# Sources & method — Vol 1, Issue 4

One section per file in this folder: what the indicator is, which
release it came from, what it does not cover, and anything the figures
should carry as a caveat.

Data freeze: three weeks before publication. After the freeze, revisions
to the underlying series go to the corrections log, not into this file.

## cotton_fig1_data.csv to cotton_fig4_data.csv, cotton_model_results.txt — Wetter summers in Punjab’s shrinking cotton belt

Figure inputs and model output for the feature, from `Adaad/data/pk-crops/drivers-2026-10-20/`
(scripts `climate_yield_panel_v2.py`, `climate_yield_test_v2.py`, `build_figures.py`,
`figures-v8/plot_figures_from_data.py`, 11 November 2026).

- **cotton_fig1_data.csv** — year; mean July–August rainfall, mm, across the thirteen Punjab cotton units
  (Punjab districts with at least 20,000 ha of cotton in 2000-01, harmonised to the 1981-82 reporting units), CHIRPS v2.0
  via ClimateSERV, 1981–2025; mean departure of cotton yield from each unit's fitted quadratic trend,
  as a proportion (each unit's residual converted with exp(r) − 1 before averaging), 1981–2024; the
  number of units reporting. Years with no crop record are blank.
- **cotton_fig2_data.csv** — coefficient and standard error of July–August rainfall (per 10 mm) in the
  yield model: cotton on the thirteen Punjab units, cotton on all nineteen units, cotton on the Punjab
  units without 1983 and 2022, cotton on the Punjab units with a fixed effect for each year, and rice on 27 units; with the 95 per cent interval and the rule used for
  it. Log-point coefficients × 100; standard errors clustered by unit and year; intervals use a t multiplier
  with units-minus-one degrees of freedom (as in the method note and the text).
- **cotton_fig3_data.csv** — area sown, thousand hectares, 2000-01 and 2024-25 and the change, by crop, for
  the sixteen Punjab districts with at least 40,000 ha of cotton in 2000-01, Chiniot counted with Jhang;
  blank where a district-crop has no record at one date. Cotton, rice and sugarcane from PBS provincial crop
  reporting via Data Darbar; maize 2024-25 from MNFSR, Crops Area and Production 2024-25, Table 3 (Kot Addu
  and Taunsa folded into Muzaffargarh and Dera Ghazi Khan).
- **cotton_fig4_data.csv** — mouzas listed in 2020 and the shares reporting a canal and a tubewell, by
  district and for all sixteen (12,008 mouzas), Chiniot with Jhang. Mouza Census 2020, PBS.
- **cotton_model_results.txt** — the printed output of the yield model on the harmonised panel, and the
  complete run of the 11 November checks (`review_checks_2026-11-11.py`: Punjab-only, Sindh-only and
  Punjab-without-1983-and-2022 cotton estimates, t(G−1) intervals, wild cluster bootstrap, the two
  scenarios with their intervals, selection-rule equivalence).

Crop reporting: PBS Agriculture Statistics via na.data.gov.pk, pulled 27 September 2026; the Sujawal
rice row that carries a provincial total is excluded; cotton production in bales of 170 kg. Rainfall:
CHIRPS v2.0 (UCSB-CHC), public domain. Temperature: ERA5-Land (Copernicus C3S) via Open-Meteo, CC BY 4.0.
Imports in the text: value of raw cotton imports in 2024-25 from PBS, Revision of Monthly Import Trade
Statistics (August 2026), Table 3; no quantities or shares are used, pending PBS's revised series.
Crop calendar in the text: Crop Reporting Service, Agriculture Department, Punjab, Cotton Yield Analysis
Punjab 2024-25 (sowing March–May, picking August–November). Waterlogging: Hussain and others (2018),
Journal of Cotton Research 1:8, doi 10.1186/s42397-018-0008-8. Seed cotton prices in October 2023: Dawn,
12 and 14 October 2023. Prices in the text:
MNFSR Agricultural Statistics of Pakistan 2024-25, tables 142–148, and the World Bank Pink Sheet
(CC BY 4.0). Credit: "PBS crop reporting via Data Darbar; CHIRPS (UCSB-CHC) via ClimateSERV;
ERA5-Land (Copernicus C3S) via Open-Meteo; Mouza Census 2020, PBS; Adaad calculation".


## flood_fig1_*.csv to flood_fig4_*.csv, flood_adp_protection_audit.csv, flood_diem_matched_districts.csv — After the flood

Figure inputs for the feature, from `Adaad/issues/2026-11-issue-4/after-the-flood/` (the recovery test of
25 October, the hunger-by-land check of 28 October and the adaptation check of 25 October), the UNOSAT
rebuild in `Adaad/data/pk-floods-2022/rebuild-2026-10-18/`, the Sindh ADP tables in `Adaad/data/pk-sindh-adp/`
and the FAO DIEM microdata in `Adaad/data/pk-diem-microdata/` (not published). Written by
`site-2026-11-01/draw_site_figures.py` beside the figures.

- **flood_fig1_rice_by_persistence.csv** — the 42 rice reporting areas with at least 10,000 hectares of rice
  before the floods and stable reporting: province; the number of the eleven UNOSAT seven-day windows in which
  at least 5 per cent of the area was mapped as flooded; the peak flooded share; pre-flood rice area and
  production (2019-20 to 2021-22 mean); and production in 2022-23, 2023-24 and 2024-25 as an index of that mean.
  Jaffarabad and Sohbatpur form one reporting area.
- **flood_fig2_food_insecurity_by_land.csv** — household-weighted share of crop-growing households in moderate
  or severe food insecurity (the mean of FAO's `p_mod`), by the land the household can cultivate (under
  1 hectare; 1 hectare or more), in Badin, Dadu, Mirpur Khas, Sanghar, Thatta and Umerkot, FAO DIEM rounds 3
  (March–April 2022), 4 (February–March 2023) and 5 (November 2023 – January 2024), with the number of
  households behind each estimate. Rounds 2 and 3 record land in bands and round 4 in four provisional values,
  so the two groups are a harmonisation; round 5 records hectares.
- **flood_fig2_income_livestock.csv** — all households in the same six districts, rounds 2 to 5: the share
  naming each source as their main income over the previous three months, the share that produced livestock
  in the past twelve months, and the share growing crops.
- **flood_fig3_rice_area_production.csv** — combined rice area, production and the derived yield for the 42
  areas grouped by observed flood windows (0, 1–10, 11), for the three harvests, with the pre-flood means and
  the indices the figure draws. The Tando Muhammad Khan 2023-24 row, which matches the ministry's print
  tables but implies about 9 tonnes per hectare, is retained and flagged in the text.
- **flood_fig4_adp_protection_by_district.csv** — the 19 Sindh districts flooded in every observed window:
  peak flooded share, WorldPop 2020 population, and the number, approved cost and recorded expenditure
  (Rs bn and Rs per head) of the audited rural flood protection and drainage schemes that entered the
  FY2023-24 and FY2024-25 budgets and are listed in the 2026-27 portfolio.
- **flood_adp_protection_audit.csv** — the hand audit of the 292 candidate schemes (every Irrigation
  Department scheme of those two budget years, plus any scheme with a flood, drain, bund, protection,
  retaining wall, escape or pumping term in its title): category (FPD = rural flood protection and drainage,
  CANAL_RESTORE, CANAL_NEW, URBAN, OTHER), the reason, districts as tagged and as used, approval status,
  cost, expenditure, recorded start date, sector and title, as in the portfolio.
- **flood_persistence_map.json** — the grid behind the unnumbered flood persistence map (added 12 November 2026):
  Kambar Shahdad Kot (PK709) and Dadu (PK703) and their surroundings on a 0.004 degree grid (about 400 m), with
  one bit per cell for each of eight UNOSAT NOAA-20 VIIRS seven-day windows, 25 August to 20 November 2022
  (polygons from the shapefile packages of products 3346 and 3457). The three winter windows are not mapped:
  their polygons (products 3470, 3471) were not downloaded, and they appear only as UNOSAT's published district
  areas. Also in the file: a district index grid, normal water (UNOSAT WaterExtent minus FloodExtent from the
  Sentinel-2 31 August and Sentinel-1 6 and 15 August scenes, which cover the eastern two-thirds of the frame),
  simplified COD-AB district outlines, GeoNames place coordinates, the published district areas per window and
  the two districts' rice indices from flood_fig1. Check: gridded flooded area reproduces UNOSAT's workbook
  figure for both districts in all eight windows to within 0.7 per cent; cloud obstruction touches 184 of
  178,200 cells, none flooded in that window. Result: 25.0 per cent of Kambar Shahdad Kot and 16.3 per cent
  of Dadu were mapped as flooded in all eight windows. Built by the scripts in
  `Adaad/issues/2026-11-issue-4/03-after-the-flood/map-2026-11-12/code/`. Derived from UNOSAT layers, so CC BY-SA.
- **flood_diem_matched_districts.csv** — medians of FAO's published district estimates across the twelve
  persistently flooded districts in both rounds 4 and 5 (Kachhi and Jaffarabad excluded for boundary
  changes), with the count of districts falling and rising.

Flood persistence: UNOSAT NOAA-20 VIIRS seven-day products on the COD district frame, eleven windows:
25–31 August, 1–7, 8–14, 15–21 September, 26 September – 2 October, 3–9 and 11–17 October, 14–20 November,
25–31 December 2022, 1–7 and 9–15 January 2023. There is no observation for 22–25 September, 18 October –
13 November or 21 November – 24 December, so persistence is "in every observed window". The August provincial
figures in the text are UNOSAT's cumulative 1–31 August 2022 product (3389). UNOSAT products are CC BY-SA, so
the flood columns are CC BY-SA ("Analysis: United Nations Satellite Centre (UNOSAT)"). Crops: PBS Agriculture
Statistics via na.data.gov.pk, stable reporting areas, the misfiled Sujawal provincial-total row excluded.
Households: FAO Data in Emergencies Monitoring household surveys, Pakistan, rounds 2 to 5; the microdata were
downloaded from https://microdata.fao.org under FAO's terms for research and aggregate reporting, and only the
aggregates above are published; the public district tables are from the Data in Emergencies Hub. Development
programme: Sindh Planning and Development Board, ADP portfolio FY2026-27, as imported on 1 October 2026.
Population: WorldPop 2020. Climate finding cited in the text: World Weather Attribution, 14 September 2022
(Otto et al., 2023). Credit: "UNOSAT; PBS Agricultural Statistics via na.data.gov.pk; FAO Data in Emergencies
Monitoring; Sindh Planning and Development Board; WorldPop; Adaad calculation".

## solar_fig2_*.csv to solar_figS2_*.csv — Pakistan’s solar boom meets its power bill

Figure inputs for the Big Read, from `Adaad/issues/2026-11-issue-4/solar-and-the-tariff/analysis-2026-10-21/`
(the DISCO ledger, recovery-and-price and tehsil folders), `Adaad/data/pk-nepra-loadshedding/`,
`Adaad/data/pk-disco-crosswalk/` and the HIES rebuild of 18 October.

- **solar_fig2_sales_by_year.csv**, **solar_fig2_sales_by_company_category.csv** — units sold by the ten
  state distribution companies, GWh, by year and by company and consumer category, FY16 to FY25. NEPRA State of
  Industry Reports 2017 to 2025, Table 68 and equivalents; FY25 transcribed from the printed table and reconciled
  to its totals. K-Electric excluded.
- **solar_fig3_generator_invoices.csv** — capacity and energy charges invoiced by all generators on the CPPA-G
  system, Rs bn, FY19 to FY25, with the ten companies' purchases and sales; the 2025 edition's columns were
  relabelled by position (flagged). NEPRA State of Industry Reports 2023 to 2025.
- **solar_fig4_fig5_companies.csv** — by company, FY25: units billed and amounts billed and realised, total and
  domestic (Table 67); recovery and losses; reported load shedding with the regulator's monitored range
  (Performance Evaluation Report FY2024-25); net-metering connections sanctioned FY23 to FY25 per 1,000
  consumers (Table 80); census solar-panel-lit share of households in the districts assigned to each company
  (Census 2023 Table 22, crosswalk in `pk-disco-crosswalk`); sales change FY22 to FY25; capacity transfer charge
  per kWh FY25 (Table 65). Satellite columns removed.
- **solar_fig6_recovery_by_category.csv** — amount realised as a share of amount billed, by company and consumer
  category, FY25. NEPRA State of Industry Report 2025, Table 67, transcribed from physical pp 203-204.
- **solar_fig7_capacity_transfer_charges.csv** — capacity and energy transfer charges invoiced by CPPA-G to each
  company and the energy taken, FY24 and FY25, with per-kWh values. FY24 from Table 64 of the 2024 edition
  (monthly rows summed; SEPCO and QESCO have separate rows, printed pp. 205 and 206, and were shown combined
  here until 18 November 2026); FY25 from Table 65 of the 2025 edition, transcribed.
- **solar_fig8_fig9_tehsils.csv** — by census-2023 tehsil polygon: populated villages and the counts answering
  All, Mostly, Some or None to the electricity-availability question and reporting solar as an alternative
  source (Mouza Census 2020 Part VI, PBS; 112 invalid or missing answers excluded); households and lighting by
  electricity and solar panel (Census 2023 Table 22); the source of the village figure (`mz_source`: tehsil,
  parent-shared, district, none) and the four-class typology at 50 per cent of villages and 25 per cent of households.
  Mauripur and Baldia (Keamari, Karachi) share a parent with seven villages, below the ten needed for a village
  figure: `mz_source` is `none`, and their village percentages and class are blank.
  The four codes were reconciled against the labelled counts the PBS dashboard returns per tehsil: exact in
  538 to 540 of 541 tehsils, the exceptions being Fort Abbas, Yazman and Liaquatpur, where the dashboard's
  Cholistan pseudo-district splits villages across two codes (`check_electricity_codes.py` beside the microdata
  in `Adaad/data/pk-mouza-2020/microdata/`, with Form 11).
- **solar_figdata.json** — the numbers behind the interactive Figures 2 to 9, as drawn (Figure 1 is the system schematic): the series above
  reshaped by `build_figdata.py` (in the Adaad folder's 02-solar-and-the-tariff/site-2026-11-18), with the FY24 capacity charges, the
  tehsil rank correlations and the classification thresholds. Nothing in it is not in the CSVs.
- **solar_tehsil_map.json** — Figure 9's geometry: the 2023 census tehsil polygons (PBS, as used for the
  Data Darbar tehsil map), simplified to 0.012 degrees and rounded to three decimals, keyed by tehsil code;
  Occupied Kashmir omitted. No attribute beyond name and province.
  Satellite columns removed.
- **solar_figS1_lighting_by_tenth.csv** — main source of lighting by population tenth, HIES 2024-25 microdata,
  household weights; tenths on consumption per adult equivalent.
- **solar_figS2_census_districts.csv** — households and lighting by electricity and solar panel by district,
  Census 2023 Table 22, on the 160-district frame, with the company assignment. Satellite columns removed.

The article quotes statistics from TransitionZero's distributed solar layer (CC BY-NC 4.0, dataset dated
October 2025) and Open Energy Transition's earthpv atlas (ODbL; github.com/open-energy-transition/earthpv;
the release published on 13 November 2026, best estimate 34,469 MWp, extracted 18 November 2026).
Those licences do not permit release under CC BY 4.0, so no table carrying their values is published here.
The five supplementary figures (S1 to S5) were dropped on 18 November 2026; the two `solar_figS*.csv`
files stay as the data behind the method note's household and district figures. Credit: "NEPRA State of Industry Reports and Performance
Evaluation Report; Census 2023 and Mouza Census 2020, PBS; HIES 2024-25 microdata, PBS; Adaad calculation".

## smog_calendar_figdata.json, smog_air_daily.json, smog_context_monthly.csv, smog_air_sensitivity.csv — The coast is clearer: Karachi emits more, but Lahore breathes worse

- **smog_calendar_figdata.json** — the numbers behind Figure 1 as drawn. Per city: fine particles (PM2.5)
  and nitrogen oxides from EDGAR Cities (2025 release, data year 2024) and sulphur dioxide from EDGAR
  v8.1 2022 gridmaps summed over the GHSL urban centre, per person (GHSL population), by source; each
  annual total split across the months by Adaad with EDGAR's monthly profiles for Pakistan (2018–2022
  average); and the city daily PM2.5 series below. Licence CC BY 4.0. Credit: European Commission JRC
  (EDGAR, GHSL); Pakistan Air Quality Initiative / Hawanama; Adaad calculation.
- **smog_air_daily.json** — city daily PM2.5, 1 Nov 2025 – 16 Oct 2026 (350 days): median across the
  Hawanama stations in Lahore (78 devices, 73 locations) and Karachi (27 devices, 24 locations), at least
  three a day, with the number reporting each day. Built from the Hawanama snapshot of 17 Oct 2026
  (Adaad/data/pk-air-quality/hawanama-stations-2026-10-17) by build_air_daily_v2.py. Daily values are
  UTC-day means of hourly readings; 17 October is dropped as incomplete. CC BY 4.0, credit Pakistan Air
  Quality Initiative / Hawanama.
- **smog_context_monthly.csv** — the two context rows of Figure 1, by city and month over the same 350 days
  (also in smog_calendar_figdata.json under `ctx`). Fire detections: NASA VIIRS 375 m active fires (nominal and
  high confidence) within 200 km of the city centre, mean per observed day, with the days missing from the Global
  Forest Watch copy and the 2013–2024 mean for the same calendar month. Weather: ERA5 boundary layer height and
  10 m wind speed at the nearest grid point (Open-Meteo historical archive), daily means over UTC days, their
  product in thousand m²/s (ventilation), and days with mean wind below 2 m/s. Built 17 Nov 2026 by
  build_context_rows.py from Adaad/data/pk-air-quality/smog-coast-2026-11-17/. CC BY 4.0. Credit: NASA VIIRS
  active fire alerts, accessed from Global Nature Watch on 17 Oct 2026 (NASA FIRMS); ERA5, Copernicus Climate
  Change Service, via Open-Meteo; Adaad calculation.
- **smog_air_sensitivity.csv** — unhealthy-or-worse, hazardous and good-or-moderate day counts under
  alternative station rules (single-hour station days dropped, fixed panels, inside the urban centre
  only, every station inside the urban centre).

## notebook_*.csv, notebook_mangrove_figdata.json — How a picture from orbit becomes a number

Figure inputs for the Method Notebook, from `Adaad/data/pk-mangroves/` (the two satellite records, pulled
22 October 2026) and the exercise in `Adaad/issues/2026-11-issue-4/notebook-satellite/exercise-2026-11-01/`.

- **notebook_mangrove_gmw_annual.csv** — Global Mangrove Watch v4.1.12 (Bunting, Hilarides, Rosenqvist et al.
  2026, Zenodo 10.5281/zenodo.21346457, CC BY 4.0): Pakistan's mangrove extent every year 1985–2025 in square
  kilometres, with the record's 95 per cent bounds, from its country statistics workbook.
- **notebook_mangrove_gilani_sites.csv** — Gilani et al., Mangrove Cover Dataset of Pakistan 1990–2020 (Zenodo
  10.5281/zenodo.10732690, CC BY 4.0): mangrove, water and other area for each of the five sites every five
  years. `mangrove_km2` is recomputed from each pixel's latitude (the Indus delta raster is on a degree grid,
  so a pixel is 27.4 by 30.0 metres at 24°N, not 30 by 30; areas are on a spherical Earth of radius 6,371.0 km, the same formula in every notebook file); `mangrove_km2_as_published` reproduces the
  dataset's own attribute tables.
- **notebook_mangrove_comparison.csv** — the two records side by side for the seven Gilani years.
- **notebook_sujawal_patch_check.csv** — for each district on the 2023 frame and each of the six five year steps from 1990: mangrove
  area at the end of the step, area new since the earlier map, area dropped, and the new area split by distance
  from the nearest pixel classed as mangrove in the earlier map (within 100 m, 100–500 m, 500 m–2 km, over 2 km)
  and by the earlier map's class. Every pixel, offshore included, goes to the nearest district polygon, so
  Thatta here includes Karachi's Keamari pixels. Earlier maps are resampled onto the 2020 grid by nearest
  neighbour.
- **notebook_new_patches.csv** — patches of 2020 mangrove (pixels connected edge to edge) that contain no
  pixel classed as mangrove in 2005, 0.5 km² and over, with area, centroid and district.
- **notebook_s2_threshold_curve.csv** — area above each NDVI threshold in a 17 by 16 km Sentinel-2 L2A window
  east of Keti Bandar (67.46–67.63°E, 23.99–24.14°N), 29 October 2026, bands 4 and 8 at 9 m in UTM 42N from
  the Copernicus Browser (Copernicus Sentinel data 2026).
- **notebook_distance_curves.csv** — behind the distance bands in the Sujawal section. For Sujawal and Thatta, the
  area classed as mangrove in 2020 and not in 2005 that lies within each distance (every 50 m to 3 km) of
  mangrove mapped in 2005, the area beyond it, and the share within; from the Gilani et al. Indus delta
  rasters, built by `build_exercise_data.py` in the exercise folder.
- **figures/notebook-s2-ndvi-window.png** — the image behind the threshold exercise after Figure 4: the same
  Sentinel-2 window averaged to a quarter of its resolution (about 36 m), with the vegetation index in the red
  channel as (NDVI + 0.2) × 255 and the Gilani et al. 2020 mangrove class in the green channel. Built with the
  extra columns of notebook_s2_threshold_curve.csv (area above each threshold that the 2020 map also classes as
  mangrove, and the map's mangrove area in the window, both at full resolution) by `build_threshold_exercise.py`
  in the exercise folder.
- **notebook_mangrove_figdata.json** — the two series and the district stack behind Figure 5, built by
  `build_mangrove_figdata.py` in `issues/2026-11-issue-4/mangroves/figure/`.

Not published: the Gilani rasters themselves (on Zenodo), the Sentinel-2 bands (reproducible from the
Copernicus Browser with the window and date above) and the Google Earth historical views used to check
the three largest new patches (Google's imagery; dates in the method note).

## District profiles: North Waziristan and Jaffarabad

- `district_profiles_2026_11.csv`: Table 1. PBS, Census 2023, Tables 22 and 23 (households by source of light, cooking fuel and drinking water; "improved" is PBS's count); Data Darbar multidimensional poverty headcount (PSLM 2019-20) and travel time by road to the nearest health facility, provinces and Pakistan as the median district; UNOSAT seven-day flood peak; UMD GLAD land cover v2, cropped land 2020.
- `districts_fig1_north_waziristan_tree_loss.csv`: Figure 1. Hansen/UMD/Google/USGS/NASA tree cover loss v1.13 and UMD tree cover loss from fires, via the Global Forest Watch Data API, 30 per cent canopy, Census-2023 district frame. CC BY 4.0.
- `districts_fig2_jaffarabad_flood.csv`: Figure 2. UNOSAT flood extents from NOAA-20 VIIRS, eleven seven-day windows, 25 August 2022 to 15 January 2023; population shares use WorldPop 2020.
- Food security figures in the text: FAO Data in Emergencies (DIEM) Pakistan household surveys, rounds 4 and 5, published district aggregates (weighted mean probability of moderate-or-severe and severe food insecurity on the FIES), retrieved 23 October 2026. Not redistributed here.
