The Most Crowded Farmland: A Country With No Land Left to Convert

Executive Summary. Bangladesh farms a larger share of its territory than any of its five comparators, and it does so under the heaviest population load of the six. Population density was 1,319 people per square kilometre in 2023, the highest of the six and 4.12 times the peer median of 320.2. Agricultural land covered 72.31% of the land area, again the highest, against a peer median of 43.8%; arable land alone covered 60.63%, also the highest and 1.956 times the median of 31.0%. There is no extensive margin left to open. The intensive margin has carried the country this far: cereal yields went from 1,583 kg per hectare in 1971 to 5,408 in 2024, 3.416 times the starting level. But the 2023 cross-section puts Bangladeshi yield only third of the six, it is bought with fertiliser use of 391.9 kg per hectare of arable land, second of the six, and the country generates 619.9 cubic metres of internal renewable freshwater per person, second lowest of the six. The crowding is not the risk. Running an exhausted land base on inputs the country does not control is.

Crowding, measured rather than asserted

In 2023 Bangladesh had 1,319 people per square kilometre of land area, the highest of the six countries compared here. India 483.7, Pakistan 321.1, Vietnam 320.2, Indonesia 148.6, Thailand 140.3. The peer median is 320.2, so Bangladesh carries 4.12 times the median comparator's density, a gap of 999 people on every square kilometre.

The load is recent enough to be within living memory. In 1971 the same series read 539.9 people per square kilometre. By 2023 it was 1,319, a rise of 779.3 people per square kilometre, or 144.3%, which puts the country at 2.443 times its 1971 density.

Source: World Bank World Development Indicators, population density (EN.POP.DNST), 2023.

Density figures usually deserve suspicion, because the denominator is total land area and most countries carry a good deal of land nobody can farm or inhabit: deserts, high mountains, permanent ice. Bangladesh has very little of that. It is a low, flat, alluvial delta, so the density reading and the usable-land reading sit closer together than they do for Pakistan or Indonesia.

The land has already been converted

The second reading closes off the usual escape route.

Agricultural land was 72.31% of Bangladesh's land area in 2023, the highest of the six. India 60.06%, Pakistan 46.69%, Thailand 43.8%, Vietnam 39.21%, Indonesia 29.13%. Against a peer median of 43.8%, Bangladesh is 28.51 percentage points higher, at 1.651 times the median share.

Arable land, the narrower category that excludes permanent pasture and permanent crops, tells the same story more sharply. Bangladesh reports 60.63% of its land area as arable, the highest of the six, against a peer median of 31.0%. That is 1.956 times the median and 29.63 percentage points above it. India, the next highest, reports 51.75%; Pakistan 39.25%, Vietnam 21.47%, Indonesia 9.4%.

Source: World Bank WDI, agricultural land (% of land area, AG.LND.AGRI.ZS), 2023.

In most countries a growing food requirement can be met partly by converting land that is not yet farmed. In Bangladesh that reserve is close to spent: with 72.31% of the surface already agricultural, land taken for a road, a factory or a housing block comes out of a field, not out of a margin. Land use here is zero-sum in a way it is not in a country farming 29.13% of its territory.

The obvious objection is that this is geography rather than achievement or failure, and it is largely correct. A flat delta with alluvial soil and a long growing season will report a high farmed share whatever anyone decides; an archipelago of mountains and rainforest will report a low one. The share is an endowment as much as a policy outcome. It is still the number a planner has to work with.

How the arithmetic held together

A land base that cannot grow, under a population load that rose 144.3% since 1971, should have produced a food crisis. It did not, and the reason is legible in one series.

Bangladesh's cereal yield was 1,583 kg per hectare in 1971. In 2024 it was 5,408, a gain of 3,825 kg per hectare, or 241.6%, taking output per hectare to 3.416 times its 1971 level. That is what absorbed the extra people.

Source: World Bank WDI, cereal yield (kg per hectare, AG.YLD.CREL.KG), Bangladesh against comparators, 1971-2024.

The comparative position is worth stating precisely, because it is easy to overclaim from a good time series. In 2023 the Bangladeshi yield was 5,321 kg per hectare, third of the six, behind Vietnam's 5,995 and Indonesia's 5,829, and ahead of Pakistan 3,634, India 3,602 and Thailand 3,072. That is 1.464 times the peer median of 3,634, a gap of 1,687 kg per hectare. So Bangladesh is a strong performer, not the frontier. Two countries in the same comparison get more grain out of a hectare, which is the encouraging half of the finding: the intensive margin is not obviously exhausted, because someone nearby is further along it.

What the intensive margin runs on

Yield gains are bought, and the two inputs that buy them show up in the data with opposite signs.

Fertiliser use was 391.9 kg per hectare of arable land in 2023, second of the six, behind Vietnam's 419.9 and ahead of Indonesia 311.1, Thailand 287.3, India 199.1 and Pakistan 160.3. That is 1.364 times the peer median of 287.3, or 104.6 kg more on every hectare than the median comparator applies. Note the denominator: this indicator divides fertiliser use by arable land, and Bangladesh's arable share is the highest of the six, so the high intensity reading is not being manufactured by an unusually small base.

Water runs the other way. Renewable internal freshwater resources came to 619.9 cubic metres per person in 2022, second lowest of the six, above only Pakistan's 225.7. Indonesia holds 7,240, Vietnam 3,606, Thailand 3,130, India 1,014. The peer median is 3,130, so Bangladesh has 19.8% of the median comparator's internal water per head, 2,510 cubic metres less per person.

That figure needs its qualification stated loudly. "Internal" means water generated inside the borders: rainfall and recharge that does not arrive as river flow from somewhere else. Bangladesh sits at the bottom of a basin whose headwaters are in other countries, so the internal measure is emphatically not a measure of how much water is physically available in an ordinary monsoon year. What it measures is how much of the country's water is produced under its own jurisdiction, and by that measure Bangladesh is second lowest of the six. Pakistan, the one country below it, is there for the same structural reason. Read correctly, the indicator is about control and exposure, not about scarcity today.

The strategy that absorbed 144.3% more people per square kilometre since 1971 was to push more nutrient and more water through a fixed area of land. One of those inputs is imported or manufactured and priced on world markets; the other largely originates upstream of the border. Neither is under domestic control, and the land that would let the country substitute area for intensity is already 72.31% committed.

The people have not moved off the land

The pressure is not being relieved from the demographic side either, or not quickly.

The rural population was 66.76% of the total in 2025, the highest of the six. India 64.31%, Vietnam 61.17%, Pakistan 60.47%, Indonesia 40.61%, Thailand 37.2%. The peer median is 60.47%, so Bangladesh sits 6.286 percentage points above it, at 1.104 times the median share. Thailand, at 37.2%, is what the far end of that transition looks like inside the same comparison group.

This is the least robust comparison in the essay. Rural and urban are defined nationally in this series, definitions differ across the six countries, and a country that classifies dense semi-urban settlement as rural ranks higher without anything different happening on the ground. The ranking is real in the data; how much of it is measurement convention, the data cannot say.

For scale: 173.6 million people, a US$450 billion economy growing 4.22% in 2024. Whatever fraction of that population is classified rural, it is fed from a land base whose area cannot increase.

What would change this conclusion

A specific, checkable test. If the next decade's WDI vintages show Bangladesh's cereal yield rising above Indonesia's current 5,829 kg per hectare while fertiliser use stays below Vietnam's 419.9 kg per hectare, the intensive margin will have kept delivering without buying growth by the tonne, and the concern in this essay was misplaced. If instead yield stalls near 5,321 while fertiliser use climbs past 419.9, the country will be paying more inputs for the same grain, which is the signature of a constraint binding rather than a technology improving.

Three moves, with owners and signals.

1. Budget farmland conversion as a quantity, not as a residual of other decisions. With 72.31% of the land already agricultural, every non-farm allocation is a subtraction from the food base, which makes the running total worth reporting rather than reconstructing after the fact. Owner: Ministry of Land, with the Planning Commission on the allocation rule. Success signal: an annually published national land-conversion account, and the arable share holding at or near 60.63% across the next two WDI vintages while urban expansion continues.
2. Target yield per hectare explicitly, and treat fertiliser tonnage as a cost rather than an output. The gap to Vietnam's 5,995 kg per hectare is the honest measure of remaining headroom, and closing it with nutrient volume alone would raise the import bill without raising resilience. Owner: Ministry of Agriculture, with the Bangladesh Agricultural Research Council on varietal and soil work. Success signal: cereal yield above 5,829 in a WDI release with fertiliser use no higher than 391.9.
3. Publish water accounts that separate internally generated flow from transboundary flow. The 619.9 figure invites reading as total availability, which it is not, and the opposite error, treating upstream flow as secure, is worse. Owner: Bangladesh Water Development Board, with BBS on the published series. Success signal: an annual basin-level series showing internal and transboundary components separately, so neither number can stand in for the other.

The counterargument

The strongest version of the objection is not that the numbers are wrong. It is that they are geology wearing the costume of a policy finding.

Bangladesh is a flat alluvial delta with almost no desert, no high mountain range and a long growing season. Of course it farms 72.31% of its land; there is very little land it cannot farm. Of course 1,319 people fit on a square kilometre; the square kilometres are habitable. Comparing that to Indonesia, an archipelago of mountains and rainforest that farms 29.13%, or to Pakistan, much of which is arid, is comparing endowments and then presenting the difference as a result. And on the part that is genuinely a policy outcome, the country has already answered: yields are 3.416 times their 1971 level and 1.464 times the peer median.

Three answers, one of which is a concession.

First, the endowment point is correct and it does not touch the constraint. Whatever produced the 72.31% figure, geology or a thousand land decisions, that is the share a planner inherits, and it means the substitution that other countries can make, more land instead of more input, is unavailable here. A constraint does not become less binding because it was not chosen.

Second, six countries is a comparison, not a sample. Nothing in this essay identifies a cause. Density, farmland share, yield, fertiliser intensity and water endowment are cross-country associations at a point in time, and every one of them is entangled with terrain, climate, cropping calendar and price policy that this indicator set does not observe. The claim being made is about levels and rankings, not about what produced them.

Third, the concession the objection earns. The argument that the intensive margin is getting harder to push rests on two rankings, fertiliser second of six and internal water second lowest of six, and not on any direct measurement of diminishing returns. The series that would settle it, nutrient-use efficiency, cropping intensity, groundwater table trends, soil organic carbon, are not in this data. Fertiliser use being high is consistent with a system approaching the limits of nutrient response, and equally consistent with a system that simply farms three crops a year where its neighbours farm two. Until those series are published, the honest statement is the narrow one: the land is spent, the water is largely someone else's, and the yield still has room to grow.

---

Data sources: World Bank World Development Indicators, retrieved from the BDPolicyLab data lake, 2026-08-10. Comparators are India, Pakistan, Indonesia, Thailand and Vietnam; "peer median" is the median of the five comparators in the stated year, excluding Bangladesh. All six countries report every indicator used here. Years differ by indicator because coverage does: density, land shares, cereal yield and fertiliser use are 2023, freshwater is 2022, rural population share is 2025, and the population, GDP and growth figures are 2024, each being the latest year with comparable coverage across the group. The cereal yield time series runs to 2024 for Bangladesh, so the 1971-2024 trend and the 2023 cross-section are stated separately and should not be conflated.

Sources

1. World Bank WDI, Population density (people per sq. km of land area), EN.POP.DNST: https://data.worldbank.org/indicator/EN.POP.DNST
2. World Bank WDI, Agricultural land (% of land area), AG.LND.AGRI.ZS: https://data.worldbank.org/indicator/AG.LND.AGRI.ZS
3. World Bank WDI, Arable land (% of land area), AG.LND.ARBL.ZS: https://data.worldbank.org/indicator/AG.LND.ARBL.ZS
4. World Bank WDI, Cereal yield (kg per hectare), AG.YLD.CREL.KG: https://data.worldbank.org/indicator/AG.YLD.CREL.KG
5. World Bank WDI, Fertilizer consumption (kg per hectare of arable land), AG.CON.FERT.ZS: https://data.worldbank.org/indicator/AG.CON.FERT.ZS
6. World Bank WDI, Renewable internal freshwater resources per capita (cubic meters), ER.H2O.INTR.PC: https://data.worldbank.org/indicator/ER.H2O.INTR.PC
7. World Bank WDI, Rural population (% of total population), SP.RUR.TOTL.ZS: https://data.worldbank.org/indicator/SP.RUR.TOTL.ZS