Soils, agricultural boundaries, and the risk of global tipping

20 October 2026

The important question is whether the systems that feed people can lose the capacity to keep doing so—and whether apparently successful production can give too little warning. Soil organic carbon matters within that question because it participates in nutrient supply, soil structure, biological activity and exchanges with the climate. A carbon concentration, however, is only one measurement of a system whose performance also depends on water, soil depth, mineral composition, management and access to resources.

The existing collection provides a substantial basis for investigating functional limits. It includes sustained production despite changing soil conditions, soils that respond poorly to additional fertilizer, slow recovery after depletion, erosion that removes the productive resource, and climate events affecting several producing regions together. These findings can be connected into arguments about agricultural vulnerability. They do not yet establish the strength of every connection, a global carbon threshold, or a date of worldwide collapse.

The useful inference is therefore broader than a correction to projected crop yields: agricultural risk should be assessed through the functions that sustain production, the support needed to maintain them, and the ability to recover when conditions worsen. This makes global boundaries a legitimate research question while leaving their form and location open. The review below distinguishes findings reported by sources from the further reasoning needed to connect them. Its evidence base is the inherited collection, with targeted readings; it is not an exhaustive search of all Earth-system tipping literature.

1. What would agricultural failure mean?

Failure needs an explicit object and timescale. A poor harvest may be followed by full recovery. A field can produce a satisfactory harvest while losing soil. A grazing landscape can retain vegetation while shifting toward a state that supports fewer animals. A farm can cease operating because restoration is unaffordable even when biological recovery is possible. These outcomes have different mechanisms and implications for feeding people. The IPCC land-degradation assessment treats resilience as the ability of interacting ecological and social systems to retain essential functions through disturbances; it also notes that judgments about successful adaptation depend on whose functions and livelihoods are preserved. Olsson et al., 2019, pp.44–45 DOI

For this review, productive capacity means the ability to sustain useful food or feed production through time under specified environmental conditions and feasible management. The definition includes reliability and recovery, as well as average output. A soil-function boundary concerns a process such as water entry, rooting or nutrient supply. An agricultural viability boundary concerns maintaining production with resources that can actually be supplied. Food security additionally depends on access, distribution and nutrition. A loss at one level can contribute to failure at another, but the connection must be shown rather than assumed.

An adequacy boundary is crossed when capacity cannot meet a stated essential production requirement. Collapse denotes a substantial, persistent loss from a stated reference capacity that prevents essential production under feasible management. Rising requirements can create an adequacy shortfall without deterioration of capacity. A landscape can remain biologically active in such a state; complete loss of vegetation or soil carbon is not required.

As an analytic illustration, consider a farming district after repeated droughts. Lower harvests are damage; a persistent inability to restore an agreed essential food or feed output over the specified recovery period, with the water and replenishment resources actually available, would meet this review's collapse criterion. That local loss could still be compensated elsewhere. A global capacity boundary concerns whether feasible compensation across the larger system also fails. This illustration defines a question; it is not a reported event or numerical threshold.

A tipping point is a stronger proposition than a damaged condition. Here it means a critical condition at which the processes governing change shift enough to produce a different pattern or state of behavior. A steep but otherwise continuous response curve does not, by itself, demonstrate that shift. Reinforcing processes can sustain the changed state. Different thresholds for deterioration and recovery support some tipping mechanisms, but are not required for every kind of transition. Slow or expensive recovery alone does not demonstrate tipping. A classification boundary can instead separate soils according to an indicator. A precautionary boundary can identify conditions where avoiding further deterioration is prudent. Such a boundary can be useful before a tipping point is demonstrated. The purpose of defining these terms is to keep the intended consequence visible: crossing a line on a map is consequential only through what that line tells us about function, vulnerability or recovery.

Soil depth illustrates the distinction. Evans and colleagues estimate how long topsoil would last under specified erosion and formation rates. Their chosen depth-exhaustion calculation is a physical accounting exercise, with explicit qualifications about deterioration before exhaustion and possible use of subsoil. It is not a measured universal time to crop failure. Nevertheless, the account identifies something fundamental: a harvest can continue while a slowly renewed productive resource is being depleted. Montgomery's comparison of erosion and soil-production rates supplies a wider basis for that concern. Evans et al., 2020, pp.3–5 DOI; Montgomery, 2007 DOI

The resulting inference is that present output cannot, by itself, establish the long-term condition of the system producing it. It also cannot establish hidden failure. To distinguish those possibilities, observations must connect production with changes in relevant functions, management demands and recovery. That is a tractable boundary question even where no single soil measurement separates safe from unsafe land.

2. What maintains productive capacity—and what could conceal deterioration?

Organic matter affects production through processes, not simply through the amount of carbon present. Its decomposition can release nutrients for plants, while its chemical properties help retain nutrients and its interaction with soil organisms and minerals helps maintain structure. That structure affects how water enters and how roots reach water and nutrients. The amount stored and the rate of nutrient release are different quantities: equal carbon stocks need not contain equally available material or support the same flows. Adding fertilizer can replace a shortage of particular nutrients; it does not by itself replace eroded soil or repair a compacted rooting zone. This explains both why compensation can work and why it may leave other constraints. Bulk carbon may also record management and soil history without identifying which process controls production. Janzen, 2006, pp.3–4 DOI; Moinet et al., 2023, p.6 DOI; Smith et al., local 2016 version, pp.13–15 DOI

Fertilizer and other management inputs can genuinely compensate for particular constraints. A synthesis of European long-term experiments found no significant average extra attainable yield from organic inputs when mineral nutrients were sufficient. It also reported no reduction in yield variability in the available analysis. Those results deserve their full weight: they challenge an assumption that lower organic matter necessarily compromises current performance in those settings. They do not test every degree of depletion, disturbance or resource limitation. Hijbeek et al., online 2016, pp.3,6,9 DOI

Other evidence establishes that yield and soil condition can diverge. An eight-year warming experiment in the North China Plain found improved surface-soil indicators under conservation management alongside similar wheat yields. The authors limit their suggested transfer to settings where water does not constrain production. In long-term Kenyan experiments, some treatments maintained or increased yields while soil carbon declined or remained approximately stable. The relationships depended on location and nitrogen supply. Read together, these studies show why observing either yield or carbon alone can miss part of the trajectory. Neither experiment establishes that stable output was necessarily concealing imminent collapse. Teng et al., 2024, pp.3,7–8 DOI; Laub et al., 2025, pp.6,10–12 DOI

The dominant constraint can also change. At the driest Kenyan site, the soil-carbon–yield relationship found at wetter sites did not hold, and yields declined across treatments. That finding limits the proposition that additional carbon can protect production against any level of water scarcity. It does not imply carbon is irrelevant wherever drought occurs. Protection depends on the amount and timing of water, atmospheric demand, rooting conditions and the other processes that determine whether plants can use it. A more informative question is which constraint becomes binding under a particular combination of conditions.

The mechanism of protection requires similar care. A Canadian experiment imposed drought after decades of contrasting crop rotations. More diverse rotations protected maize yield, but the measured water-retention, infiltration and rooting pathways did not explain the association with soil organic matter, and diversification had not significantly increased the total organic matter itself. A separate synthesis found that the direct increase in plant-available water per unit of added carbon was generally modest. Together these findings support soil- and management-related drought protection while challenging a simple explanation in which more bulk carbon acts mainly as a larger water reservoir. Renwick et al., 2021, pp.6,8,10 DOI; Minasny and McBratney, online 2017, pp.2–4,7 DOI

This matters for inferring boundaries. A weak average relationship between carbon and yield may reflect real substitution, a different limiting factor, an incomplete indicator, or a restricted range of conditions. A strong relationship may partly reflect management or the reverse pathway, in which productive plants supply more residues to soil. The county-scale drought analysis by Kane and colleagues explicitly cautions against interpreting its spatial relationship as the effect of changing carbon on an individual farm. Repeating such associations in several analyses does not remove that limitation. Kane et al., 2021, p.8 DOI

The evidence therefore supports a specific investigation of concealed deterioration: compare systems with similar output but different soil histories, functions and input requirements, then examine their responses to stress and recovery. It does not support treating all input-supported production as temporary concealment. In a North China Plain experiment, diversified rotations improved measured soil condition and selected production outcomes while reducing nitrogen fertilizer and irrigation requirements. These results support an alternative to increasing dependence, within the tested crops, conditions and study period. Yang et al., 2024, pp.3,8 DOI The unresolved issue is where compensation remains durable, where it becomes harder to sustain, and whether it preserves the capacity to respond when the environment changes.

3. When could deterioration reinforce itself or become difficult to reverse?

Several mechanisms could connect soil deterioration to further deterioration. Reduced plant growth can reduce residue inputs and protective cover. Loss of cover can expose soil to erosion and alter water entry. Loss of soil and nutrients can then make plant recovery harder. These links suggest a possible reinforcing cycle. Demonstrating individual links does not establish that the complete cycle dominates everywhere: management can interrupt it, other constraints can dominate, and greater plant growth can sometimes increase carbon inputs. The question is which feedback prevails under which conditions.

Long-term African maize experiments provide partial evidence for the proposed cycle and for its interruption. Couëdel and colleagues report concurrent soil-carbon and yield declines in two of four sites without external carbon and nitrogen inputs; the other two did not show significant joint decline. A follow-up using the same trials finds maintained yields with combined mineral and organic inputs at all four sites, while carbon increased, remained stable or declined depending on location. These are contrasting treatments in reused experiments, not independent confirmation of a universal feedback. They show that continuing replenishment can interrupt production decline without producing a uniform carbon trajectory. Neither analysis demonstrates a self-maintaining collapse state after the initiating pressure is removed. Couëdel et al., 2024, pp.7–8,13–15 DOI; Couëdel et al., 2026, pp.4–5,8 DOI

The strongest relevant examples in the collection include management history and recovery. Tittonell and Giller describe African long-term sequences in which depleted plots responded poorly to fertilizer and recovered only gradually after organic and mineral inputs were restored. Their Benin example records very low yields after years without replenishment, followed by a lengthy recovery that did not readily reproduce the performance of continuously supplied plots. They also distinguish poorly responsive fertile fields from poorly responsive degraded fields. The same weak fertilizer response can therefore signal very different states. Tittonell and Giller, 2013, pp.7–9 DOI

These examples support an inference about history: preventing deterioration and rebuilding after it need not require the same effort or timescale. They do not isolate carbon as the sole cause, establish that recovery is impossible, or identify a universal transition point. The practical boundary may concern how much time, material and money recovery requires relative to what farmers can sustain. An agricultural system can face a consequential limit even when its soil is not biologically sterile.

Evidence outside cropland strengthens the reason to examine persistent transitions. The IPCC land-degradation chapter describes a dryland grazing example where reduced ground cover is associated with poorer infiltration, more erosion, changing vegetation and lower productivity that requires substantial intervention to reverse. The chapter also discusses forest-regeneration failure under interacting fire and water pressures. These are assessment-level examples, drawing on underlying studies not all independently available in this collection. They establish relevant mechanisms and questions about recovery; their numerical thresholds cannot simply be transferred to crop soils. Olsson et al., 2019, p.45 DOI

Erosion adds a different form of persistence. Removing soil can diminish rooting volume and export nutrients as well as carbon. Replacing one nutrient does not necessarily replace the lost physical material or restore the previous water regime. Alewell and colleagues connect modeled erosion losses with phosphorus balances and the need for replacement inputs. Their scenario without chemical phosphorus fertilizer exposes dependence on replenishment; it is not a forecast that fertilizer will disappear on a particular date. The distinction matters because loss of productive capacity can arise through both physical depletion and constraints on maintaining the compensating supply. Alewell et al., 2020, pp.2,5 DOI

Across these cases, the useful inference is that recovery belongs inside the definition of risk. A system may remain productive in ordinary years yet recover slowly from a severe disturbance. If damaging events arrive before recovery is complete, further deterioration is a plausible pathway requiring investigation. The collection supplies components of that argument, not a calibrated global rate at which it becomes self-sustaining. Testing recovery, repeated stress and the feasibility of intervention would address the proposed mechanism more directly than extrapolating a short-term yield–carbon curve alone.

4. What boundaries can the evidence actually constrain?

The collection contains several kinds of carbon boundary. Keeping their endpoints visible makes them more useful. A limit to how much carbon minerals can retain concerns storage. A carbon-to-clay ratio can describe structural condition. A change in the response to fertilizer concerns a particular treatment and outcome. None automatically supplies the minimum carbon needed to sustain food production.

The best-developed structural example is the soil-organic-carbon-to-clay ratio. In a Swiss study, ratios around 1:8, 1:10 and 1:13 described average very good, target and poor structural conditions, respectively. But the observed classes overlapped. High ratios did not prevent mechanical damage, and some samples with low ratios retained good structure. These were shallow samples from a particular soil group, evaluated for structure rather than crop collapse. The attached correction changes the original abstract's poor-structure criterion to 1:13 or smaller; relying on the abstract alone would give the wrong number. Johannes et al., 2017, pp.3,6–7 and correction p.10 DOI

British data supported a structural gradient while rejecting the exact ratios as absolute thresholds. French tests using different structural measurements found the ratios less useful than carbon alone. These findings constrain transfer; they do not establish that structure is irrelevant or that a boundary can never be defined. Soil depth, mineral composition, acidity, mechanical disturbance and the function measured all affect what a carbon value means. Forest-soil evidence from Amazonia further shows that similar clay contents can coexist with different carbon relationships across mineralogical groups. Prout et al., online 2020, pp.10–11 DOI; Rabot et al., 2024, pp.4,11 DOI; Quesada et al., 2020, pp.13,23 DOI

A crucial disagreement concerns the reference condition. Comparing a soil with others under similar climate and texture can improve interpretation. Yet a region's current average can already reflect widespread depletion. A soil can therefore rank favorably without meeting a functional need. Feeney and colleagues' contextual benchmarks and Sauzet and colleagues' criticism, read together, reveal two separate tasks: describing how unusual a soil is, and deciding what condition should be maintained. The first cannot settle the second. A precautionary boundary can be justified by vulnerability and the difficulty of repair before collapse is observed, provided its protective purpose and uncertainty are explicit. Feeney et al., 2024, pp.3,5,11 DOI; Sauzet et al., 2024, pp.3–4 DOI

Storage ceilings answer a different question again. Estimates of mineral-associated carbon capacity concern upper limits or attainable accumulation. Experiments showing continued accumulation beyond an older estimate can challenge that estimate without establishing unlimited durable storage. Neither result identifies a lower agricultural safety boundary. Likewise, the fertilizer-response breakpoints reported by Ling and colleagues concern changes in carbon fractions under nitrogen addition, not measured transitions to failed production. Calling each a “threshold” does not make the endpoints interchangeable. Georgiou et al., 2022, pp.2,5–6 DOI; Heinemann et al., online 2024, pp.14–15 DOI; Ling et al., 2025, pp.8–9 DOI

The synthesis points toward boundaries defined by combinations of conditions: how much production, water regulation or recovery remains possible, at a given soil depth and condition, under a specified climate and feasible management. Carbon concentration, total carbon stock and carbon relative to clay each describe different aspects of those conditions. No one measure can stand in for all three. This framework preserves the possibility of preventive limits and meaningful soil-carbon boundaries, while requiring each proposed limit to say what it protects. Different local limits also leave open the possibility of a consequential global boundary.

5. How might local deterioration become a global agricultural problem?

A global boundary need not be a single carbon value reached everywhere. It could arise because productive capacity is lost across enough places, because damaging events coincide across regions, or because changes in one part of the Earth system amplify changes elsewhere. These routes have different evidentiary requirements. Local variation does not logically exclude any of them, but a map of local degradation cannot establish their global consequences on its own.

The first route is accumulation. Losses in soil depth, water regulation or nutrient supply can reduce the production that a region can sustain. Regional modeling already compares food-energy production with soil properties held constant against production when soils change under specified management. Balkovič and colleagues' European experiment makes that comparison explicit, while omitting several degradation processes and treating management and water supply through scenario assumptions. It demonstrates a way to connect soil change to production; it does not observe a global boundary. Balkovič et al., 2018, pp.9–10,13–14 DOI

Extending this reasoning globally requires more than adding degraded hectares. Their contribution to production differs; some losses can be compensated elsewhere, and some alternatives require additional land, water or nutrients. The relevant inference is conditional: a global capacity limit could be crossed when losses and recovery demands exceed the feasible capacity to compensate, even if most soils remain productive. Establishing where that happens requires accounting for the amount, location and timing of both losses and replacement capacity. The collection does not close that account. A gradual aggregate shortfall would also differ from a self-reinforcing tipping transition, although either could be consequential.

The second route is shared stress. Kornhuber and colleagues connect large atmospheric circulation patterns with concurrent below-trend harvests in several major producing regions. Their comparison finds shortcomings in simulated regional impacts and disagreement about future changes in concurrence. “Below trend” does not mean catastrophic crop failure. Nevertheless, the observed connection matters: geographical separation does not guarantee that losses will offset each other. Kornhuber et al., 2023, pp.2,5–6 DOI

Recovery provides essential counterevidence. Lesk and colleagues find that national cereal production generally recovered after the drought and heat disasters they analyzed. Their aggregate result can coexist with local damage, but it prevents treating a weather-related production loss as proof of persistent capacity loss. Lesk et al., 2016, pp.2–3 DOI

Combining that evidence with the site-level evidence about soil condition suggests a further hypothesis. Shared weather stress could encounter landscapes with differing, and possibly declining, ability to withstand it. Loss of local buffering and simultaneous exposure could then erode the practical value of geographical diversity. That interaction has not been demonstrated simply by placing the two literatures together. Its decisive missing link is whether measured soil and management histories change how often regions lose production together, how severe those losses are, and how long recovery takes under the same large-scale events. This is a stronger global question than whether average carbon predicts average yield.

The third route involves coupling: changes in land alter an environmental driver that then affects land and production elsewhere. Soil-carbon exchange with the climate provides a relevant physical connection, but its sign must be interpreted carefully. Varney and colleagues find warming-driven carbon losses in isolated model experiments while net global soil-plus-litter carbon still increases in the combined idealized experiments, where rising carbon dioxide also stimulates plant inputs. They identify stronger climate sensitivity under greater forcing and important limitations in represented processes. These results support examining feedbacks; they do not establish inevitable net carbon loss, a runaway transition, or an agricultural collapse threshold. Varney et al., 2024, pp.5–6,12–13 DOI

Forests, grazing lands and watersheds belong in this question wherever their changes affect the water, climate or ecological processes supporting food production. The assessed dryland and forest transitions justify considering persistent change beyond cropland. They do not, within this collection, quantify the strength of a forest-to-rainfall-to-crop feedback or demonstrate that several Earth-system transitions combine into a global agricultural tipping point. Those connections remain missing links, rather than reasons to exclude the ecosystems involved.

Dependence on common supporting resources provides another possible connection across all three routes. Maintaining output locally may require more replenishment or restoration. If many places need the same constrained resources at once, compensation feasible for one site may be infeasible collectively. Tittonell and Giller summarize a village-scale analysis in which manure from the local livestock population, constrained by the carrying capacity of surrounding grasslands, could cover only part of the cropland. This is concrete local evidence of a collective resource limit, reported through a review; it is not a global inventory or a constraint that necessarily binds everywhere. Tittonell and Giller, 2013, p.7 DOI The phosphorus-budget and rehabilitation evidence therefore grounds a wider conditional inference, while leaving a global failure mechanism unmeasured. The global question therefore concerns the joint behavior of productive landscapes, shared stresses, supporting resources and recovery. The strongest conclusion from this collection is that these connections deserve explicit assessment; neither a universal soil-carbon cutoff nor an observed worldwide collapse is a prerequisite for asking where protective boundaries should lie.

6. What can be inferred now, and what would reveal approaching danger?

The collection supports a change in what counts as evidence of agricultural security. Current harvests, carbon stocks and model accuracy each answer useful questions; none alone establishes the capacity to withstand worsening conditions and recover afterward. The cross-paper inference is to examine their relationship through time. Similar output accompanied by increasing replenishment needs, worsening rooting or water-entry conditions, and slower recovery would be more concerning than any one signal alone. Conversely, stable inputs, maintained functions and reliable recovery would support the interpretation that management is providing durable compensation. These are discriminating expectations, not a validated early-warning formula.

Recovery deserves particular attention because it connects present condition with future capacity. The rehabilitation examples suggest that the intervention needed after degradation can exceed what would have prevented it. The drought experiments show that management history can influence response without a simple change in total carbon explaining the result. The shared-weather evidence adds the possibility that many regions could need recovery support together. Together, these findings justify asking whether damage is accumulating faster than functions can be restored, and whether the resources required for restoration remain feasible. They do not yet supply a numerical global recovery boundary.

A warning system would also have to distinguish real change from changing measurement. Yu and colleagues caution against treating differences between successive global crop maps as trends because their inputs and methods differ. The WoSIS soil compilation documents uneven sampling, missing dates and incomplete comparability between measurements. Such products are valuable for locating evidence and gaps, but repeat observations at comparable depths and with known methods are needed to establish trajectories. Surface-carbon changes cannot automatically represent changes in the rooting zone. Yu et al., 2020, p.20 DOI; Batjes et al., 2024, pp.7–11 DOI

More sophisticated mapping does not remove this requirement. Peter-Contesse and colleagues test spatial transfer of a soil-mapping model and explicitly distinguish that achievement from validation of change through time. Their change analysis does not propagate prediction uncertainty. This limits a claim of detected deterioration; it does not prove that deterioration is absent. Errors across dates can reinforce or partly cancel each other, so the uncertainty of change needs its own evaluation. Peter-Contesse et al., 2026, pp.6–7,16 DOI

Crop-model validation presents a related problem. Folberth and colleagues show that holding soil profiles constant or supplying ample nutrients can improve reproduction of national year-to-year yield variation. They also explain why changing soils must be represented to investigate degradation, and why those processes need additional evaluation. Müller and colleagues similarly find that acceptable final-yield behavior can conceal different responses to underlying stresses. A good historical yield fit therefore cannot validate a boundary involving a process that was held constant, omitted or insufficiently tested. Folberth et al., 2019, pp.21–23 DOI; Müller et al., 2024, pp.14–15 DOI

This keeps model choice in its proper place. The prior question is what evidence would distinguish declining productive capacity from durable adaptation, and a temporary disturbance from a persistent transition. Methods should then be judged by whether they represent and test those distinctions. Adding a more detailed carbon module cannot, by itself, answer questions about depleted soil depth, constrained water supply, coincident regional stress or the feasibility of recovery.

Consequential boundaries could emerge when deteriorating functions, shared stresses and the demands of compensation exceed feasible recovery and replacement. That argument has identifiable empirical premises, important counterexamples and missing links. The collection does not establish a global collapse state or locate a global tipping point. It does establish why neither present production nor the absence of a universal carbon threshold settles the question. Protecting food-producing capacity requires attention to the conditions that keep production possible—and to whether those conditions can still be maintained after the next disturbance.