In Pakpattan, a district in the middle of Punjab's cotton belt, cotton covered 52,000 hectares in 2000-01 and 4,500 in 2024-25, according to the provincial crop reporting series published by the Pakistan Bureau of Statistics and held in Data Darbar. Across the sixteen Punjab districts that each sowed at least 40,000 hectares of cotton in 2000-01, the area under cotton has fallen from 2.34mn hectares to 1.24mn over the same twenty-four years, while the area under rice, maize and sugarcane has grown.
Nationally, the harvest has halved from 14.0mn bales in 2014-15 to 7.1mn in 2024-25, and the textile industry that spins it is the country's largest exporter. Pakistan's raw cotton imports cost $2.7bn in 2024-25, according to revised figures from the Pakistan Bureau of Statistics.
Pests and seed on one side and prices on the other are the usual explanations for cotton's retreat. This piece adds the weather: summers in Punjab's cotton districts have been wetter in 2015–2024 than in 1981–2010, and cotton's district yields are lower in wetter summers. That association is what the analysis establishes. The expansion of other crops in the same districts is described alongside it, and the consequences for the belt's canals and groundwater are raised but not measured.
How the comparison is made
We compare year-to-year changes in each district's cotton yield with the rain that fell in that district in July and August, allowing each district its own long-run trend, over the kharif (summer) seasons from 1981 to 2024. The same comparison is run for rice, maize and sugarcane. District boundaries have moved since 1981, so districts that enter the crop series after 1981-82 are counted with the district they were carved from, which gives thirteen units covering Punjab's cotton country and nineteen with Sindh's. Rainfall is CHIRPS, a satellite and gauge blend; temperature is ERA5-Land, a reanalysis; both are averaged over the district, which understates local extremes. PIDE set district cotton against weather for thirty districts over 1981–2010; this analysis runs to 2024 and adds the other kharif crops. Sources and the model are set out in the method note.
Punjab's cotton summers have been wetter
Cotton in Punjab is sown from March to May and picked from August to November, on the calendar used by the province's Crop Reporting Service. July and August fall between sowing and the main picking, when the crop is flowering and forming bolls, which is why they are the rainfall window tested here. September–October rain, during picking, is tested alongside and no clear association is detected, the estimate being far too imprecise to say either way. Figure 1 shows how much July–August rain the thirteen Punjab units have had each year since 1981, with the cotton yield of the same units beneath.
Across the thirteen units, July–August rainfall averaged 92 mm over 1981–2010, 130 mm over 2011–2024 and 144 mm over 2015–2024, an increase of 56 per cent on the earlier period. The year 2022 is exceptional at 249 mm; leaving it out, the 2015–2024 mean is still 132 mm, 44 per cent above the earlier period. The daytime heat indicators have not moved the same way: days at or above 40°C in May and June averaged 39 a year in 1981–2010 and 35 since 2011, and nights at or above 28°C in July–September are unchanged at 44 a season. On these measures, the change in the cotton summer is in the rain. Whether it persists, and what lies behind it, this record does not say.
The lower panel shows what wet summers have meant for cotton. The three worst years in the Punjab record, 1983, 2022 and 2015, when yields across the units averaged 54, 27 and 18 per cent below trend, are all years when July–August rain was far above the earlier average.
Cotton has the largest estimated yield penalty in wet summers
A wet summer is bad for cotton; the question for the belt is whether it is bad for the crops growers moved into. Figure 2 shows the association between July–August rainfall and district yield for cotton and for rice, estimated with the same model on each crop's own set of districts.
For cotton in Punjab, 10 mm more July–August rain is associated with a yield 2.3 per cent lower, with an interval from 0.3 to 4.2 per cent. On all nineteen units, Sindh included, the association is 1.8 per cent per 10 mm, from 0.6 to 3.1. Two exceptional harvests carry much of the Punjab estimate. Without 1983 and 2022 it is 0.8 per cent per 10 mm, with an interval from −1.8 to +0.1 that includes zero. Comparing districts within the same season, which sets aside whatever all of Punjab shared in a given year, including the common part of the monsoon, also gives 0.8, with an interval from −2.4 to +0.8. That is a narrower question, and on it the estimate is statistically uncertain. Nor was 1983, the worst harvest in the series, only a wet year: a paper in the Pakistan Development Review two years later recorded that pest attack destroyed nearly 40 per cent of that year's cotton (Akhtar, 1985). For rice it is 0.4 per cent per 10 mm, from 0.1 to 0.7. Maize and sugarcane show no clear penalty in this window, but the comparison stops there: Punjab reports spring and autumn maize together, and only the autumn crop stands in the July–August rain, so a near-zero annual coefficient cannot show that maize tolerates it. Cotton has the largest estimated negative association of the four; that describes the estimates, each on its own districts, not a ranking of crops exposed to the same rain.
Most of the year-to-year variation in district cotton yields remains unexplained by these five weather measures, which account for about a seventh of it once trends are removed; pests, inputs and prices are among the possible additional influences, alongside local shocks and the limits of the data. Nor does the model identify the pathway from rain to yield. Waterlogging is one documented candidate: in trials at the Cotton Research Station in Multan, cotton kept waterlogged grew shorter and yielded less seed cotton (Hussain and others, 2018). This regression does not test that pathway or any other.
How crop areas changed
Figure 3 shows, district by district, how sown area changed between 2000-01 and 2024-25.
Cotton fell in fifteen of the sixteen districts, by 1.10mn hectares in all, and rose only in Bahawalnagar, where rice also tripled. Rice area rose in all fifteen districts with a comparison, by 0.69mn hectares; maize in eleven of sixteen, by 0.65mn; sugarcane in eight of sixteen, by 0.27mn. The shift is sharpest in the centre of the belt. In Pakpattan, rice went from 15,000 hectares to 123,000 and maize from 35,000 to 140,000 as cotton fell to 4,500. In Vehari, cotton went from 238,000 hectares to 50,000 and maize from 12,500 to 215,000. In Rahim Yar Khan, cotton fell by a third and sugarcane went from 28,000 hectares to 217,000.
These are sown area, not unique physical land area. The same field can carry a spring and an autumn crop, changes in cropping intensity could contribute, and each crop's reporting covers a different set of fields, so the 1.6mn hectares added across the three crops exceed the 1.1mn that left cotton and are not a destination account for former cotton fields. The decline in cotton is of the area planted; the association in Figure 2 is with the yield on what was planted, and the two are different outcomes.
Why cotton planting shrank is harder to establish. Comparable farm-gate returns net of costs are not available for the four crops in the belt, so no ranking of returns is offered here. Cotton's repeated failures since 2015 are a plausible influence on planting, though the district record shows no consistent next-season response to a bad harvest. Prices have not always held either. In 2023-24 the government set an intervention price of Rs 8,500 per 40 kg for seed cotton, and by mid-October 2023 Dawn was reporting prices of Rs 7,500 and then Rs 7,000, the lower figure from the chairman of the Cotton Ginners Forum.
What the replacement crops need
A switch away from cotton is not, in itself, a switch to thirstier farming. On FAO's seasonal crop water needs, cotton at 700 to 1,300 mm sits above maize (500 to 800) and wheat (450 to 650) and below sugarcane (1,500 to 2,500); rice's crop use, 450 to 700 mm, is low, but a paddy field is kept flooded and the water applied to it is far more than the plant uses. Three quantities are easily confused here: what a crop uses, what is applied to the field, and what is pumped from the ground. Crop areas measure none of them; paired with FAO's ranges, which are requirements under favourable conditions rather than observed Punjab use, they support only an illustrative calculation. Rain enters the sum too: part of a wetter summer substitutes for irrigation, although runoff and timing limit how much, so wetter years do not simply add to water pressure.
The belt was irrigated before the shift and is irrigated after it. What the Mouza Census can show is how widely each source is reported. Figure 4 shows the irrigation sources villages in the sixteen districts reported in 2020.
Across the sixteen districts, 80 per cent of mouzas report a canal and 78 per cent a tubewell, so most report both. In the centre of the belt, Vehari, Pakpattan, Sahiwal, Lodhran and Khanewal, both are above 85 per cent. On the western margin the canal share drops: 50 per cent in Rajanpur and 33 per cent in Dera Ghazi Khan, where tubewells are reported by two-thirds and half of villages. A rice field wants a flooded surface through the months the canal is running; a cane field occupies the ground for longer than cotton and on a different irrigation calendar, with spring and autumn plantings and an October–March harvest (Agriculture Policy Institute, Sugarcane Policy Analysis 2022-23, section 2). Where canal deliveries fall short of what those crops want, the tubewell is the possible response in a belt where most villages report one, and pumping has its own cost in diesel, grid electricity or, increasingly, a solar array, the subject of this issue's Big Read. A cross-section of sources in 2020 cannot say how reliance on either source has changed over twenty-four years, who owns the pumps, how much is pumped, or what the water table has done.
What this means
Punjab's cotton summers have been wetter for a decade, and cotton's yields are lower in wet summers: the largest negative estimate among the separately fitted crop models, and one that weakens to statistical uncertainty once the two exceptional harvests are set aside or districts are compared within the same season. That is consistent with rainfall being one pressure on the crop. It is not an identified share of cotton's decline, and it describes the yield on a planted field, not the decision to stop planting one. That decision, and the parts played by pests, seed and prices, remain to be weighed.
What the record does establish is the contraction of cotton and the expansion of other crops. In Pakpattan, cotton has all but gone, and rice and maize now occupy much more of the district's sown area. In the 2020 census, 92 per cent of the district's villages reported a canal and 93 per cent a tubewell. As rice and maize spread across the district, how much of the water for those crops came through the canal, and how much did farmers have to pump?
---
Crop data. Crop areas, production and yields are PBS's district series from the provincial crop reporting services, pulled from na.data.gov.pk on 27 September 2026 and audited on 19 October. The Sujawal row that carries a provincial total is excluded, and cotton production is in bales of 170 kg. Maize area for 2024-25 is from MNFSR's Crops Area and Production 2024-25, Table 3, whose 2023-24 column matches the PBS series. The maize series combines spring and autumn crops.
The sixteen-district belt. Every Punjab district with at least 40,000 hectares of cotton in 2000-01. Chiniot is counted with Jhang, and Kot Addu and Taunsa, separated in 2022, with Muzaffargarh and Dera Ghazi Khan. District changes compare 2000-01 with 2024-25 and are reported only for district-crops observed at both dates: fifteen districts for rice, sixteen for the other crops.
Units for the weather analysis. The analysis starts from the 24 districts nationally with at least 20,000 hectares of cotton in 2000-01, harmonised to the reporting units the crop series carries from 1981-82: every district that enters the series later is merged into its parent for all years. Toba Tek Singh, for example, is reported separately from 1981-82, although it formally became a district on 1 July 1982, and is kept separate from Faisalabad. Selecting the modern districts first and merging, or merging first and applying the threshold to units, gives the same nineteen units, thirteen in Punjab and six in Sindh. Yields are area-weighted across a unit's members and weather is cell-weighted across their polygons. A unit-year enters only when every member reporting that crop in that period is present.
Weather. Rainfall is CHIRPS v2.0 monthly district means. Temperature is ERA5-Land daily district means via Open-Meteo.
Model. Log unit yield is fitted on July–August and September–October rainfall (per 10 mm), days at or above 40°C in May–June, days at or above 38°C in July–August and nights at or above 28°C in July–September, with unit intercepts and unit-specific quadratic trends fitted jointly, on units with at least fifteen observed years. Figure 1's lower panel converts each unit's residual to a percentage departure before averaging.
Sample. The crop record has gaps, so the 1981–2024 span holds 36 observed seasons for cotton: 465 unit-years on the thirteen Punjab units and 671 on all nineteen. Rice has 1,005 unit-years on 27 units, maize 759 on 20 and sugarcane 963 on 25, over 39 observed seasons.
Intervals. Standard errors are clustered by unit and by year. Every interval quoted, in the text and in Figure 2, uses a t multiplier with units-minus-one degrees of freedom (2.18 for thirteen units, 2.10 for nineteen, 2.06 for 27). As one check on the small number of clusters, a wild cluster bootstrap by unit (Rademacher, 1,999 draws) gives p = 0.015 for the Punjab cotton estimate and below 0.001 for the nineteen-unit estimate. It is that check, not a resolution of every small-sample concern.
Within-season check. Adding a fixed effect for each year to the Punjab model, so that districts are compared only with each other in the same season, gives −0.81 per cent per 10 mm (interval −2.43 to +0.81) on the same 465 unit-years.
Scenario. The Punjab coefficient (−2.26 per cent per 10 mm, interval −4.20 to −0.31) is applied to the 51.8 mm difference between the 1981–2010 and 2015–2024 means on the thirteen Punjab units, giving −11.0 per cent (−19.6 to −1.6). The sensitivity refits the same thirteen units without 1983 and 2022 (−0.81 per 10 mm, interval −1.75 to +0.14) and gives −4.1 per cent (−8.7 to +0.7). Both ranges carry only the coefficient's uncertainty, and both are illustrative scenarios under this model, not measured shares of the decline.
Variation explained. The five weather terms explain 14.7 per cent of the within-unit variation in detrended Punjab cotton yields, and 14.3 per cent on nineteen units.
Imports. The value of raw cotton imports in 2024-25 is from PBS's Revision of Monthly Import Trade Statistics (August 2026), Table 3. No import share or tonnage is given: PBS revised its import series for 2020-21 to 2024-25 after finding customs declarations omitted from its source data, and revised quantities by commodity are not yet published.
Irrigation sources. Mouza shares divide by all mouzas listed in each district in the Mouza Census 2020, Chiniot included with Jhang.
Not used. The physical cropped footprint from satellite (GLAD, 2000–2020) is published as a separate data release and is not used here.
Files. The numbers behind Figures 1 to 4 and the printed output of the yield model and its checks are downloadable below.
Sources
Agriculture Policy Institute (2022). Sugarcane Policy Analysis for 2022-23 Crop. Ministry of National Food Security and Research.
Akhtar, S. (1985). Effects of the timing and the number of sprays on cotton yields in Sind: an exploratory analysis. The Pakistan Development Review 24(2), 173–181.
Brouwer, C. and Heibloem, M. (1986). Irrigation Water Management: Irrigation Water Needs. Training Manual No. 3. FAO.
Crop Reporting Service, Agriculture Department, Punjab. Cotton Yield Analysis Punjab 2024-25.
Dawn (2023). Govt urged to keep promise of lint purchase. 12 October 2023.
Dawn (2023). TCP seeks ECC approval for cotton purchase. 14 October 2023.
Hussain, A., Farooq, J., Ahmad, S. and others (2018). Hypoxia tolerance studies for yield, fiber and physiological traits in cotton (Gossypium hirsutum L.). Journal of Cotton Research 1, 8. doi:10.1186/s42397-018-0008-8
Ministry of National Food Security and Research. Crops Area and Production (by districts) 2024-25.
Pakistan Bureau of Statistics (2026). Revision of Monthly Import Trade Statistics, 2021 to 2025. August 2026.
Pakistan Bureau of Statistics. Mouza Census 2020; district crop reporting series via na.data.gov.pk and Data Darbar.
Pakistan Institute of Development Economics. Climate Change Working Paper 9.
Data and methods downloads
- Figure 1 data: rainfall and cotton yield departures (CSV)
- Figure 2 data: rain coefficients, cotton and rice (CSV)
- Figure 3 data: area by district, 2000-01 and 2024-25 (CSV)
- Figure 4 data: irrigation sources by district, 2020 (CSV)
- Yield model output and the 11 November checks (text)
- PBS Agriculture Statistics (na.data.gov.pk) ↗
- CHIRPS v2.0 (UCSB-CHC) ↗
- ERA5-Land via Open-Meteo ↗
- Mouza Census 2020 (PBS) ↗
- MNFSR Agricultural Statistics of Pakistan 2024-25 ↗
- Source and method notes (Markdown)