Soil, connected Earth systems and the limits of food-producing capacity: a historical reassessment

20 October 2026

The first Boundary Review asked whether the inherited collection supports a soil-carbon threshold for agricultural collapse. It concluded that it does not establish a universal carbon cutoff or a global tipping point, and that productive capacity depends on several functions, not on carbon alone. That conclusion still holds. It does, however, leave a broader question unanswered: what does the literature show about how food-producing capacity has actually been lost, held and regained, and how soil degradation connects to the climate, water and vegetation systems that set those limits?

This reassessment answers that broader question. It adds a historical perspective, covering regions where the ability to grow crops was lost or sharply reduced. It also adds the modern record of desertification and the places where land is now recovering. Three findings stand out:

1. Regional thresholds in food-producing capacity have been crossed repeatedly. Over five millennia this has happened through salinization, erosion down to infertile subsoil, wind erosion and the loss of water supply. The physical mechanisms are well documented. Claims that soil loss caused the collapse of particular societies are much weaker and often contested.
2. The crossings that mattered most were coupled events. In each case a climate or water stress met land whose buffering had already been reduced, and in several cases the degraded land then fed back on the regional climate. The literature supports these soil–climate–water connections far better than it supports any single soil indicator.
3. Large-scale recovery is real and documented. It has happened in Niger, northern Ethiopia, the Loess Plateau, Iceland and the North Aral Sea. Degradation is therefore not generally irreversible. But recovery has taken decades, has depended on governance and rainfall, and has sometimes used water drawn from elsewhere. Recovery moves boundaries rather than abolishing them.

Evidence base and labels. The review draws on the 173 usable works in the inherited collection plus three newly added PDFs (Zhao et al. 2026, Hijbeek 2026, and a published copy of Bossio et al. 2020). Because the collection contains little historical or dryland literature, it also uses a targeted set of external sources. These were checked against their abstracts, repository records or summaries, but not read in full, and the review marks them as external. Each conclusion carries a status label:

• Supported: established within the stated scope.
• Conditional: depends on stated assumptions.
• Contested: credible sources disagree.
• Unresolved: a decisive link is missing.

1. Reframing: what kind of boundary limits food production?

The original review defined productive capacity as the ability to sustain useful food or feed production under specified conditions and feasible management. It separated damage, collapse and tipping. Those definitions carry over unchanged. The historical record calls for one more distinction, between the different kinds of process through which a region loses the ability to grow crops:

These types interact. Salinization is both chemical and hydrological. Erosion removes nutrients as well as depth, and wind erosion in a drought can change the drought itself. A boundary in food-producing capacity is therefore best described as a combination of conditions that a region can no longer meet. It is not a single crossed value. This matches the original review's conclusion that boundaries must state what they protect. Olsson et al., 2019, pp.44–45DOI

The newly added evidence supports this framing from the farm side. Zhao and colleagues used causal machine-learning methods on 1,246 field trials and 192 working farms in Northeast China. Higher soil organic carbon raised maize yields in trials by 8–13% up to 30–40 g kg⁻¹. On farms it had no detectable effect; farm yields rose only up to about 10 g kg⁻¹ and then fell back. The authors attribute the difference mainly to fertilizer applications about 50% higher on farms, to acidification, and to poorly implemented conservation practices. On farms, growing-season rainfall, not carbon or nitrogen, dominated yield variation. Zhao et al., 2026, pp.4–7DOI

In the accompanying commentary, Hijbeek concludes that soil-carbon functions matter only where they relieve the constraint that is currently limiting yield. She argues that "global or continental soil carbon thresholds will have little practical meaning for farmers." Hijbeek, 2026, pp.3–4DOI

Neither paper says soil condition is unimportant. Northeast China is the region where black-soil topsoil has thinned and carbon has been declining for six decades. Both papers say the limiting variable has to be found, not assumed.

2. The historical record of crossed thresholds

2.1 Salt: southern Mesopotamia and the irrigated drylands

Irrigation in a hot, flat, poorly drained basin raises the water table and leaves salts behind as water evaporates. The physical mechanism is not in dispute.

Jacobsen and Adams (1958, Science; external) argued from temple archives that salt accumulation in southern Mesopotamia drove a shift from wheat to the more salt-tolerant barley. They linked this to falling reported yields over the late third and early second millennia BCE, and to a northward shift of political and agricultural centres. Their interpretation was challenged by Powell (1985; external), who disputed whether the textual evidence demonstrates progressive salinization. It was defended by Artzy and Hillel (1988; external). The magnitude of the yield decline and its role in political change therefore remain contested. That irrigated land in that setting can cross a chemical threshold beyond which staple crops fail is supported.

The same process is operating today at a far larger scale. Prăvălie's synthesis reports that salt-affected soils cover roughly 1 billion hectares. More than 20% of the global irrigated area is affected by salinization and waterlogging, a share that could exceed 50% by 2050 without intervention. Agricultural losses from salinization reach 18–43% across the world's arid and semi-arid lands. Iran is the most heavily affected country, with about 19% of its territory salinized. Prăvălie, 2021, pp.9,11DOI

The ancient case is therefore not just a historical curiosity. Irrigation, the technology that compensates for water limits, carries a slow chemical boundary of its own.

The Aral Sea basin is the clearest modern case of this coupling. Diverting the Amu Darya and Syr Darya for irrigated cotton shrank the sea, salinized the land around it and exposed a salt-dust source. Prăvălie identifies intensified salinization in Kazakhstan and Uzbekistan as a direct result. Prăvălie, 2021, p.11DOI

2.2 Erosion: civilizations, islands and the plough

Montgomery compiled erosion and soil-production rates worldwide. Conventionally ploughed fields erode at a median rate one to two orders of magnitude faster than soil forms. That imbalance predicts a time to strip a typical hillslope soil profile of roughly 500 to a few thousand years, comparable to the lifespan of many agricultural civilizations. Montgomery cites archaeological evidence of pronounced erosion during the rise and decline of societies in the Middle East, Greece, Rome and Mesoamerica. The most extreme case is the island of Mangaia, where upland Polynesian agriculture stripped soil from most of the island. Violent conflict over the remaining depositional patches followed. Montgomery, 2007, pp.1,4DOI

Lal makes the stronger claim that soil degradation "caused the downfall" of the Harappan, Mesopotamian, Lydian and Maya civilizations. He relies on mid-twentieth-century authorities for this. Lal, 2001, p.1DOI That causal attribution is older and less secure than the erosion evidence itself.

The Maya lowlands show why. Geomorphic work by Beach and colleagues (external) documents region-wide "Maya Clay" deposits: eroded upland soil buried in footslopes, lakes and wetlands, accumulating from the Preclassic through the Late Classic. Erosion clearly reduced upland soil. But Evans and colleagues (2018, Science; external) used gypsum hydration water from Lake Chichancanab to estimate that annual rainfall fell by 41–54% during the Terminal Classic, with intervals of up to 70% reduction at peak drought. The authors caution that drought cannot be singled out as the cause of collapse. The supportable inference is coupling: a society whose agricultural base had already been altered by centuries of erosion met a severe multidecadal drought. Neither factor alone explains the outcome. The physical losses are supported; the decisive cause of societal collapse is contested.

Iceland gives a cleaner example of an ecological state shift. Settlement in the late ninth century was followed by deforestation of most of the original birch woodland and by sustained grazing. Soil conservation authorities estimate that about half of the vegetated cover may have been lost and that some 3 million hectares became eroded (Runólfsson and Arnalds; external). Once the volcanic soils lost cover, wind and frost erosion could proceed without further grazing pressure. This is a reinforcing feedback of the kind the original review could only infer from fragmentary cropland evidence.

2.3 The Dust Bowl: degraded land changes the drought

The 1930s Great Plains Dust Bowl is the best-documented case of a land–atmosphere feedback that degraded food-producing capacity. The IPCC land-degradation chapter cites modelling showing that dust emissions from degraded cropland amplified the intensity, extent and duration of the drought. Olsson et al., 2019, p.33DOI

The underlying study is Cook, Miller and Seager (2009, PNAS; external). Sea-surface temperatures alone produced a drought that was too weak and centred too far south-west. Matching the observed pattern required adding the loss of crop cover and a dust aerosol source from failed fields. The authors note that the drought was not meteorologically extreme by nineteenth- and twentieth-century standards; ploughed, unprotected land made it a catastrophe. Prăvălie likewise attributes the event to the simultaneous action of climatic, socio-economic and agricultural drivers. Prăvălie, 2021, p.17DOI

The economic consequences show what "crossing" meant in practice. Hornbeck (2012, American Economic Review; external) found that more-eroded counties suffered immediate and persistent declines in land values and agricultural revenue relative to less-eroded counties. Shifting land from crops to livestock recovered only 14–28% of the initial cost. The main adjustment was population loss. The land was not abandoned to desert, but its productive value was persistently reduced, and the people who farmed it left. The IPCC chapter calls the Dust Bowl "an important wake-up call" that led to soil conservation research and institutions. Olsson et al., 2019, p.37DOI

The Soviet Virgin Lands campaign repeated the pattern. Roughly 23 million hectares of northern Kazakh steppe were ploughed between 1954 and 1963, followed by severe wind erosion and dust storms in the early 1960s. Much cropland was abandoned after 1991 (external remote-sensing and historical sources). The common ingredients were semi-arid grassland, rapid conversion, bare fallow and a dry run of years.

2.4 The Sahel drought: a contested feedback

The Sahel endured one of the strongest multidecadal rainfall shifts recorded anywhere: wet decades in the 1950s–60s were followed by two decades of deficient rain, with a 20–30% decline by the mid-to-late 1980s, accompanied by famine. Bezner Kerr et al., 2022, p.73DOI

Charney's albedo-feedback hypothesis (1975; external) proposed that overgrazing brightened the surface and suppressed rainfall. Desertification would then be self-reinforcing. Later work shifted the primary cause to ocean forcing. Giannini and colleagues (2003, Science; external) reproduced the long-term drying from sea-surface temperatures alone, with Indian Ocean warming most important and land-surface processes acting mainly as amplifiers.

This matters for the boundary question. The Sahel is often cited as proof of human-driven desertification, but the evidence shows external climate forcing with land feedback as a secondary amplifier. The land feedback is conditional; the externally forced drought is supported.

2.5 The slow modern crossings: topsoil, black soils and aquifers

Several present-day breadbaskets are approaching physical boundaries that output statistics do not reveal.

• US Corn Belt. Thaler, Larsen and Yu (2021, PNAS; external) combined satellite imagery with LiDAR and estimated that about 35% of cultivated area has lost its entire A-horizon, mostly on convex hilltops where tillage moves soil downslope. They estimate a yield penalty of about 6% and roughly US$2.8 billion a year in losses. Tillage erosion is not simulated in the national erosion models used to judge sustainability.
• Northeast China and other black-soil regions. Liao and colleagues report, citing secondary sources, that black soils cover 17% of global cropland yet produce about 30% of wheat. In 70 years, cultivation has removed 20–50% of their original soil carbon, and Songnen Plain topsoil has thinned from 60–80 cm in the 1950s to 20–40 cm today. Topsoil thinning there is linked to yield reductions of about 27%. Liao et al., 2026, pp.2,4DOI
• Mollisols worldwide. Meng and colleagues map a 5.59% decline in surface soil carbon across global Mollisol croplands from 1990 to 2020, with the decadal rate of loss slowing from 2.98% to 1.36%. Meng et al., 2026, pp.1–2DOI
• Soil lifespans. Evans and colleagues find that just under a third of conventionally managed soils in their dataset would lose a 30 cm topsoil in under 200 years, and 16% in under 100 years. Conservation measures extend many lifespans beyond 10,000 years. Evans et al., 2020, p.2DOI
• Aquifers under irrigated soil. Scanlon and colleagues (2012, PNAS; external) project that at current depletion rates about 35% of the southern High Plains will be unable to support irrigation within 30 years. This is a hydrological boundary on soils that remain physically intact.

These cases bear directly on the hidden-deterioration question raised in the original review. In the Corn Belt, output has continued to rise while the A-horizon has been lost from a third of fields. Harvests can continue while a slowly renewed resource is depleted, and the historical record shows where that ends when a drought arrives.

2.6 What the historical cases have in common

Inference (supported): the episodes that most reduced regional food-producing capacity were not passages of a single soil indicator through a fixed value. Each combined three things: (i) a slow loss of buffering — topsoil, cover, drainage or aquifer storage; (ii) a fast stress, usually drought; and (iii) limited options to substitute, relocate or recover. The same drought on intact land was usually survivable. Degraded land without a stress kept producing. Inference (contested): that soil degradation alone brought down specific civilizations. Most modern scholarship treats soil loss as one strand among climate, political and economic causes.

3. Desertification today: extent, attribution and thresholds

3.1 How much land, and why the estimates disagree

Global estimates of degraded land range across almost an order of magnitude, from under 10 to 60 million km² (8–45% of ice-free land), depending on whether they rest on expert opinion, satellite productivity trends, biophysical models or abandoned-land inventories. Olsson et al., 2019, p.21DOI; Prăvălie, 2021, p.2DOI The spread reflects different definitions, not only different data. A boundary analysis must state which kind of degradation it counts.

Satellite records add a further complication. Vegetation has greened at the global scale, probably partly through CO₂ fertilization, while regional browning has intensified since about 2000. Prăvălie, 2021, p.16DOI Burrell, Evans and De Kauwe (2020, Nature Communications; external) attempted the first observation-based attribution of desertification. They separated climate change, climate variability, CO₂ fertilization and land use. Between 1982 and 2015, 6% of drylands underwent desertification driven by unsustainable land use compounded by climate change. Anthropogenic climate change alone degraded 12.6% of drylands (5.43 million km²), affecting 213 million people, 93% of them in developing economies. Average greening concealed regional degradation.

The UN Convention to Combat Desertification's 2024 assessment (external) reports that 77.6% of Earth's land experienced drier conditions in 1990–2020 than in the preceding 30 years. Drylands expanded by about 4.3 million km² and now cover about 40.6% of land outside Antarctica, and about 7.6% of global land crossed an aridity-class threshold. These are climate classifications, not measured losses of production, but they describe a moving boundary for rain-fed agriculture.

3.2 The clearest evidence for a soil-functional threshold

The strongest support in the literature for genuine threshold behaviour in soil function comes from drylands, not from cropland carbon studies. Berdugo and colleagues (2020, Science; external) measured 20 ecosystem attributes across global dryland sites. As aridity increased, ecosystems changed abruptly in three phases: plant productivity declined sharply near an aridity of 0.54, soil fertility near 0.7, and plant cover and richness near 0.8 (aridity = 1 − precipitation/potential evapotranspiration). They estimate that more than 20% of the land surface could cross one or more of these thresholds by 2100.

This is closer to what the initial proposal wanted than any SOC:clay ratio. It ties a soil-fertility transition to an Earth-system driver, and projects when regions will cross it. Its limits are equally important:

• it is a space-for-time comparison of natural drylands, not observed transitions;
• it concerns ecosystem attributes, not crop yields under management;
• the authors note the projections still need validation.

The finding is supported for dryland ecosystems and conditional for food production.

Sanderman, Hengl and Fiske estimate that agriculture has removed 133 Pg C from the top 2 m of soil over 12,000 years (revised to 116 Pg C in their 2018 correction), with losses accelerating over the past two centuries. Hotspots coincide with major cropping regions and with degraded semi-arid grazing land, especially the rangelands of Argentina, southern Africa and Australia. Sanderman et al., 2017, pp.1–2DOI The overlap between historic carbon loss and dryland grazing matches the IPCC's description of rangeland transitions that need substantial intervention to reverse. Olsson et al., 2019, p.45DOI

3.3 Planetary-scale accounting

Three external syntheses place regional degradation in a planetary frame:

• Richardson et al. (2023, Science Advances) find six of nine planetary boundaries transgressed, including land-system change and freshwater. They show that climate and land-system transgressions have to be assessed as a connected system.
• Wang-Erlandsson et al. (2022, Nature Reviews Earth & Environment) propose a "green water" boundary: the share of land where root-zone soil moisture departs from Holocene variability. They conclude, provisionally, that it is already transgressed. This is the soil-moisture variable that rain-fed crops depend on.
• Gerten et al. (2020, Nature Sustainability) estimate that almost half of current food production depends on transgressing four terrestrial boundaries: biosphere integrity, land-system change, freshwater and nitrogen. Under strict compliance and the present food system, the land could provide a balanced diet for only 3.4 billion people. With redistributed cropland, better water and nutrient management, less waste and dietary change, it could support 10.2 billion.

Gerten's result is the closest the literature comes to quantifying a global boundary in food-producing capacity. It shows that current output is partly borrowed from the stability of other Earth systems. It also shows that the boundary depends on how food is produced and consumed, not only on biophysics. It does not describe a self-reinforcing global collapse. Its status is conditional on the boundary definitions, which remain debated.

4. De-desertification: where land is recovering

The historical record could be read as a one-way story. Recent evidence shows otherwise. Large regions are recovering, and the way they recover is informative about the boundaries themselves.

4.1 Niger and the West African Sahel

From the mid-1980s, farmers in Niger's Maradi and Zinder regions began protecting and pruning tree regrowth from stumps and roots in their fields: farmer-managed natural regeneration. Sendzimir, Reij and Magnuszewski (2011, Ecology and Society; external) document regreening across more than 5 million hectares over about two decades. No single actor or policy produced it. Governance changed first, especially farmers' de facto rights to the trees on their land, then livelihoods, and the biophysical environment last. The authors caution that the regions' future remains uncertain under climate change and rapid population growth. Its effect on crop yields is plausible but not yet rigorously measured; a randomized evaluation in Niger is under way (J-PAL; external).

At larger scale, the IPCC notes general greening in the Sahel, probably driven by rainfall recovery and rising CO₂. Olsson et al., 2019, p.24DOI Brandt and colleagues (2020, Nature; external) mapped more than 1.8 billion individual trees in the West African Sahara and Sahel from sub-metre imagery. They argue this challenges prevailing narratives of dryland desertification.

The African Great Green Wall is less advanced than often reported. Its 2020 status report (UNCCD; external) found about 4% of the strict intervention zone restored, and about 18–20 million hectares when wider national efforts are included, against a target of 100 million hectares by 2030.

The rainfall recovery itself is partial. It comes with false starts, early cessations, longer dry spells and more intense rain. Bezner Kerr et al., 2022, p.73DOI Human-intensified heat and rainfall extremes have reduced West African millet and sorghum yields by 10–20%. Bezner Kerr et al., 2022, p.12DOI Regreening and agricultural risk are rising together.

4.2 Northern Ethiopia

Nyssen and colleagues (2009, Science of the Total Environment; external) repeated photographs taken by the 1868 British military expedition. They found a significant improvement in vegetation cover and soil protection in northern Ethiopia over 140 years, attributed to exclosures, terracing and community labour programmes intensified since the 1980s–90s. A 2014 follow-up extended the comparison to 361 landscapes. The authors argue the findings challenge the view that dryland degradation is irreversible. They also note that recovery has not healed all scars, and that exclosures compete with livestock feed. This review did not assess conditions after the 2020–2022 conflict in Tigray.

4.3 China: the Loess Plateau and the sandy lands

China's restoration programmes are the largest deliberate reversal of land degradation on record. After severe droughts in 1997 and floods in 1998, the government launched the Natural Forest Protection Program and Grain for Green, which converts sloping cropland to forest or grassland. By the late 2000s Grain for Green had increased vegetative cover by about 25 million hectares. Yellow River sediment concentration fell 38% during 2000–2007, and Grain for Green plots retained more soil moisture and phosphorus than comparison plots. Satellite data confirm rising vegetation cover and declining bare soil in 2001–2015. Olsson et al., 2019, p.52DOI China and India together account for about a third of global greening since 2000. In China, 42% of that greening came from forests and 32% from cropland; in India, 82% came from cropland. Olsson et al., 2019, p.21DOI

Wang and colleagues (2016, Nature Geoscience; external) attribute a roughly 90% decline in Yellow River sediment load over 60 years mainly to human action. Terraces and check dams dominated until the 1990s, and vegetation restoration has dominated since. China's national monitoring (external, official) reports that desertified land has shrunk for four consecutive five-year monitoring periods, by an average of 2,424 km² a year between 2009 and 2014.

The Chinese case also shows the limits of recovery:

• Feng and colleagues (2016, Nature Climate Change; external) found that revegetation raised evapotranspiration and reduced runoff. Loess Plateau productivity was already close to an estimated limit of about 400 g C m⁻² yr⁻¹, beyond which human water demand could not be met.
• The IPCC questions the sustainability of plantations kept alive by drip irrigation from groundwater. Olsson et al., 2019, p.52DOI
• Cao and colleagues (2011, Earth-Science Reviews; external) argue that afforestation unsuited to local climate and soils in arid and semi-arid China has in places degraded soils, reduced cover and worsened water shortages.

India's greening, concentrated in irrigated cropland, depends partly on groundwater that is itself being depleted. Greening that borrows water from another part of the hydrological system relocates the boundary rather than removing it.

4.4 Other recoveries

• Iceland. More than a century of state-led land reclamation, beginning with a soil conservation service founded in 1907, has reduced erosion. A UNCCD case study (external) reports that reclamation and natural healing now outpace degradation.
• North Aral Sea. Kazakhstan's Kok-Aral dike, completed in 2005, raised water levels and lowered salinity. It allowed commercial fisheries to return to the northern sea. The southern remnant has continued to shrink and salinize (external).
• US Great Plains. Conservation institutions created after the Dust Bowl, and later the Conservation Reserve Program, have converted erodible cropland to perennial cover with measurable carbon gains. Olsson et al., 2019, p.50DOI
• Mollisols. The rate of surface-carbon loss on global Mollisol croplands has roughly halved since the 1990s, and about 39% of Mollisol area gained carbon from 1990 to 2020. Meng et al., 2026, p.2DOI

4.5 What recovery teaches about boundaries

Recovery evidence revises the picture in four ways:

• Irreversibility is not general (supported). Landscapes described as desertified in the 1970s–80s have regained vegetation and soil protection within decades.
• Recovery is not symmetrical with degradation (supported). It has required the pressure to be removed, often secure tenure and sustained investment, and frequently a favourable shift in rainfall. This parallels the slow plot-level rehabilitation in African smallholder systems described in the original review. Tittonell and Giller, 2013, pp.7–9DOI
• Greening is not the same as restored food-producing capacity (supported). Most recovery indicators are vegetation indices, tree counts or sediment loads. Few studies measure whether the land can again produce food reliably, and some of the best-documented restoration deliberately took cropland out of production.
• Recovery can borrow from other systems (conditional). When restoration uses groundwater or reduces downstream runoff, the boundary has moved within the connected water system.

5. Connected Earth systems: how soil degradation reaches beyond the field

The original review found the global connections the least developed part of the inherited collection. The broader literature fills several of those gaps. The mechanisms, and how firmly each is established, are:

Land–atmosphere feedback (supported regionally). Degraded land can change the climate that stresses it. The Dust Bowl is the strongest case: crop loss and dust made the drought hotter, stronger and further north. Olsson et al., 2019, p.33DOI The Sahel case shows the opposite caution. External ocean forcing was primary, and the land feedback was an amplifier. In both, the land feedback acts on a drought that already exists.

Forest–rainfall–agriculture coupling (supported regionally). Leite-Filho and colleagues (2021, Nature Communications; external) found that in the southern Brazilian Amazon rainfall falls steeply once forest loss passes about 25–30% at 112 km scale (45–50% at 56 km scale). They estimate that avoiding further deforestation would prevent agricultural losses of up to US$1 billion a year. Clearing land for soy and cattle reduces the rainfall those systems depend on: a negative-sum trade within one connected system. This is the clearest quantified example of a non-soil Earth-system threshold that limits food production.

Water borrowed across time and space (supported). Irrigation compensates for dry soils by drawing on rivers and aquifers. That compensation can end through salinization, as in Mesopotamia and the Aral basin, or through aquifer depletion, as in the High Plains. Revegetation competes with agriculture and people for the same water. Green-water departures suggest the soil-moisture base of rain-fed production is shifting at planetary scale (Wang-Erlandsson et al., external).

Soil carbon–climate feedback (conditional, and of uncertain sign). CMIP6 models show warming-driven soil-carbon losses offset by CO₂-stimulated inputs, with large structural uncertainty. Varney et al., 2024, p.6DOI Observed Mollisol carbon shows a nonlinear response to the rate of warming: net loss turned toward gain above about 0.031 °C a year, a threshold that rises toward lower latitudes. About 65% of Mollisol cropland may exceed it by 2050 if recent relationships hold. Meng et al., 2026, p.1DOI This is a reminder that warming does not translate simply into soil-carbon loss, and that a threshold in soil carbon need not point toward degradation.

Nutrient supply chains (conditional). Erosion exports phosphorus and increases dependence on mined fertilizer, linking field soil loss to a finite global resource. Alewell et al., 2020, pp.2,5DOI

Synchronized stress (supported) versus recovery (supported). Atmospheric circulation patterns can cause simultaneous below-trend harvests in several breadbaskets. Kornhuber et al., 2023, p.5DOI National cereal production has nonetheless generally recovered after drought and heat disasters. Lesk et al., 2016, p.2DOI Human-induced warming has also slowed global agricultural productivity growth by about 21% since 1961, relative to a world without warming. Bezner Kerr et al., 2022, pp.12,16DOI The historical record suggests what links these findings: recovery is fast where land retains its buffering and slow or absent where buffering has been spent. Whether depleted soils make synchronized losses deeper or longer at the global scale has not been measured. That is the decisive missing link.

6. What the literature now supports

The reframed conclusions, with their status:

Overall answer. The literature supports strong connections between soil degradation, the climate and water systems around it, and consequential limits on regional food production. The historical record contains repeated crossings, and the modern record shows the same processes operating faster and over larger areas. It also shows recovery at scales once thought impossible. What it does not yet support is a single global threshold, or the claim that local crossings are combining into planetary failure. The most defensible global statement is Gerten's: current output depends partly on transgressing other Earth-system boundaries, and whether it can be sustained depends on how production changes.

7. Implications for the Global Soil Transformer project

The historical reframing changes what a threshold study should measure:

1. Use boundary variables that history shows were decisive. These are topsoil depth relative to rooting needs, salinity in irrigated land, aridity combined with soil fertility (following Berdugo), root-zone soil moisture, and dependence on inputs or groundwater. A SOC:clay class can be one input, not the boundary itself.
2. Treat time of emergence as a coupled quantity. The relevant moment is when buffering has fallen far enough that a plausible drought produces non-recovering losses, not when a soil indicator crosses a line.
3. Use historical and recovering regions as tests. The Dust Bowl counties, the Virgin Lands, the Corn Belt hilltops, Northeast China's black soils, Niger and the Loess Plateau provide natural experiments. A method that cannot distinguish their degrading and recovering trajectories in hindcast should not be trusted to project future crossings.
4. Report recovery capacity alongside degradation. Where rehabilitation is feasible, and what water and institutional resources it would require, is as relevant to food security as where degradation is projected.
5. Keep the null result publishable. The farm-level evidence already shows that carbon effects can vanish under real management. A study designed to test, not assert, an underestimated soil feedback remains the defensible design.

8. Limits of this reassessment

• External sources are not yet in the collection. Many historical and dryland conclusions rest on sources checked at abstract or repository level only. Their figures should be confirmed against full texts before publication.
• Historical yield evidence is thin. Ancient cases rely on archives, sediments and archaeology. Their causal interpretation is debated, and dramatic cases are over-represented.
• Recovery indicators are mostly vegetation-based. Few studies measure whether recovered land again supports reliable food production.
• No systematic search was made. This is a targeted extension, not a systematic review of desertification or collapse literature. Absence of a case here is not evidence that it does not exist.
• Further verification is needed. Claims based on abstracts and summaries remain provisional pending examination of their full texts.

Read the boundary review