State of
the planet.

The 2026 Indicators of Global Climate Change assessment puts observed warming over 2016–2025 at 1.26°C above 1850–1900, with a range of 1.13–1.36°C; human-induced warming over the same decade is 1.24°C. Human-induced warming is now advancing at about 0.27°C per decade, with a likely range of 0.2–0.4°C. These decade-scale figures measure the sustained shift rather than any single year.⁹

The HadCRUT5 record shows the same structure. Its 2016–2025 mean is 1.28°C above 1850–1900; 2024 stands at 1.53°C and 2025 at 1.40°C. Ordinary least-squares trends steepen from 0.21°C per decade over 1975–2025 to 0.25°C over 2000–2025 and 0.36°C over 2011–2025. The last figure is inflated by the 2023–2024 peak and by natural variability within a short window; it is a description of the record, not an estimate of the forced rate.^2,12

WMO reports that August 2026 was the warmest August on record and June–August the warmest Northern Hemisphere summer, with ocean-surface temperatures setting a daily record during August. El Niño added to those conditions on top of the long-term forcing.¹³

The energy behind the trend

Earth’s energy imbalance—the gap between absorbed sunlight and emitted heat—is the quantity that drives accumulation. A synthesis of satellite and ocean observations finds that it has more than doubled in recent decades and now runs well above what climate models simulated for the period. The authors link much of the discrepancy to falling solar reflectivity and warn that the capacity to observe the imbalance is deteriorating as satellites are decommissioned.¹⁰

The reflectivity change has a specific signature. In 2023 global temperature rose almost to 1.5°C above preindustrial, beating the previous record by about 0.17°C; known drivers including anthropogenic warming and El Niño’s onset fall about 0.2°C short of explaining it. Satellite and reanalysis data identify a record-low planetary albedo as the main bridge, largely through reduced low cloud over the northern mid-latitudes and tropics. Whether that cloud decline is internal variability, fewer aerosols or an emerging feedback remains open.¹⁴

Carbon dioxide continues to accumulate faster each decade. At Mauna Loa the average annual increase rose from 0.88 ppm per year in the 1960s to 2.41 in 2010–2020 and 2.63 over 2015–2025. The 2025 annual mean was 427.35 ppm, 2.74 ppm above 2024, after a 3.53 ppm rise from 2023 to 2024.^2,15

Water, ice and the sea retain the history of stress

WMO’s September 2026 water assessment reports a continuing decline in global freshwater storage and widespread glacier loss, with about 1,400 gigatonnes of glacier mass lost over 2023–2025. Groundwater, snow and glaciers each carry water across seasons, so their depletion sets a worse starting position for the next dry period. Local water security still turns on withdrawals, storage, catchment conditions and management.¹⁶

Satellite altimetry shows global mean sea level rising 111 mm from 1993 to 2023, with the rate increasing from about 2.1 mm per year at the start of the record to about 4.5 mm per year at its end. Continuing that trajectory adds about 169 mm over the next three decades, comparable to mid-range IPCC projections. Ocean and ice-sheet responses commit further rise over centuries.^11,17

The Planetary Health Check 2026 finds seven of nine planetary boundaries transgressed: climate change, biosphere integrity, land-system change, freshwater, nutrient cycles, novel entities and ocean acidification. The boundaries are assessed guardrails for Earth-system processes, so seven of those processes now operate outside the range considered safe.¹⁸

Coral reefs show why exposure, damage and recovery are separate measurements. NOAA reports that about 84% of the world’s coral reef area experienced bleaching-level heat stress between early 2023 and mid-2025, and its June 2026 assessment concludes that the fourth global bleaching event likely ended in mid-2025. The end of an event marks the end of widespread heat stress; recovery of the injured reef communities is a separate and slower process.¹⁹

●A documented severe outcome

After the 2023 marine heatwave, a study of Florida’s Coral Reef found 97.8–100% mortality among the monitored Acropora palmata and Acropora cervicornis colonies in the Florida Keys and Dry Tortugas, and 37.9% mortality offshore southeast Florida. The authors conclude that these two reef-building corals are now functionally extinct on Florida’s Coral Reef.²⁰ A heatwave striking an intact reef, a full reservoir or a rested workforce is a different event from the same heatwave striking a depleted one. The relevant reserve is the function that remains when the next hazard arrives.

Climate now registers in Earth’s rotation

Melting ice moves mass from the poles toward the equator, increasing Earth’s oblateness and lengthening the day. Reconstructions of surface mass change since 1900 put the climate-induced trend in length of day between 0.3 and 1.0 milliseconds per century through the twentieth century and 1.33 ± 0.03 ms per century since 2000. Under high-emission scenarios it may reach 2.62 ± 0.79 ms per century by 2100, overtaking lunar tidal friction as the largest single contributor to long-term change.²¹ The effect is tiny for daily life and large as a fingerprint: surface climate change now shows up in planetary geodesy alongside the core and the Moon.

Circulation: a contested tipping element

The Atlantic Meridional Overturning Circulation illustrates a mechanism whose direction is agreed and whose threshold is not. A physics-based early-warning analysis in a state-of-the-art model reports the circulation to be on a tipping course.²² An analysis of 34 climate models under extreme greenhouse and freshwater forcing finds it weakened but sustained by Southern Ocean upwelling in every case, concluding that a twenty-first-century collapse is unlikely.²³ Both findings are model-based. The disagreement concerns whether models capture the stabilizing and destabilizing pathways, which observations alone cannot yet settle.

? Unresolved  A collapse of the Atlantic overturning circulation this century.

Long-lived damage leaves further warming in human hands

Global surface temperature responds differently from the slow components: once CO₂ emissions reach zero, the model comparison finds a central multidecadal temperature commitment near zero, with spread across models. Continued long-lived damage and the ability to halt further warming are both true at once.^17,24 For infrastructure, this separates two tasks. Existing exposure requires protection and recovery capability now. Future forcing sets the burden those protections will face.

Adequacy in normal conditions is different from endurance under stress

Before summer 2026, NERC assessed anticipated resources as adequate for normal peak conditions in every North American bulk-power assessment area, while identifying elevated risks in selected areas under more extreme conditions. That was a prospective adequacy assessment, rather than a record of every local outage.²⁵ Its longer-horizon assessment found resource-adequacy challenges in 13 of 23 areas over 2026–2035, under the demand and resource plans supplied to it. NERC also records later demand revisions and accelerated resource programmes that can improve the outlook. The risk is a changing balance of demand, usable supply and timely investment—not a uniform decline in every region.²⁶

Four-epoch comparison plate: 1900, 1945, 1985 and 2025

The dipole is declining, and the decline has eased

The dipole moment—the strength of the field’s dominant, bar-magnet-like part—fell from 8.32 to 7.69 × 10²² A m² between 1900 and 2025, a decline of 7.6%, or 6.1% per century on average. The rate has not been constant. In five-year steps it peaked near 9.2% per century around 1915–1920 and again near 8.7% around 1980–1985; since 2015 it has run at 4.1–4.7% per century, and IGRF-14’s secular variation implies about 5.3% for 2025–2030.^1,2

The Holocene comparison is larger—and the denominator must be named

A 1900 baseline measures the instrument-era change. It does not measure the retreat from the strongest part of the Holocene. In GGF100k, the sampled Holocene high is 9.824 × 10²² A m²; the strongest 1,000-year average is 9.603. Against the 2025 total dipole of 7.690, those differences are −21.7% and −19.9%. The strongest 500- and 2,000-year averages give −21.0% and −18.3%. Selecting a sustained high is a meaningful answer to “maximum average,” but the averaging interval is part of the definition.⁹²

Ebihara and colleagues used 11.64 × 10²² A m² as their strong-dipole experiment: 1.5 times their standard 7.76, linked in their discussion to the Holocene maximum about 2,650 years ago reported by Knudsen and colleagues. Today’s total dipole is 33.9% below that input. This supports a substantially larger high-to-present comparison than 7.6%, but it is not a newly measured maximum, a Holocene-wide mean, or a maximum millennium average.^94,93

Strength, energy and stand-off distance have different percentages. Relative to 11.64, the 2025 total dipole retains 66.1% of the moment. Squaring that ratio gives 43.6%, or a 56.4% reduction in an idealized dipole magnetic-energy scaling at fixed geometry—not 56.4% less biological shielding. A simple dipole pressure-balance model at unchanged solar-wind pressure gives a characteristic radius proportional to M^1/3, hence about 12.9% smaller. These are transformations under stated assumptions; neither predicts ground dose or grid damage.^92,94

The Holocene-wide mean of the same reconstruction is 8.098; today’s dipole is 5.0% below that mean. The 100,000-year mean is 6.827, making today 12.6% higher. A high-to-present loss, a mean-relative anomaly and a present rate of change are three different facts. Choosing only the least dramatic or only the most dramatic denominator would conceal that structure.⁹²

Newer reconstructions also challenge confidence in any single Holocene curve. Nilsson and colleagues’ 2024 joint reconstruction argues that hemispheric data imbalance has biased some dipole estimates low and that stronger Holocene fields can explain part of the slow cosmogenic-isotope variation often attributed to the Sun. That makes the baseline question more important—not settled by one smoothed series. The same result also weakens the independence of a supposed solar cycle inferred after an incorrect geomagnetic correction.⁹⁵

Two published analyses frame what the decline does and does not imply. A stochastic model of the axial dipole built from two million years of paleomagnetic data finds the current rate of decline unusual but within natural variability.³⁰ A follow-up shows that the historical trend would reverse the field in under 2,000 years if sustained, and that sustaining it that long is highly unlikely given how trends of that length vary in the paleomagnetic record.³¹

Supported The decline can be produced by natural geomagnetic variability. This is a statement about mechanism, not a guarantee of stability or harmless exposure.^30,98

Reweighting toward smaller scales

GeoPhasix tracks one part of the reorganization with the ratio of quadrupole to dipole degree power, R₂/R₁. The ratio grew 1.948-fold between 1945 and 2025 and about 2.75-fold since 1900. The Spectral Migration Index, SMI = ln(R₂/R₁), expresses that relative weighting on a logarithmic scale.²⁹

Most of the degree-two growth has a simple geometric meaning: the best-fitting dipole is moving away from Earth’s centre. This equivalent eccentric dipole sat about 407 km from the centre in 1945 and 609 km in August 2026. It is moving faster than in earlier periods of this record, at about 2.8 km per year since 2020 against 1.75 km per year over 1900–1950. This is a mathematical description of field geometry, not a physical displacement of the core or the planet’s centre of mass. The dipole remains the dominant component.²⁹

◐Reconstruction: IGRF-14 and CHAOS-8.7

Two descriptions of the same change

The rate of reweighting fluctuates even as the displacement accelerates. Its highest five-year values came in the late 1970s and its satellite-era peak around 2013; over January 2025–August 2026 it stood at about 0.0075 per year, slightly below its 1945–2025 average. The recent easing comes mainly from the non-offset twist turning negative, while the offset part, which carries the displacement, still supplies most of the growth.²⁹

◐Reconstruction: IGRF-14 and CHAOS-8.7

The reweighting rate fluctuates and has eased recently

The north pole surged toward Siberia, then braked

The north dip pole—where the field points straight down—moved slowly around northern Canada for centuries. NOAA’s annual positions, from the gufm1 historical model and IGRF, put its speed under 15 km per year for most of the record, rising sharply after 1990 to a peak near 58 km per year around 2010, then slowing every year since to about 37 km per year by 2024. The British Geological Survey describes the deceleration from about 50 to 35 km per year as the largest it has observed.^2,32,33

Satellite-era analyses attribute the surge to a change in the pattern of magnetic flux at the core surface: a lobe of strong flux beneath Canada elongated and weakened while a Siberian lobe strengthened, shifting the balance of the tug between them.³⁴ The south dip pole has moved far less, at 3–9 km per year since 1985 in IGRF-14. Pole motion and dipole decline are different measurements of the same core flow, and they need not accelerate together.

A second weak-field feature reaches about 250 km

The August 2026 reconstruction contains a secondary surface minimum of about 23,304 nT. Followed upward, it remains a distinct minimum to about 249 km, where it merges with a neighbouring saddle point. Above that height the field has a single minimum, including at the 450 km altitude of many low-orbit spacecraft.²⁹

The merger has the form of a generic minimum–saddle fold: near the merging height, the strength difference between minimum and saddle shrinks as the three-halves power of the remaining distance. The feature’s reach is climbing about 12 km per year, slower than its mid-2010s peak, and its estimated height shifts by roughly ten kilometres with the model’s spatial resolution.²⁹ Its operational meaning runs through particle transport. Radiation changes can appear before a second minimum reaches a spacecraft’s altitude, and a new minimum adds a second dose maximum only where particles populate it.^35,36

The core runs on several clocks at once

Satellite data now resolve interannual structure in the core itself. Recurrent non-axisymmetric flow patterns with periods of about seven years have been identified as magneto-Coriolis waves, with related activity in bands centred near 3.5 and 15 years.^37,38 The arrival of such wave energy at the core surface accounts for many features of interannual field change, including the sharp changes in secular variation known as geomagnetic jerks; slower fluctuations of a planetary-scale gyre account for decadal changes in the length of day.³⁹

Deeper still, repeating earthquakes recorded between 1991 and 2023 show that the solid inner core began to rotate more slowly than the mantle around 2010, backtracking relative to the surface for the first time in decades. The authors attribute the change to coupling with the outer core’s flow and to gravitational pull from dense mantle regions; its effect on day length is of order a thousandth of a second.⁴⁰

✓ Supported The inner core’s motion relative to the mantle changes. ? Unresolved A particular inner-core change as the cause of the modern South Atlantic trajectory. The emerging coupling research below distinguishes modeled pathways from that larger attribution.

The layers interact. They do not share one clock.

Some connections are direct physics: mantle heat extraction helps drive core convection; conducting fluid generates the magnetic field; angular momentum moves between the core, mantle and surface fluids. Other connections are hypotheses about which process drives a particular observed oscillation. Those are different levels of knowledge. The same word—reversal—can describe a regional flow changing direction, the inner core changing its motion relative to the mantle, or a global magnetic polarity transition. Only the last is a geomagnetic reversal.^39,40,41

A regional flow reversal is not a polarity reversal

Madsen and colleagues infer core-surface flow from magnetic observations spanning 1997–2025. Beneath the equatorial Pacific, their reconstruction changes from weakly westward to strongly eastward around 2010, then weakens after 2020. They propose a connection with contemporaneous inner-core behaviour; the timing motivates that hypothesis but does not establish it. This is evidence of reorganizing flow within a working dynamo—not an observation of the magnetic poles exchanging global polarity.⁴¹

A separate September 2026 study by Zhang and Dumberry reconstructs the torques associated with multidecadal changes in day length. Its observation-constrained model favours gravitational coupling between the inner core and mantle as the driver, with electromagnetic and topographic torques generally resisting the variation. The result depends on the inferred inner-core rotation and core-flow histories. It strengthens a specific deep-Earth coupling explanation; it does not establish an impending excursion or a driver of modern warming.⁴²

The six-year connection is a serious question—with an unresolved mechanism

Cazenave and colleagues review an approximately six-year band in core magnetism, Earth rotation and several climate-related records, including atmospheric angular momentum, sea level and hydrology. The day-length component is of order 0.1 milliseconds, and the reported period varies with the record rather than behaving like a precise metronome. They regard the pattern as a substantive research problem while leaving its system-wide origin unresolved.⁴⁶

For interpretation, matching spectral peaks is a starting point. A shared-driver explanation must also reproduce spatial pattern, phase, amplitude and a viable transfer of energy or angular momentum. A band-pass filter makes a selected band visible; it does not identify the cause. Neither the presence of a six-year mode nor a multidecadal torque cycle supplies a countdown to a millennial-scale excursion.

Climate already changes rotation—without moving the magnetic pole

Redistributing ice and water changes the planet’s mass distribution. A 2024 model attribution assigns about 90% of the interannual and multidecadal variation in rotational polar motion to surface mass redistribution, while mantle processes dominate the long-term drift. This concerns the rotation axis relative to the crust, not the north magnetic pole that travels kilometres per year. Feedback involving core dynamics is a proposed component of the attribution, not a demonstrated explanation for the South Atlantic Anomaly.⁴³

Past weak-field states ended in different ways

Historical and paleomagnetic reconstructions contain SAA-like structures that did not progress into excursions or reversals. A 9,000-year reconstruction links such anomalies to millennial dipole variation and proposes the asymmetric field around 600 BCE as an analogue. Its suggested disappearance of the modern SAA within a few centuries is an analogue-based projection, not a validated recovery timetable; newer ensemble reconstructions limit confidence in the ancient comparison.^47,48,49,96

Qualified SAA-like structures have historical analogues, but the modern anomaly is more pronounced in one newer model comparison. Neither its precise uniqueness nor a reversal already in progress is established. ^96,49

The modern field’s trajectory is therefore read from its own evolution. Resemblance to a weak-field configuration leaves the endpoint open, and the consequences are measured as exposure: where particles concentrate and where induced currents flow.

The South Atlantic may retain a mantle imprint without remaining continuously weak

A 2024 study of Trindade volcanic directions and regional paleomagnetic records finds persistent strong non-dipole behaviour in the South Atlantic over the past ten million years. It supports a long-lived influence of heterogeneous conditions at the core–mantle boundary. But directional non-dipole structure is not the same observable as low field intensity: normal and reversed flux patches can both increase directional dispersion. The evidence favours a persistent regional preference, not an uninterrupted modern-style anomaly or a short-term reversal clock.⁵⁰

That distinction allows two ideas to coexist: mantle structure may shape where anomalies preferentially develop, while faster core flow determines their evolution and disappearance. A long-lived geographical influence does not force a monotonic decline at the surface.^39,50

LLSVPs and jerks are relevant—but they are not independent countdowns

The large low-shear-velocity provinces, or LLSVPs, are deep-mantle structures beneath Africa and the Pacific identified seismically. They can influence the thermal boundary conditions above the liquid core. A persistent boundary pattern can organize much faster flows and waves; a slow mantle does not require every magnetic response to be slow. But seismic structure is not a direct measurement of current core–mantle heat flux, nor of a newly crossed dynamo threshold.^39,50,99

Rogers, Beggan and Whaler’s 2025 analysis finds substantial regional secular-variation structure associated with LLVP regions, but no systematic suppression of secular variation beneath them. It is useful evidence against treating the structures as simple magnetic dead zones. It does not turn their existence into a date for field collapse. The relevant question is whether a changing boundary condition or core response is independently observed, rather than whether a persistent structure exists.⁹⁹

A geomagnetic jerk is an abrupt change in secular acceleration: the slope of the field’s rate of change changes. Aubert and Finlay reproduced jerks through fast hydromagnetic wave packets focusing near the core surface. This supplies a mechanism for sharp, localized magnetic changes without a global polarity transition. A jerk can still matter for model forecasts and field geometry; its presence alone does not diagnose excursion onset.¹⁰⁰

The recent evidence also limits an overly confident recovery story. ArchKalmag8k, published in 2026, finds the modern SAA more pronounced than the selected ancient anomalies in its comparison, with wide uncertainty and very different configurations among early ensemble members. Its results do not establish uninterrupted modern-style anomalies sweeping around the planet. The older suggestion that the SAA could fade within a few centuries remains an analogue-based possibility—not a scheduled recovery, and not something on which to base reassurance.^96,49

Shorter-looking intervals need a consistent clock and a complete catalogue

Core-wave bands, jerk dates, excursion ages, solar-cycle phases and a star’s Galactic trajectory do not measure the same process. The core studies discussed above identify multiple bands, while the six-year rotation–climate connection remains an open coupling problem. Selecting a run of shorter gaps from a mixed catalogue can manufacture apparent acceleration. A changing detection threshold can do the same when newer observations resolve events that older records miss.^37,38,46

A defensible contraction claim must survive one event definition, one detection procedure, uncertainty in ages, regional versus global classification, the changing observation window and a suitable non-periodic comparison. Periods discovered after trying many bands need a multiple-testing correction. The next interval must then be forecast before it is observed. This assessment does not derive a contraction law from the small list of named excursions, or a near-term hazard multiplier from increasingly detailed satellite observations.

Below Holocene highs, above the 100,000-year mean

GGF100k provides a long reconstructed context, not a universal normal state. Its displayed time series averages 6.827 × 10²² A m² over 100,000 years, while the 2025 total dipole is 7.690. The Holocene-high comparisons above are appreciably less reassuring than that mean-relative statement. Both belong in the assessment. A present value far above the deepest Laschamps minimum does not, on its own, rule out an early trajectory toward a weak-field interval.^92,51

What would “onset conditions” actually mean?

There is published support for asking this question quantitatively. Knudsen and colleagues suggested an empirical dipole level near 4 × 10²² A m² below which excursions and reversals occur in their reconstruction. That is not a universal bifurcation boundary. Their quantity is an axial dipole estimate; the matching 2025 IGRF axial moment is 7.591, derived from g₁⁰, rather than the total dipole’s 7.690. The present axial value is about 1.90 times that proposed level and would have to fall a further 47.3% to reach it. This comparison gives scale, not a waiting time. The proposed level does not identify the earliest precursor, and remaining above it is not an all-clear signal.^93,118

Gwirtz and colleagues make the distinction explicit: their low-dipole event begins below 10% of the long-term mean and ends above 80%; the optimized warning threshold is a different quantity. Forecast skill varies across dynamo models and paleomagnetic reconstructions. Neither 10%, 80%, nor a percentage lost from the Holocene maximum can be borrowed as a ready-made countdown for today.⁹⁷

The strongest warning would not be “today already looks like the Laschamps minimum.” It would be persistent global dipole erosion accompanied by spatial reorganization that demonstrably approaches a pre-transition state. The strongest counterexample is substantial weakening followed by recovery. Buffett’s 2024 analysis treats ordinary variation, excursions and reversals as potentially different expressions of shared dynamics: how long a trend persists matters. A natural origin does not make an extreme outcome impossible; a short natural trend does not establish its continuation.⁹⁸

Being closer in amplitude is not the same as being close in time. The high-to-present comparison increases the measured loss. It does not supply the missing dynamical boundary, present distance to that boundary, or conditional transition probability. Those must be estimated using matched spatial resolution, age uncertainty and non-transition controls—not only an attractive overlay with one past event.

What an excursion looks like

The Laschamps excursion, about 41,000 years ago, is the best-documented weak-field event. In LSMOD.2 it is governed mainly by axial dipole decay and recovery: the axial dipole falls almost to zero but does not reverse. The underlying process lasts about 5,000 years, from 43 to 38 thousand years ago, and its surface expression about 2,000 years. The following Mono Lake interval may contain several excursions between 36 and 30 thousand years ago, with dipole and non-dipole power at the core surface becoming comparable. This is a description of that model, not a rule that every reconstruction must share.⁵²

Evaluating LSMOD.2 directly reproduces that behaviour. Its axial dipole coefficient never changes sign over 49.9–30.05 thousand years ago, and its dipole moment bottoms out at 0.39 × 10²² A m², about a twentieth of today’s. At that minimum the model’s average surface field is about 8,800 nT against about 42,200 nT for the 2025 field truncated to the same resolution, and the whole surface lies below 32,000 nT, against 21% today.^2,52 A space-plasma reconstruction of the same epoch puts the field near 10% of modern values, with dipole weakening and tilting over about 300 years, and shows auroral regions and open field lines expanding and wandering toward low latitudes.⁵³

✓ Supported  During a global excursion, auroral and open-field-line exposure extends toward low latitudes.

The depth of an excursion depends on how the record is resolved

GGFSS70 reconstructs 15–70 thousand years ago from nine selected high-resolution sediment records. It places the Laschamps dipole minimum at 0.12 × 10²² A m² around 40.97 ka, and includes a roughly 300-year reversal of the axial dipole between 41.25 and 40.93 ka. LSMOD.2 does not reverse that coefficient. The models agree on a profound weak-field episode, but not on every detail of its deepest state.^52,54

GGFSS70 also reports dipole minima of 1.87 at 64.87 ka for the Norwegian–Greenland Sea event and 3.30 at 34.47 ka for Mono Lake/Auckland. Even within one reconstruction, named excursions need not reach the same depth. A single “excursion-strength field” is therefore an incomplete input to an exposure calculation.⁵⁴

How often, and how fast

Excursions occur at paleointensity minima and last a few thousand years or less. About seven are well documented and age-calibrated in the present Brunhes chron; reviews list more candidates, and some appear only regionally.^51,55,56 North Atlantic sediments record four late-Brunhes excursions in the intervals 32–34, 39–41, 180–188 and 205–225 thousand years ago, associated with the Mono Lake, Laschamps, Iceland Basin and Pringle Falls events.⁵⁷ Gaps between documented excursions range from several thousand years to several hundred thousand.

Directional change during these events can be very rapid. Comparing geodynamo simulations with a 100,000-year field model, one study finds maximum rates of directional change near 10° per year, typically while field strength is falling—almost a hundred times faster than current changes. The clearest observed example is about 2.5° per year some 39,000 years ago.⁵⁹

Intensity rates are harder to compare. The modern dipole decline, about 0.5 × 10²² A m² per century averaged since 1900, is close to the steepest decline in LSMOD.2 (0.60 per century, near 41.1 thousand years ago) and above the steepest in GGF100k (0.47). Both paleomagnetic models smooth rapid change, so their rates are resolution-dependent smoothed estimates and the comparison with instrumental data is not like for like.² The published stochastic assessment, which accounts for that variability, finds the modern rate within natural variation.³⁰

Reversals: rare, slow, and irregular in rate

The last full reversal, from the Matuyama to the Brunhes chron, is dated at 773,000 years ago in the 2020 Geologic Time Scale.⁵⁸ A synthesis of lava flows, ocean sediments and Antarctic ice cores finds that it took at least 22,000 years to complete, with a final directional reversal of under 4,000 years preceded by a long interval of instability.⁶⁰

The polarity record shows that reversal frequency itself changes over tens of millions of years. Counting the boundaries in GTS2020 gives under one reversal per million years just after the Cretaceous Normal Superchron ended 83 million years ago, rising to about 4.4–4.8 per million years over the last 20 million. Over the last 10 million years the mean polarity interval is about 196,000 years and the median 132,000.^2,58

The present chron, at 773,000 years, is longer than 46 of the 47 earlier complete polarity intervals of the last ten million years. That makes it unusual, not demonstrably overdue. Those intervals are irregular, reversal rates change through geological time, and a fitted interval distribution does not by itself establish a memoryless process or a future hazard rate.^2,58 A published stochastic assessment conditioned on a present axial dipole of 7.6 × 10²² A m² gives under 2% probability of being in the reversed state after 20,000 years and 11% after 50,000 years. Those are results of that paleomagnetic model, not a universal clock.³⁰

✕ Not supported  The field is about to reverse.

Claimed consequences of weak-field intervals

A weaker field admits more cosmic radiation to the atmosphere, raising the production of cosmogenic isotopes such as beryllium-10, which is how ice cores record the Matuyama–Brunhes reversal.⁶⁰ Beyond that well-established chain, several larger claims remain contested. One study links the Laschamps excursion to a global environmental crisis about 42,000 years ago, including atmospheric ozone changes and megafaunal and human effects; the paper drew a published critique in the same journal.^61,62 The space-plasma reconstruction notes that ochre use, tailored clothing and cave occupation appear in regions its model places under the expanded auroral zone, and presents that as a fresh perspective rather than a demonstrated cause.⁵³

? Unresolved  The Laschamps excursion caused a global environmental crisis or shaped Neanderthal and human outcomes.

A separate claim connects the field and the Sun to modern climate through cosmic rays seeding clouds. CERN chamber experiments establish ion-enhanced particle nucleation, but cloud-relevant changes also depend on particle growth and survival. The evaluated solar-cycle effect on cloud condensation nuclei is small; independent aerosol modelling finds a response about two orders of magnitude too small to explain the cloud changes at issue. That constrains a proposed explanation of present warming without denying the underlying ionization chemistry.^63,64

✕ Not supported  Changes in the geomagnetic field or cosmic-ray flux are a significant driver of present warming.

Four records of “extreme activity” are not four estimates of the same risk

Occurrence rates only become comparable after defining the event, the observing window and the detection threshold. A magnetic storm, a particle shower and a radiocarbon spike overlap in their solar origins, but they are not interchangeable outcomes.

In 67 ground-level enhancements, Owens and colleagues found occurrence approximately four times more likely around solar maximum than minimum, with additional cycle structure. Their result does not turn the post-maximum Kp pattern into a particle-event forecast. The severe event definition—and not simply the word “solar”—determines the relevant cadence.⁶⁵

For radiocarbon, Usoskin, Miyake and Golubenko show that background solar modulation and carbon-cycle response can imitate or obscure smaller candidate events. In their model, events below roughly 0.5–0.7 of the specified reference cannot be robustly identified without knowing the background. That reference is 100 times the 20 January 2005 particle event; it is not the 1956 event used in the flight-dose example below. An archive’s apparent gaps therefore cannot automatically be read as physically quiet centuries.⁶⁶

At the stellar scale, Vasilyev and colleagues identified 2,889 superflares on 2,527 of 56,450 Sun-like stars, above 10³⁴ erg. Their approximate once-per-century population recurrence is important evidence about energetic stellar behaviour. It is not a once-per-century forecast for an Earth-directed solar particle event, still less for a catastrophic geomagnetic storm: stellar selection, radiated energy, particle acceleration and Earth interception intervene.⁶⁷

The storm record has a season and a phase

GFZ’s planetary Kp index runs continuously from 1932. Counting days whose maximum three-hour Kp reaches 8− or higher, and treating severe days separated by no more than one quiet day as a single storm, the record to 1 October 2026 holds 278 severe storms, 68 of which reached Kp 9− or higher.^2,71 About 46% begin in March, April, September or October. About 32% are followed by another severe storm within 30 days, compared with about 30% in the season-and-cycle comparison. That particular comparison explains most of the observed follow-on fraction; it is not a test of every possible source of event dependence.⁷¹

Cycle 25’s 13-month smoothed sunspot number peaked at 160.9 in October 2024.⁷² By 1 October 2026 it had produced 13 severe episodes, four reaching Kp 9− or above: May and October 2024, November 2025 and January 2026. That is fewer than cycles 17–23 had at the same elapsed cycle time (18–45) and more than cycle 24, which had six. Six of cycle 25’s 13 occurred after maximum. These are observed counts, not a prediction of how the rest of the cycle will unfold.^2,71

The declining phase is not a guarantee of quiet. The follow-on analysis did not detect a material excess beyond its season-and-cycle comparison. That result is specific to its event definition and null model, not proof that solar storms never cluster through additional processes.⁷¹

The declining phase is longer. Count the years as well as the storms.

Across completed solar cycles 17–24, 69.2% of severe storm onsets occurred after maximum—but that phase occupied 62.5% of the observed time. Dividing by duration gives 3.38 storms per year after maximum against 2.51 before: a pooled rate ratio of 1.35, not the 2.25 ratio of raw counts. Only four of the eight cycles have a higher post-maximum annual rate.³

The result becomes less pronounced for the Kp ≥ 9− subset: 0.78 versus 0.68 episodes per year, a pooled ratio of 1.14. Resampling whole cycles gives broad 95% intervals—0.96–2.37 for severe storms and 0.66–2.91 for the extreme subset—both spanning equal rates. With only eight cycles and substantial between-cycle variation, the useful conclusion is continued exposure after maximum, not a universal post-maximum escalation.³

The May 2024 storm as a reference event

The storm of 10–11 May 2024, the Gannon storm, reached a Dst minimum of −412 nT, the seventh largest by Dst since 1957, and a one-minute SYM-H peak of −518 nT, the second largest since 1981 after March 1989. Its integrated recovery-phase disturbance exceeded those of the 1989 and 2003 storms.⁷³ The heated, expanded thermosphere accelerated orbital decay: one tracked Starlink satellite reached reentry about eleven days earlier than predicted before the storm, according to a preprint that has not been peer reviewed.⁷⁴ The storm also tested grid operations, which the recovery analysis examines.

Geometry and particle flux move independently

A six-year CSES study finds a double-peaked distribution for 2–10 MeV protons and a single peak at 10–20 MeV. Its proton-defined SAA area shrank by about 6.1% over 2019–2024, while studies of the magnetic field document a growing weak-field area over a different period. Different energies, thresholds and observables draw different boundaries, so a magnetic footprint can grow while a particle footprint contracts.^28,35

Field-line tracing turns the map into an orbital screen. In the internal-field calculation, the zone whose minimum mirror altitude stays above 200 km covers 7.6% of the 450 km shell in 1950 and 6.2% in August 2026, a contraction of about a fifth, drifting west while the surface weak-field area grows. Truncating the field at degree two makes this screening zone about 2.6 times larger. The result identifies sensitivity to detailed geometry; the 200 km threshold is a screening convention, and particle lifetime and dose require transport and atmospheric-loss calculations.^29,36

◐Reconstruction: internal field, traced here

The long-lived zone at 450 km has contracted

The particle signal combines source and loss

Trapped-particle populations follow a balance, dN/dt = Q − νN: the population N changes through its source Q and loss rate ν. With the source fixed, slower loss raises the steady population. A record of N alone cannot separate the two, because any admissible loss history pairs with the source Q = dN/dt + νN.⁷⁵

An exact construction makes this concrete: the same 11-year, ±5% population cycle arises from reference loss times of one year and ten years, each with a positive source and a positive loss rate. Particle lifetimes are identified from independent information, such as several energy and pitch-angle channels, measured injections or direct loss measurements, rather than from the amplitude or lag of the solar-cycle response.⁷⁵

Climate change also reaches orbit through the upper atmosphere. Greenhouse-gas-driven cooling contracts the thermosphere and reduces drag, so debris stays in orbit longer; modelling finds substantial losses of sustainable satellite carrying capacity under specified emissions scenarios. Storms act the other way on short timescales, heating and expanding the thermosphere and raising drag for days.⁷⁰

Weak-field exposure is a geography, not one global multiplier

A 2026 Laschamps reconstruction traces particle access through the full non-dipole field, then calculates atmospheric radiation. The modeled area with cutoff rigidity below 1 gigavolt grows from 13.6% for 2020 to a maximum 48.3% at 41,050 years before present. Lower cutoff admits lower-rigidity particles, but route, altitude, spectrum and field geometry still determine exposure.⁴⁴

Three layers of protection—and different ways to lose margin

Arsenović and colleagues model an extreme solar particle event under several magnetic conditions. In the no-geomagnetic-field boundary case, they obtain a 20–25% increase in UV index and a 40–50% increase in a solar-induced DNA-damage proxy, with elevated surface UV lasting up to six years. These are modeled relative changes in a deliberately extreme scenario; the biological weighting is not a measured increase in disease, and no field is not today’s field—or the full structured Laschamps field.⁴⁵

The important co-factor is the conjunction: magnetic configuration × particle spectrum × atmospheric response. It can produce a prolonged consequence from a short forcing event. It does not show that an excursion caused a specific extinction, that a modern anomaly is equivalent to a missing field, or that particle effects explain the present warming trend.

Ground impacts can now be traced from physics to economic exposure

A September 2026 U.S. study couples geoelectric forcing, grid assets, transformer thermal vulnerability and economic activity. Its 250-year-return-period scenario affects about 5.1 million people and causes about US$1.81 billion in daily economic loss, with a conditional 95% interval of US$1.65–1.96 billion.⁶⁹ The study ends where this assessment continues. Restoration duration, grid reconfiguration and backup at critical facilities lie outside its calculation, so its daily figure becomes a total only through a restoration trajectory, the subject of the recovery and continuity analyses.

A larger solar-event envelope does not require an imminent magnetic reversal

The prehistoric record expands the severity envelope beyond the satellite era. Golubenko and colleagues’ 2025 chemistry–climate reconstruction of the 12350 BC radiocarbon spike estimates an extreme solar particle event 18 ± 11% stronger than AD 775. A nearly doubled radiocarbon spike did not imply a doubled particle event: glacial atmospheric and carbon-cycle conditions changed the recorded response. Event dating and inferred particle strength are not equivalent to a prediction of the next event.¹¹⁴

The current cycle is already past its smoothed sunspot maximum: SILSO identifies October 2024, at 161. A separate SIDC analysis of cycles 21–24 finds strong X5+ flares concentrated in some post-maximum years, including the equal-duration bin represented by May 2026–April 2027 in cycle 25. This is relevant to continued vigilance. It measures soft-X-ray flare timing—not the rate of Earth-directed ejecta, ground-level particles or grid-damaging storms. The exposure-normalized Kp analysis above therefore remains a separate result.^115,116

The physical branches can move in opposite directions: solar magnetic activity modulates incoming galactic cosmic rays while flares and ejecta create their own radiation and magnetic disturbances. A quieter phase in one channel need not quiet another. Event spectrum, magnetic connection to Earth, arrival direction and duration matter more than a single generic label of “solar energy.”^44,65,66

A weaker field can intensify currents despite receiving less energy

Ebihara and colleagues’ MHD experiments show why ionospheric conductivity is a consequential co-factor. Lowering the model dipole from 7.76 to 5.17 made the peak westward-electrojet proxy about 24% stronger with their conductivity scaling—but about 20% weaker without that scaling. The model excludes quadrupole structure and simplifies precipitation physics. This is a response calculation, not a measured modern trend or an outage forecast.⁹⁴

The important inference is that reduced field strength need not mean reduced electrical disturbance simply because a smaller magnetosphere intercepts less energy. A changed coupling can move a particular response in the other direction. An infrastructure operating limit can therefore be crossed without Earth’s dynamo approaching a reversal threshold.

The outer shield responds to its surroundings

The heliosphere is the solar-wind cavity in the interstellar medium. Its size and particle filtering depend on the external gas, magnetic environment and the Sun’s own outflow. Opher’s 2026 review treats changing Galactic surroundings as a potentially important environmental influence over geological time. Repeated encounters are physically credible; recurrence on million-year scales is not a phase-locked millennial clock, and neither establishes a 2026 arrival date.¹¹⁷

The recent interstellar environment is not perfectly constant

A May 2026 Antarctic-ice study provides direct material evidence. Koll and colleagues measured supernova-derived iron-60 in 295 kg of ice spanning 40–81 thousand years ago. The inferred deposition, 0.22^+0.14_−0.12 atoms cm⁻² yr⁻¹, is lower than recent Antarctic-snow and younger sediment estimates, supporting a change in the interstellar material encountered by the Solar System. The paper summarizes a possible exit from the Local Interstellar Cloud within the next 2,000–6,000 years and discusses possible mixing with the G-cloud at our current location. This is a changing-influx result, not a measurement of an ongoing cosmic-ray surge or a new nearby supernova.¹⁰⁷

The local-cloud geometry is itself under discussion. Linsky and Redfield review competing descriptions of nearby warm gas, including discrete clouds and continuous structure. A boundary can alter the heliosphere, but “near a cloud boundary” does not specify its density contrast, crossing date, thickness or radiation effect. Those are the quantities needed to distinguish a modest environmental change from a severe compression.¹⁰⁸

Vannier and colleagues’ 2025 Hubble/STIS study strengthens the case for local environmental variability. Eight sight lines along the Sun’s historical trajectory reveal small-scale structure and support changing density in its recent past and near future. The authors discuss a possible present LIC/G-cloud interface and an estimated LIC exit in less than about 3,000 years. They detect no cold, dense material along their observed sight lines. That non-detection is spatially limited, not a complete census of every nearby cloud.¹¹⁹

A cold-cloud encounter could transform exposure dramatically

Opher, Loeb and Peek’s 2024 simulation asks what happens if the Sun meets a cold, dense cloud of the specified type. It places the upwind heliopause at about 0.22 ± 0.01 AU, inside Earth’s 1-AU orbit, for a proposed encounter 2–3 million years ago. This is a serious physical mechanism for losing much of the heliosphere’s usual filtering. In that calculation the interstellar magnetic field is neglected because gas ram pressure dominates: calling the result a demonstrated “Galactic magnetic pulse” would misidentify the driver.¹⁰²

The 2026 extension adds particle acceleration, not only reduced filtering. In its compressed-heliosphere scenarios, Opher and colleagues obtain heliospheric energetic particles with sub-10-MeV proton intensities at least an order of magnitude above extreme modern solar-particle events, with exposure lasting months along relevant orbital segments. Outside the compressed cavity, sub-GeV galactic protons are also enhanced. These are conditional spectra outside Earth’s magnetosphere, not a calculated tenfold increase in surface dose. Their atmospheric significance depends on energy, geographic access and duration.¹⁰³

The past-event attribution is actively contested. In April 2026, Koll and colleagues argued that cosmogenic nuclides do not show the prolonged large cosmic-ray increase expected from a strong 2–3-million-year-ago exposure. Opher and colleagues replied that encounter duration, density structure and latitude-dependent isotope production can change what survives in the geological record. The disagreement is about a testable physical history—not a reason to treat the mechanism as either established history or impossible.^104,105

Other astronomical messengers must retain their own histories

The 2026 iron–plutonium–curium study separates supernova-material delivery from older r-process material: the iron-60 and plutonium-244 profiles are not simply correlated. A 2025 marine beryllium-10 anomaly near 10 million years ago admits production and ocean-transport explanations. Neither record independently supplies a recurring near-term catastrophe date. Deposited dust, cosmogenic production and a burst of radiation are different observables, with different transport and dating filters.^109,110

There are also published Galactic-centre hypotheses. LaViolette’s 1987 paper proposes cosmic-ray volleys and dust-related environmental consequences; its existence should not be erased from an exploratory assessment. Establishing a present episode would require a source history, propagation model and independently dated terrestrial signatures that distinguish it from solar modulation, geomagnetic change and archive effects. The primary abstract examined here does not supply a calibrated current-arrival forecast.¹¹²

Likewise, the observed oscillation of the Radcliffe Wave concerns the motion of a large Galactic gas structure, not detection of a coherent electromagnetic front approaching Earth. A spatial pattern becomes an exposure history only after calculating the Sun’s path, the structure’s evolution and their encounter uncertainty.¹¹¹

Direct astronomical effects on Earth are demonstrable. Piersanti and colleagues linked the 9 October 2022 gamma-ray burst GRB 221009A to an upper-ionospheric electric-field perturbation. That is evidence for an atmospheric response to a distant event. It is not evidence that such a pulse penetrated to the core or triggered a geomagnetic excursion. Photons, charged particles and interstellar dust cannot be merged into one mechanism merely because all originate beyond Earth.¹¹³

Published millennial cycles deserve evaluation—not automatic attribution

Beer and colleagues’ 2025 analysis identifies persistent bands in a 145,000-year Antarctic beryllium-10 record and reports close period agreement between its first and last approximately 70,000 years. The authors describe 17 new solar-activity cycles, but explicitly leave potential geomagnetic origins unresolved because paleomagnetic reconstructions disagree. Their separation of production from archive effects is essential: a period in an isotope record is not necessarily a period in external radiation.¹⁰¹

Nilsson and colleagues’ 2024 joint reconstruction illustrates why attribution can change: hemispheric geomagnetic evolution alters polar cosmic-ray proxies differently, and the approximately 2,400-year Hallstatt signal can be strongly affected by that separation. This does not settle every longer band in the 2025 analysis. It does mean that solar histories corrected with a field model cannot then be treated as wholly independent evidence confirming that same magnetic history.⁹⁵

Three claims that require three different tests

The same caution applies to “timing contraction.” A changing period, an increasing number of detected events, and a falling time-to-failure are different hypotheses. A convincing case would predict the next interval and the affected observables before seeing them. Mixing a millennial cosmogenic band with year-scale core waves and day-scale solar disturbances does not perform that test.

A weak perturbation can tip a nonlinear system—but only through a coupling

Rejecting every weak external driver because it cannot supply the whole energy of a transition would be too crude. A perturbation can trigger the release of internally stored energy, shift a near-threshold system, or synchronize an existing oscillator. The opposite shortcut is equally unsound: invoking “resonance” or “tipping” does not establish that a particular forcing couples efficiently to the core, ocean or atmosphere.

A useful evaluation identifies the affected state variable, the coupling, its gain and phase, the relevant relaxation time, and the boundary separating recovery from transition. It then asks whether observations constrain those quantities today. An external-to-core proposal needs an induction or torque pathway through the intervening layers; an external-to-ozone proposal needs particle spectra and chemistry. Demonstrating the second does not establish the first. Surface calm cannot prove a deep system is safe; an unseen threshold cannot prove it is near.

The influence map: where each factor earns its place

These pathways are not equally established, equally severe or independent. The map separates measured conditions, conditional response calculations and unresolved attribution. It is an evidence map, not a summed risk score.

The most consequential combination is an interaction, not a pile of percentages

Consider a strong energetic-particle spectrum meeting a geographically altered magnetic field and a vulnerable atmospheric state. That can be more consequential than changing any one factor alone. At the same time, a high-latitude route already open to a particle population can change little while lower latitudes change much more. The route-specific weak-field results above demonstrate why “global shielding down by X%” is an inadequate exposure model.⁴⁴

The proposed compressed-heliosphere/weak-geomagnetic-field combination is therefore worth explicit calculation. Its conditional severity cannot substitute for evidence that the Solar System is currently entering the required cloud. Separately, an extreme solar event can create a major operational shock under today’s field. It does not need a reversal, an interstellar encounter or a planetary synchronization to matter.^103,106,45,6

The serious interpretation is asymmetric. The baseline correction increases the long-term magnetic decline that must be acknowledged. New space-environment work enlarges the set of physically plausible severe scenarios. Neither supplies a calibrated imminent global tipping probability. Present exposure, potential conditional severity and event timing must remain three separate conclusions.

During Tropical Storm Imelda in 2019, more than 95% of outages farther than 30 km from flooded roads were repaired within ten hours, while 40% of those within 5 km remained unrepaired after ten hours. Spatial and prior-period controls point to localized impairment from flooded access rather than a uniform shortage across the service area.⁷

The mechanism produces three distinct conditions: a busy but recoverable repair queue; a queue whose effective capacity declines as damage accumulates; and a system that recovers but misses the deadlines of the services it supports.

Nominal capacity can conceal a shrinking recovery margin

A minimal model shows the second condition. Let B be unfinished repair work, a the continuing arrival of new work and c nominal repair capacity. Let b set how quickly throughput rises when work is available, and K the backlog at which damaged support begins to constrain throughput:

The last factor expresses damage-dependent impairment of repair. Without it, throughput rises toward c; with it, effective throughput peaks and then falls. When incoming work stays below that peak, the model has a stable low backlog and an upper recovery boundary, beyond which repair cannot catch up under unchanged conditions.⁷⁵

With a = 8, c = 20, b = 10 and K = 100, the stable backlog is 7.55 work units and the recovery boundary 132.45. A 20% loss of capacity moves the boundary to 77.02 and cuts the margin between equilibrium and boundary from 124.90 to 64.03 work units: a 20% loss of capacity removes 49% of the margin.⁷⁵

Disturbances also interact through their timing. A 100-unit work shock alone is recoverable, and so is a 30-day reduction of capacity from 20 to 16. Together they leave about 151 units of unfinished work when the impairment ends, above the original boundary, so restoring ordinary capacity no longer reverses the trajectory. Extra crews or protected support during days 30–90 do.⁷⁵

The severe implication is direct. Recovery becomes harder when recovery itself loses support. Where that feedback operates, protecting access, communications, fuel and repair logistics changes the course of the emergency, not merely its initial damage.

In Louisiana after Hurricane Ida, a published assessment counts about 35 million customer-hours of simultaneous power outage and heat exposure above an outdoor heat index of 37.8°C (100°F). A separate three-city simulation found that multiday blackouts during heatwaves more than doubled modeled heat-related mortality rates in Atlanta, Detroit and Phoenix.^76,77

Restoration time and backup endurance have to be paired

Let T be time until normal service returns and A the usable autonomy of a substitute. Under full substitute service until A, followed by full restoration at T, uncovered time is:

Usable autonomy reflects actual load, equipment condition and available replenishment. Nameplate energy alone does not specify it; intermittent service and repeated refuelling require a time-resolved service trajectory. In a four-site example, restoration times and autonomies are both 1, 2, 4 and 8 days. Each averages 3.75 days under every arrangement. Changing only the pairing moves mean uncovered service from zero to 2.25 days per site, and the share with any gap from zero to 75%.⁷⁵

The arrangement that interrupts the most sites differs from the one that loses the most time: cyclic mismatch interrupts three of four sites for an average of 1.75 days, opposite pairing two of four for 2.25 days. Counting failures and integrating their duration answer different questions.

When the pairing is unknown, let F_A(t) be the fraction whose autonomy has run out by time t, and F_T(t) the fraction restored by then. The sharp bounds on mean uncovered duration are:

Matching equal quantiles attains the lower bound; matching opposite quantiles attains the upper. A positive lower bound proves a continuity shortfall even under the most favourable arrangement. These bounds apply to a specified population and operating condition with finite mean durations.^75,78 A simpler deadline guarantee requires no dependence assumption. At a common deadline d, Pr(T > A) ≤ Pr(A < d) + Pr(T > d). If no more than 2.5% of sites have autonomy below 48 hours and no more than 2.5% take longer than 48 hours to restore, at most 5% have an uncovered interval in this model. This is a design condition, not a probability estimated from the hurricane counts.⁷⁵

Resupply has a deadline, not just a total

Under uninterrupted operation, usable energy must remain sufficient at every time before restoration. With initial usable energy E₀, cumulative usable deliveries D_E(t) and required power p(t), the energy condition is:

In an exact example, a site starts with 24 energy units, uses one per hour and regains normal supply after 48 hours. A 30-unit replenishment at hour 12 preserves service; the same delivery at hour 30 leaves a six-hour interruption. Total supply is 54 units in both cases, exceeding the 48-unit requirement. Delivery timing alone changes continuity. The example assumes adequate storage and discharge power and no other failed dependency.⁷⁵

Fuel inventory, battery capacity and logistics have to be evaluated against the next exhaustion time. If required load rises, the same energy lasts less time. A communication link or pump can also remain unavailable with ample energy when another necessary component is down.

Helene left a long power-related service gap

FCC disaster reports count cell sites off the air by reported cause—power, backhaul transport or physical damage—and, for Helene, sites operating on backup power. Here the power-observed subset means sites reported off for power plus sites operating on backup. Its conditional power-off share is P/(P + B), where P and B are the corresponding counts. Sites classified under transport or damage are outside that denominator, even when they also lack grid power.^4,79

On 28 September 2024, roughly a day after landfall, the conditional power-off shares were 89% in western North Carolina, 64% in South Carolina and 62% in Georgia. By 4 October, about a week after landfall, they were 51%, 15% and 18%. The North Carolina share reached 40% on 7 October. These counts show prolonged unmet power needs; a power-coded outage alone does not identify whether backup was absent, exhausted, unable to start or otherwise insufficient.⁴

Access was severely impaired during the same emergency. NCDOT reported about 390 road closures, including more than 60 on primary routes, on 28 September. CRS describes damaged fibre and obstacles to reaching and resupplying cell sites. Those records establish a credible access mechanism alongside the outage pattern. They do not isolate the proportion of the regional contrast caused by roads, fuel, backhaul or the initial intensity of damage.^82,83

Exposure reverses the apparent ranking of performance

For the same fixed Helene panels and the same nine elapsed days, the power-off burden separates exactly into two factors. Let P_d be power-caused outage site-days, G_d power-observed site-days, N panel sites and W the observation window:

Georgia’s conditional uncovered share was lower than South Carolina’s—36.8% against 43.0%—but its power-observed exposure was 3.56 times greater. Consequently, its power-off time per panel site was 3.05 times greater. Western North Carolina’s burden was 4.05 times South Carolina’s, combining 2.53 times the exposure with 1.60 times the conditional uncovered share. These are exact decompositions of the reported integrated totals, not estimates of a causal treatment effect.⁴

Conditional performance and total burden must travel together. A lower failure share during exposure can coexist with more lost service if exposure grows faster. Conversely, a worsening conditional share can coexist with less total lost service when exposure falls.

Ida provides a different, complementary observation

After Ida, Louisiana’s power-caused site outages fell 65.9% between the first report and four days later, while the site-weighted customer outage share fell 21.8%. Power-related wireless availability improved faster than the customer-grid proxy. Over 30 August–14 September 2021, the record contains 3,080 power-caused site-days against 12,733 estimated exposure site-days: a proxy uncovered share of about 24%. Its denominator and 15-day window differ from Helene’s direct category counts.^4,80,81

The county rearrangement analysis also constrains the interpretation. Observed totals lie 0.45 and 0.50 of the way between the most and least favourable pairings for Ida and Helene, versus 0.47 and 0.43 under independent pairing. At that resolution, the evidence does not establish strong adverse pairing as the dominant explanation. The large regional differences remain to be explained by matched hazard, access, equipment and restoration records.^4,78

Service resilience is therefore the fit between exposure, usable protection and restoration deadlines, for the people and functions that must be sustained under the conditions the hazard itself creates.

In the repair model, any finite increase in c or K leaves dependent throughput falling toward zero at large backlog. Such increases enlarge the recoverable region, and an upper boundary remains. Add a separately supported repair capacity e that shares the low-backlog availability factor but not the damage-dependent support factor:⁷⁵

At large backlog, R_total approaches e. If e is maintained above the continuing arrival rate a, the model has a single positive equilibrium and no upper boundary: every finite backlog returns to that equilibrium after an isolated shock, whatever its size. With a = 8, adding e = 10 turns a 500-unit backlog from continued growth into recovery toward 3.76 units. A 20% increase in dependent capacity and a doubled support scale both leave that backlog above their boundaries, where it keeps growing. The interactive model above shows each case through its presets.⁷⁵

Portable generators, delivered fuel and satellite or microwave backhaul can bypass particular local failures. Their availability still depends on their own supply and access. In a complementary operator account from Beryl, AT&T reported more than 200 generators deployed by 10 July 2024 and more than 98% of its wireless network operating normally by 13 July, while crews continued refuelling. That is one operator’s operational account, not an all-carrier comparison with Helene.^83,84

Preserve recovery. Bridge the wait. Secure the supply line.

Repair support changes how and when normal service returns. Service backup changes what happens before that return. Both draw on access, fuel and communications that a single hazard can remove at once, so a complete protection strategy secures the supply line that keeps backup running, alongside the backup itself.

Locally independent support can still be regionally scarce

The independent-channel result requires e to remain available. A crew, generator, fuel tanker or replacement component already committed elsewhere cannot be counted again. Independence from local grid damage is different from independence from regional competition. A two-region allocation makes this exact.⁷⁵

The corresponding feasibility test checks every set of dependent destinations: their required work must fit within the capacity that can actually serve that set before the deadline. Under divisible work, known routes and a common deadline, this is an exact flow-allocation condition. Travel delays, specialist skills and task precedence make the real constraint tighter. A regional reserve total alone cannot pass the test.⁷⁵

Local protection also has to be coordinated with the network

NERC’s review of a July 2024 Eastern Interconnection disturbance identifies about 1,500 MW of data-centre load transferring away from the grid to backup systems after voltage disturbances. Frequency and voltage rose; operators corrected voltage, and neither rise reached a level posing a reliability risk in that event. Protective behaviour that sustains individual facilities changes the network seen by everyone else, so transfer and reconnection dynamics belong in the resilience assessment.⁸⁵

Protection has to reach people, not just facilities

A mobility study of 101 cooling centres during Beryl found more visits from census block groups closer to the centres and from lower-income areas. It measured area-level use rather than individual heat dose or avoided deaths. The planning implication is to measure reachable, operating and usable refuge—not simply count designated buildings.⁸⁶

Stable dynamics can still produce intolerable transients

The same distinction appears without any recovery boundary. In a two-variable linear model, an upstream disturbance decays over ten days and the downstream service relaxes over two days. Quadrupling the directional coupling raises the peak downstream deficit from 6.7% to 26.7% and the cumulative deficit fourfold, while both decay rates stay unchanged. Both systems recover; one inflicts four times the harm on the way.⁷⁵ Long-run stability and transient amplification are separate properties, and work on stochastic systems likewise separates recovery within a state, escape from it and the observable used to assess either.⁸⁷

Operational adaptation is an observed counterweight

During the May 2024 Gannon storm, New Zealand used targeted transmission-line disconnections to cut induced currents at key transformers while maintaining supply. A monitored Dunedin transformer peaked at about 113 A, against roughly 200 A modeled without mitigation, and consumers kept their supply through the storm.⁸⁸ The relevant comparison is the same hazard under different feasible operating states. Heat, unavailable equipment or reduced reserve can constrain a protective configuration; preparation and alternative paths preserve it.

This expression distinguishes a brief outage from a prolonged one and basic function from nominal asset availability. Its units follow the weights. Heat-weighted customer-hours, unserved electricity and loss of communications remain separate outcomes unless an explicit valuation connects them.^75,76 A useful compound comparison holds the service, time horizon and operating information fixed. Let L₀ be ordinary loss, L_H loss under environmental or repair stress, L_S loss under a storm and L_HS loss under both. The interaction is:

A positive value means the joint loss exceeds the sum of the separate increases on the chosen scale. A zero interaction can accompany catastrophic additive losses, and the loss unit matters: unserved megawatt-hours, customer-hours, heat-weighted exposure, repair work and economic cost measure different harms.^8,75