In 2022-23, Kambar Shahdadkot’s rice production dropped to 16 per cent of its three-year pre-flood average, according to Pakistan Bureau of Statistics figures. In a February 2023 FAO survey, 91 per cent of households answering a question about their harvest reported gathering less than half their normal crop. By 2023-24 the district’s production had reached 136 per cent of its pre-flood average. Dadu followed a similar arc, from 36 per cent in the flood year to 106 per cent the following year.
UNOSAT’s satellite imaging detected flooding across at least 5 per cent of Kambar Shahdadkot in all eleven observation windows between late August 2022 and mid-January 2023. Across Sindh, it identified at least 49,000 square kilometres under water during August 2022, about 35 per cent of the province, inhabited by an estimated 18.6mn people. World Weather Attribution found that human-induced warming likely intensified the extreme rainfall, although its contribution could not be quantified reliably.
Food insecurity among farming households predated the floods: in six Sindh districts surveyed by FAO in spring 2022, between four and five in ten crop growers already faced moderate or severe food insecurity.
Four years on, the return of rice production offers an incomplete picture of recovery. How did the harvest fare where flooding was recorded most often, and did the gains endure? Did household food security recover, and for whom? And is Sindh better protected against another flood? We take these questions through UNOSAT’s satellite observations, the Pakistan Bureau of Statistics’ district crop series, successive FAO household surveys and the Sindh government’s development portfolio, which differ in the places, samples and periods they cover.
Windows under water, of 8
Share of district under water, by window
Where rice harvests fell hardest
The losses in Kambar Shahdadkot and Dadu formed part of a wider pattern. Among 42 rice-growing reporting areas in the country, the ten with the most persistently detected flooding produced 71 per cent of their pre-flood rice output in 2022-23, against 94 per cent in areas with fewer flood observations and 97 per cent where none met the threshold.
The comparison sets district crop statistics against UNOSAT’s eleven seven-day flood maps, spanning 25 August 2022 to 15 January 2023. An observation counts towards an area’s flood record if at least 5 per cent of its land was mapped as flooded; the areas are those with consistent crop records and at least 10,000 hectares of rice before the floods. Gaps of several weeks separate some observations, so the count captures how often substantial flooding was detected across an area, not how long particular fields remained under water. Even areas with no qualifying observation may have flooded: the record excludes July and early August, and anything below the threshold. Figure 1 plots the count against rice production in the three harvests since, relative to the 2019-20 to 2021-22 average.
By the following harvest the gap had largely closed. Production in the ten areas reached 114 per cent of its pre-flood average in 2023-24, with eight individually exceeding their earlier output, and the other groups rose to 121 and 117 per cent. Repeated observations of flooding went with much greater losses in the flood year but offered little indication of the scale of the rebound. Alternative thresholds, minimum rice areas and baseline periods leave this pattern intact.
The figures also reflect a pronounced geographical divide. All nine Sindh rice-growing areas in the sample were flooded in every window, and the tenth is the combined Jaffarabad and Sohbatpur area in Balochistan, so the comparison with less frequently flooded areas largely crosses into Punjab. A model that allows for province and peak flooded extent gives an imprecise flood-year estimate whose interval spans zero. How much repeated flooding itself explains the differences in production remains difficult to establish.
From the harvest to the households
Estimated food insecurity eased during 2023 in most of the repeatedly flooded districts surveyed by FAO. Nine of the twelve districts with comparable post-flood survey data recorded a decline. The median estimated prevalence of moderate or severe food insecurity fell from 59 per cent in early 2023 to 37 per cent by the end of the year.
Did that decline reach growers with little land as much as those with more? Six of these districts, Badin, Dadu, Mirpur Khas, Sanghar, Thatta and Umerkot, had also been surveyed in spring 2022, before the floods. Their household records allow a comparison of growers cultivating less than a hectare, about two and a half acres, with those cultivating at least a hectare, across the pre-flood survey and the two post-flood surveys of early and late 2023 (Figure 2).
Before the floods, estimated food insecurity stood at 49 per cent among growers cultivating less than a hectare, against 42 per cent among those with more land. In the first post-flood survey, both estimates rose to about 57 per cent, all but closing the difference between the groups. These are point estimates from fresh samples: they show the gap closing in that survey, not that holding less land had stopped mattering. By late 2023 the estimates had declined to 41 and 28 per cent respectively. The decline was considerably greater among those farming more land, at 29 percentage points against 16, so growers with less land once again had the higher estimated rate. Holding each survey’s district mix fixed narrows that late-2023 gap from 13 points to about 7. The surveys, however, reached different households in different agricultural seasons. They cannot establish whether the same families became more food secure, or whether the disparity between land groups ended wider or narrower than before the floods.
Households’ main sources of income also underwent a near reversal. Before the floods, 41 per cent of households in the six districts relied primarily on their own crops and 19 per cent on casual work outside farming. By early 2023 the shares were 19 and 40 per cent. By late 2023, crops were again the main source for 39 per cent, close to their pre-flood share.
Livestock production followed a similar course, from 53 per cent of households before the floods to 35 per cent in early 2023 and 48 per cent by late 2023. Yet casual work outside farming remained the main source of income for 27 per cent of households, against 19 per cent before the floods. These changes describe successive survey samples rather than the fortunes of individual households. Nor can the land comparison distinguish owners from tenants or sharecroppers: the released data record how much land growers can cultivate, not who owns it.
Recovery did not settle everything
The rebound in rice production did not last everywhere. By 2024-25, six of the ten persistently flooded areas were again producing less rice than before the floods, and the group’s total output had fallen to 95 per cent of its pre-flood average. Production in areas with fewer or no qualifying flood observations held above earlier levels, at 115 and 120 per cent. Figure 3 separates the land planted with rice from the amount produced.
The land under rice had scarcely changed: across the ten areas it stood at about 106 per cent of its pre-flood average in both 2023-24 and 2024-25. Yields, however, fell from 107 to 90 per cent of their earlier level, so the smaller harvest reflected less rice per hectare, not less land under rice. How much of that decline, if any, is a lingering effect of the 2022 floods rather than of water availability, weather and farming conditions in 2024-25 cannot be determined from the crop record.
One district figure is in dispute. The federal crop statistics credit Tando Muhammad Khan with 145,306 tonnes of rice from 16,082 hectares in 2023-24, a yield of about 9 tonnes per hectare and nearly three times its pre-flood average, while Sindh’s own statistical volume assigns the same area and production to neighbouring Tando Allahyar and records 1,436 tonnes for Tando Muhammad Khan. Without Tando Muhammad Khan, the remaining nine areas produced 71, 111 and 95 per cent of their pre-flood average across the three harvests. The disputed figure affects the scale of the recovery, not the finding that its gains were short-lived.
The floodplain in the budget
Rice regained its pre-flood share of the land under the three major summer crops. Across eleven persistently flooded reporting areas with comparable figures for rice, cotton and sugarcane, its share fell from 72 per cent before the floods to 63 per cent in 2022-23 and was back at 72 per cent by 2024-25. The return concealed differences between areas: in five of the eleven, at least one crop’s share still differed from its pre-flood level by more than three points.
The provincial government’s response appears in the Sindh Planning and Development Board’s 2026-27 portfolio, which lists 21 rural flood protection and drainage schemes that entered the FY2023-24 and FY2024-25 budgets, the first two presented after the floods. They encompass dhoro revivals, outfall and SCARP drains, saline tubewells, pumping stations, protective bunds and dikes, with an approved cost of Rs 35.6bn, of which Rs 11.8bn, roughly a third, is recorded as spent. Their listed start dates run from August 2024 to July 2025. Figure 4 compares approved cost and spending per resident across the 19 Sindh districts flooded in every observation window.
The allocations bear little relation to the extent of flooding at its peak. Umerkot, a quarter of which was inundated, has the highest approved cost per resident, Rs 8,360 for the Pithoro and Hingorno drains. Matiari follows at Rs 5,460, then Jacobabad at Rs 3,740, although floodwater covered 95 per cent of Jacobabad at the peak; its scheme, listed as starting in April 2025, has recorded spending of 3 per cent of its approved cost. Shikarpur and Kashmore, 91 and 70 per cent flooded at the peak, have no schemes in this selection.
The portfolio also lists a Rs 16.9bn dam on the Mazarani torrent above Kambar and protective spurs along the Indus, both marked Not Approved. Nor does the selection capture every aspect of the government’s response: schemes completed or dropped before 2026-27 are absent, and the Rs 93bn housing programme is recorded province-wide. Approved costs are budget ceilings, and recorded expenditure does not establish whether drains and defences have been completed or are working.
Beyond recovery: the test of adaptation
Rice production rose above pre-flood levels within a year across the ten persistently flooded reporting areas, and estimated food insecurity eased in most of the repeatedly flooded Sindh districts surveyed by FAO. These are two observations with their own geography. The surveyed households grew crops of many kinds, and the surveys do not trace the people who produced the rice in the harvest series; only three of the six districts in the land comparison, Badin, Dadu and Thatta, are among the ten rice areas, and their combined output in 2023-24 was 93 per cent of the pre-flood average. These were signs of recovery, although the later fall in yields and the disparities between growers complicate any account of lasting progress. Adaptation demands a more exacting measure: whether fields, infrastructure and households are better equipped to withstand another spell of extreme rainfall, which a warming climate can intensify.
On that, the evidence is thin. Rice has regained its share of the land under the three main summer crops, but the figures reveal little about changes in varieties, sowing dates or irrigation. Growers with less than a hectare again had the higher estimated rate of food insecurity by late 2023. FAO’s late 2023 survey asked farmers who had noticed changes in the climate how they were responding, including through climate-resistant varieties, altered sowing times and drainage, but those answers are absent from the released data. The selected drainage schemes have recorded start dates from August 2024 to July 2025 and no measure of completed works.
Four years after the floods, the missing account is of protection on the ground: which drains are operating, which fields they serve, and whether the households depending on them face less exposure to another flood.
Flood persistence. UNOSAT NOAA-20 VIIRS flood products. The August 2022 provincial figures are UNOSAT’s cumulative 1–31 August 2022 extent for Sindh (product 3389: 49,041 km², 34.8 per cent of analysed area, 18.57mn people in the flooded area on WorldPop 2020). The persistence series uses the eleven seven-day products from 25 August 2022 to 15 January 2023 on a consistent district frame, with gaps between some windows; the dates of all eleven are listed in the issue’s sources note. Persistence is the number of windows in which at least 5 per cent of a unit’s area was mapped as flooded; 2 and 10 per cent thresholds, the share remaining in mid-January and the mean observed share are reported as sensitivities. District area is not rice-field area.
Crops. PBS Agricultural Statistics via na.data.gov.pk, stable reporting areas, excluding the misfiled Sujawal provincial-total row. The baseline is the 2019-20 to 2021-22 mean. Figures 1 and 3 use the 42 rice areas with at least 10,000 hectares, grouped by observed flood windows. The federal series credits Tando Muhammad Khan with 145,306 tonnes of rice on 16,082 hectares in 2023-24 (about 9 tonnes per hectare); Sindh’s published 2023-24 statistics assign the same area and production to Tando Allahyar and record 1,436 tonnes for Tando Muhammad Khan, and the federal series carries no Tando Allahyar rice row after 2014-15. The conflicting attributions are unresolved. The analysis keeps the federal series and reports results excluding Tando Muhammad Khan from the baseline and all three harvests. Group indices divide combined area or production by the corresponding pre-flood average. The Kharif crop-mix result uses the eleven persistent areas with stable reporting from 2019-20 to 2024-25 (see Crop mix below); eight persistent areas are excluded because they lack a complete record for at least one of the three crops, most often because PBS began reporting their rice only in 2021-22 or later, which is a change in reporting and not in cropping. Maize is excluded because the source has no 2020-21 or 2024-25.
Climate. World Weather Attribution, “Climate change likely increased extreme monsoon rainfall, flooding highly vulnerable communities in Pakistan”, 14 September 2022 (peer-reviewed version Otto et al., 2023). The study’s observational trend analysis finds maximum five-day rainfall over Sindh and Balochistan about 75 per cent more intense than in a climate without 1.2°C of warming; some climate models suggest an increase of up to 50 per cent attributable to climate change. The authors state that the overall role of human-induced warming could not be reliably quantified, because of rainfall variability and model limitations. This article does not attribute any district or household outcome to climate change.
Households, district tables. FAO Data in Emergencies monitoring, Pakistan rounds 4 (12 February to 15 March 2023) and 5 (22 November 2023 to 6 January 2024), public district tables retrieved 23 October 2026. Kachhi and Jaffarabad are excluded for unresolved boundary changes, leaving twelve persistently flooded districts in both rounds; the medians are of FAO’s district estimates.
Households, microdata. FAO DIEM household microdata, Pakistan rounds 2 to 6, used under FAO’s terms for research and aggregate reporting only; no household record is published. Data in Emergencies Hub, FAO. Data in Emergencies Monitoring Household Survey, Pakistan, rounds 2 (2021), 3 (2022), 4 (2023), 5 (2023) and 6 (2024); datasets downloaded from https://microdata.fao.org. The rounds are repeated cross-sections: no household identifier links them. Figure 2 uses the six persistently flooded districts present in rounds 2 to 5: the land comparison uses rounds 3 to 5, the income and livestock panel rounds 2 to 5. The land comparison is restricted to households reporting crop production in the past year (246 to 624 households per point; under one hectare 246 to 305, one hectare or more 556 to 624); the income and livestock panel covers all households. Fieldwork periods differ (August to September 2021, March to April 2022, February to March 2023, November 2023 to January 2024), and seasonal effects cannot be separated from post-flood change. Cultivable land (crp_landsize, crp_landsize_ha) is reported by crop growers only. Rounds 2 and 3 record it in labelled bands (under 1 ha, 1 to 5 ha and above); round 4 stores four values (0.5, 2.5, 7 and 11 ha) that appear to represent bands whose cut points the released metadata do not confirm; round 5 records hectares. Two groups are therefore used, under 1 hectare and 1 hectare or more, as a provisional harmonisation. Round 6 is not used. A five-hectare split was tested but rests on 40 to 90 households per survey and is not shown. Food insecurity is the household-weighted mean of FAO’s probability of moderate or severe food insecurity (p_mod); weighted whole-district estimates reproduce FAO’s published district figures. Sampling strata and clusters are not released, so design-based intervals cannot be computed. Income categories are FAO’s main-income codes for the previous three months; livestock is whether the household produced livestock in the past twelve months, not current ownership. Two checks for changing district composition leave the convergence and later separation in place. Holding each round’s district mix of crop growers fixed at the early 2023 distribution gives gaps (under one hectare minus one hectare or more) of about 4.3, 1.0 and 6.8 percentage points in the three surveys. A household-weighted regression comparing the land groups within districts gives about 5.2, 0.2 and 9.6 points. The round 5 questionnaire contains climate adaptation and land tenure questions whose answers are not in the released files.
Development programme. Sindh Planning and Development Board, ADP portfolio for FY2026-27 (3,715 schemes, data as imported on 1 October 2026). The year a scheme entered the programme is read from its identifier. Sindh’s FY2022-23 budget was presented on 14 June 2022, before the floods, so the post-flood cohort is schemes with FY2023-24 and FY2024-25 identifiers. Because the portfolio is live, post-flood schemes completed or dropped before 2026-27 are absent, so the cohort undercounts; it is described as listed projects, not as a reconstruction of the two budgets. Every scheme in the period that is an Irrigation Department scheme or has a flood, drain, bund, protection, retaining wall, escape or pumping term in its title (292 schemes) was read and classified by hand. Twenty-one are rural flood protection and drainage, including saline tubewells in Badin, which drain waterlogged land vertically as in the SCARP schemes; the rest are canal repair or lining, town and village drainage and sewerage, groundwater supply, or unrelated works such as a college whose place name contains “dhoro”. Jacobabad’s scheme combines a branch canal with rain drainage and is counted in full. The Sohni dhoro revival is split among the three districts its title names. Schemes marked Not Approved are excluded. Cost is approved total cost and spending is cumulative expenditure as recorded in the portfolio, whose cutoff date is not stated; neither measures completed or functioning works. Start dates are as recorded in the portfolio. The housing programme (Rs 75bn World Bank, Rs 11.2bn ADB and Rs 7bn EIB lines) and STAR WASH (Rs 25.2bn) are province-wide. The full audit is among the files published with this piece.
Crop mix. Shares are of the combined area under rice, cotton and sugarcane, not all Kharif crops. In four of the 33 area–crop combinations a crop was never reported in 2019-20 to 2024-25 and is counted as zero. On the seven areas that report all three crops in every year, rice’s share was 80.6 per cent before the floods and 81.5 per cent in 2024-25.
Adaptation checks. Two pre-flood Mouza Census 2020 measures, the number of major crops named per village and the share of villages with both canal and tubewell water, were tested against the 2023-24 rebound after allowing for persistence, peak extent and province; neither gives a consistent explanation. The findings do not attribute the floods to human-induced climate change or evaluate any named intervention. UNOSAT products are CC BY-SA. Crop and development data are government statistics; FAO district tables are public; FAO household microdata are used under FAO’s access terms, reported only as aggregates; the climate finding is cited from World Weather Attribution.
Data and methods downloads
- Map data: flood persistence grid, Kambar Shahdadkot and Dadu, eight windows (JSON, CC BY-SA)
- Figure 1 data: rice production by observed flood windows, 42 areas (CSV)
- Figure 2 data: food insecurity by cultivable land, six districts (CSV)
- Figure 2 data: main income source and livestock, six districts (CSV)
- Figure 3 data: rice area and production by persistence group (CSV)
- Figure 4 data: drainage and protection schemes per head, 19 districts (CSV)
- Hand audit of the 292 candidate ADP schemes (CSV)
- Matched-district medians, FAO rounds 4 and 5 (CSV)
- UNOSAT flood products, Pakistan 2022 ↗
- PBS Agriculture Statistics (na.data.gov.pk) ↗
- FAO Data in Emergencies Hub ↗
- FAO microdata catalogue ↗
- Sindh ADP portfolio (P&D Board) ↗
- World Weather Attribution, Pakistan 2022 ↗
- Source and method notes (Markdown)