index derives from. This page states both findings and reconciles neither. Reconciling them would take work that neither analysis has done.

Every source here links to its published DOI, and the AR6 material links to the chapter and Summary for Policymakers PDFs. Full citations sit on the Library page.