Area or yield: the land-sparing is real, and it is fading

Reachability swept first: yes — semantics as well as status. Three ingest routes run through fetchOwidPoints and cross-checked against the multi-column grapher to the tonne. Vintage checked: yes — all three series are FAO (2025) via OWID, lastUpdated 2026-02-25, nextUpdate 2027-02-25. One producer, one vintage.

Revisions

2026-09-23 — two corrections, both found when the data was INGESTED rather than queried. The verdict stands; one showcase example does not.

1. "2023 is the latest year" was wrong: all three cereal series carry 2024 (64 points, 1961–2024), and the forest series runs to 2025. Every figure below is anchored at 2023 and is correct for that endpoint, but the window stops a year short of the data. World yield FELL 0.5% in 2024 (4.2283 → 4.2058), which strengthens the fading-sparing argument rather than weakening it — and which the sheet could not see.
2. ARGENTINA'S −20% YIELD IS AN ENDPOINT ON A DROUGHT, and this sheet defended it as real. Open questions distinguishes Europe's −340% (an artifact of a near-zero denominator) from Argentina's fall, and concludes "the test is the size of the denominator, not whether the share looks strange." That test is right and it is not the only one. Argentina's yield runs 5.41 (2021), 4.82 (2022), 3.99 (2023), 4.85 (2024) — 2022–23 is the La Niña drought and 2023 is the trough. Measured 2010→2024: production ×1.84 (not ×1.40), area +88.6%, yield −2.5% (not −20%). The area-driven story survives — area still grew far faster than production — but "its yield actually fell by a fifth" is a drought, not a trend, and must not be spoken. THE LESSON, and it is a THIRD kind of window error this sheet did not have: it warns twice about a window flattering a number, and checks denominators for small-number artifacts. Neither catches an endpoint landing on a weather year. For an annual agricultural series, check the neighbouring years before believing a terminal move.

Segment: Area or yield

• Source theme(s): Batch 54 #8
• Spine role: main story
• The framing people carry: that feeding more people means clearing more land — or, on the other side, that farming has become so productive it no longer needs to expand.
• Hypothesis (from the bank), four sub-claims: (1) global cereal production growth decomposes "overwhelmingly into yield" since the 1960s with area close to flat — the land-sparing case; (2) the opposite decomposition holds "for specific crops and specific regions, where nearly all growth came from area and the environmental consequence followed"; (3) the yield term is stalling in some major producers "while area resumed growing"; (4) the two terms are "negatively correlated across regions".
• Verdict: CONFIRMED on two of the bank's four sub-claims, partly on a third, and one untested — with a turn the bank did not predict. (1) Land-sparing at world scale: confirmed. (2) The opposite decomposition in specific regions and crops: confirmed for regions, untested for crops (cereals only), and its "environmental consequence" not shown by this data. (3) Yield stalling while area resumed growing: confirmed. (4) The two terms negatively correlated across regions: NOT tested. Over 1961–2023 world production grew 3.57× and 90% of that came from yield. But the 90% is a long-run blend that is fading: yield's share was 82% in 1961–80 and 73% in 2010–23, while world yield growth has roughly halved, from 2.49% a year to 1.33%. Area expansion is coming back, and it is concentrated in South America.

The numbers (every row sourced — no exceptions)

All rows: FAO (2025) via OWID, lastUpdated 2026-02-25, nextUpdate 2027-02-25. Tier A throughout. Cereals (FAO item 1717).

The texture (what narrative flattens)

• The world mostly fed itself by getting more from the same land. Area harvested grew 14% in sixty-two years while production grew 257%. That is the land-sparing case, and at world scale it is real.
• But the sparing is fading. Yield growth has roughly halved since the 1960s, and in 2010–23 area supplied over a quarter of world growth, up from near-zero-or-negative in 1980–2000. China's yield growth fell from 4.86% a year to 1.16%. India is the exception, with yield growth reaccelerating to 2.34%.
• Two regimes, geographically sorted. Asia is almost entirely yield. Africa is roughly half area. And South America's recent growth is mostly area — Brazil and Argentina together added about 17.6 million hectares of cereal land in 2010–23. Argentina is the starkest case in the dataset: production up 40% while area rose 75%, so its yield actually fell by a fifth. More than all of its growth came from new land.

Datoid artifact (Tier-A only)

• Nothing is ingested yet, and all three terms have a clean route, RUN through fetchOwidPoints (64 points each, 1961–2023): cereal-production and cereal-yield are single-column; area harvested is index-of-cereal-production-yield-and-land-use, whose cols[3] happens to be area_harvested — so the connector ingests the right column from a four-column grapher. The per-series runs match the grapher to the tonne.
• A NAMING TRAP, verified: that grapher is called "index-of" and serves RAW values. 647,997,184 hectares for World 1961, not 100. OWID displays it as an index; the CSV does not. Read the column, not the slug.
• Its cols[3] is area only by column ORDER, and the grapher has four value columns (area, production, yield, population). If OWID reorders them, the seed would silently ingest a different quantity. Worth a guard: assert the ingested World 1961 value is ~648M before trusting it.
• No VALUE_DIVISORS entries exist for these slugs. Production in tonnes (billions) and area in hectares (hundreds of millions) would sit on one axis about 4× apart — tolerable, but production and yield must not share an axis (billions against single digits). Settle axis assignment when the panel is built.

Story justification (decide here — do NOT defer)

• Does the argument have a reveal that needs sequencing? YES — BUILD.
• Because: there are two reveals and both need withholding. First, the reader sees production soaring and assumes land was cleared; then area is shown nearly flat — the land-sparing surprise. Second, the yield-growth rate is revealed to be halving and area creeping back, which turns a good-news story into an open question. One view cannot do both.

The questions this raises (and whether we can answer them)

Stage 1 rule 6, added 2026-09-23. Everything above verifies a hypothesis; this names what a READER asks next and forces the decision before Stage 2 runs — fetch it, attribute it, or declare it open. Every slug named here was swept for reachability first.

1. What did South America's new cropland replace — forest, pasture, or other crops? — FETCHABLE, and it is this sheet's own named gap. annual-change-forest-area (swept 200, net_change_forest_area) and share-of-land-area-used-for-agriculture (200) per country. Note that forest-area 404s; the net-change slug is the reachable one. Until one of them is on the page, Brazil's +10.6M ha must not be spoken as deforestation — which is exactly what Open questions already forbids.
2. What is behind the yield slowdown — fertiliser, water, breeding or weather? — PARTLY FETCHABLE. fertilizer-use-per-hectare-of-cropland (200) tests whether the input side plateaued alongside the output side. Breeding and water need attribution, not a series.
3. Why did Argentina's yield fall 20% while its area rose 75%? — ATTRIBUTABLE. Drought years, a shift of planting onto marginal land, a crop-mix change: all plausible, none visible in these three series. Needs a named agronomic or FAO commentary source, or it stays unsaid.
4. Does the decomposition hold for oilseeds and pulses? — FETCHABLE in principle and deliberately NOT fetched. The verdict is scoped to cereals; widening the crop set widens the CLAIM rather than answering a reader's question about this one. It is a theme candidate, not this episode's work.
5. Is the yield-growth slowdown a plateau or a pause? — OPEN.

Open questions / where I'm guessing

• THE IDENTITY CHECK IS NEAR-VACUOUS HERE, and must not be cited as corroboration. FAO defines yield as production ÷ area, so production ÷ (area × yield) equals 1 by construction; the 0.99999–1.00001 residual is only the rounding of the published yield. This is the same weakness as the World = 0 migration check on the previous sheet. It proves the series are one producer's, which is useful, and nothing about whether they are right.
• THE 90% IS A LONG-RUN BLEND — do not lead with it as a current fact. It averages an era when yield exceeded 100% of growth (1980–2000, area shrinking) with a recent era at 73%. The durable claim is the trend: sparing strong historically, weakening now. This is the fourth sheet in a row where the cleanest number was a window choice.
• TWO WAYS A SHARE CAN LIE, AND THIS BULLET HAS NOW BEEN WRONG BOTH WAYS. Europe's yield share of −340% for 1980–2000 is a denominator artifact: production barely moved, so the ratio explodes. It must not be spoken. Argentina's −66% for 2010–23 is an ENDPOINT artifact, which is a different failure and one this bullet did not have: production did grow 40% (ln 0.336), so the denominator is sound — but the yield trough is the 2022–23 La Niña drought, and 2024 recovers to −2.5% from 2010. See Revisions. The area-driven claim for Argentina survives on either endpoint, because area grew far faster than production throughout; what does not survive is "yield genuinely fell 20%", which is weather. The history is the lesson. Draft one filed Argentina under "artifacts, do not speak" — wrong. Draft two called it the strongest evidence in the sheet and concluded "the test is the size of the denominator, not whether the share looks strange" — right about the test, wrong that it was the only one. A denominator check is necessary and not sufficient; for an annual agricultural series, read the neighbouring years too.
• THE ENVIRONMENTAL CONSEQUENCE IS NOT SHOWN. The bank says area-driven growth is where "the environmental consequence followed". This data shows cereal area expanding; it does not show what that land was before — forest, pasture, or other cropland switched to cereals. Brazil's +10.6M ha must not be spoken as deforestation without a land-cover source. Soy, the crop most associated with that frontier, is not a cereal and is not in these figures.
• Cereals only — so the bank's "specific crops" claim is untested. FAO item 1717 aggregates cereals; oilseeds, pulses and all-crops may decompose differently.
• The bank's fourth sub-claim is NOT tested — that area and yield are "negatively correlated across regions in a way that complicates any single global narrative". That needs a cross-regional correlation that was not run.
• 2024 is the latest year for all three cereal series, and 2025 for forest (corrected 2026-09-23 — this said 2023, and every figure above is anchored there). FAO revises recent years, so the 2010–24 figures are the most exposed.