Adaad · Volume 1, Issue 4 · Wednesday 18 November 2026 · Climate, energy and resources
Climate/Volume 1, Issue 4 · Feature

Pakistan’s summer nights have warmed more than its days

In a reconstruction of past weather covering the whole country, Pakistan’s May to September nights were 0.4 °C warmer over 2011–25 than over 1981–2010, while its days were almost unchanged. Shorter records from 19 weather stations support the night–day contrast, but do not establish where nights have warmed most.

Hiba Sameen · analysis
Adaad data desk · charts and files
Wednesday 18 November 2026 · Volume 1, Issue 4
9 min read · Download the data (CSV)

ERA5-Land, a dataset from the European weather forecasting centre, reconstructs past weather for every part of Pakistan by combining a weather model with satellite and station observations, a method known as reanalysis. In it, Pakistan's summer nights have warmed and its summer days have barely moved. Averaged over May to September and over the whole country, the night-time minimum was 0.4 °C higher over 2011–25 than over 1981–2010, and the daytime maximum 0.04 °C higher. Weather stations, over shorter windows, show the same ordering. Nights rose more than days at 15 of 19 stations with comparable records. Across the sample, the median gap between night and day changes was about 0.33 °C, compared with 0.34 °C in the reanalysis on the same dates.

The two records

The first record is ERA5-Land, produced by the European Centre for Medium-Range Weather Forecasts on a grid of about 9 km. It runs a model of the land surface driven by ERA5, the centre's global reanalysis, so its observations come in through ERA5 rather than directly. Daily maximum and minimum temperatures were averaged over every grid cell in Pakistan from 1981, the start date chosen for this analysis, to 2025.

The second is the daily record of Pakistan's synoptic weather stations, as archived by the US National Centers for Environmental Information in its Global Summary of the Day. Most of these stations are at airports, and their records have gaps. Because ERA5 assimilates station observations, the two records are not fully independent, but the station series do not pass through a weather model.

Unless otherwise stated, reanalysis changes compare 2011–25 with 1981–2010, the fixed reference period chosen for this analysis. Station comparisons use the qualifying years available at each site, described below. The World Meteorological Organization's current standard is 1991–2020. Each monthly anomaly is the difference from that location's average for the same calendar month in 1981–2010.

Nights have warmed more than days

Figure 1 shows the reanalysis record for Pakistan across the whole year, nights above and days below, each year coloured by its anomaly. Over the whole year the gap is smaller than in summer: nights were 0.4 °C warmer over 2011–25 than over 1981–2010 and days 0.2 °C warmer. The warmest nights in the record were in 2024 and 2025, at 0.9 and 1.0 °C above the 1981–2010 average.

Figure 1 · Warming stripes · interactive
Pakistan’s nights have warmed about twice as much as its days since 1981–2010
2025
Yearly mean of the daily minimum (nights) and daily maximum (days) across Pakistan, minus the 1981–2010 mean, °C, 1981–2025. Area-weighted mean across Pakistan in ERA5-Land. Switch to May–September or press play; hover, tap or focus a stripe for the year. Source: ERA5-Land (Copernicus C3S) via Open-Meteo; Adaad calculation.

Figure 2 shows in which months the change falls. Nights have warmed in every month except December and January, most in October (0.8 °C) and least in May (0.2 °C) and August (0.3 °C). Days have warmed in February, March and October and barely changed from May to September. Counts beyond fixed thresholds point the same way. Counted from the average temperature of each of the 160 areas the national series is built from, then weighted by area, nights at or above 25 °C rose from 69 a year to 75, while days at or above 40 °C fell from 32 to 30. Days above each area's own 90th percentile for that calendar day rose from 36 to 42, so the change in daytime heat depends on the measure used.

Figure 2 · Seasonal anomaly · interactive
Nights have warmed in every month but December and January, most in October
2011–2025
Mean night-time minimum and daytime maximum by month, 2011–25, minus the 1981–2010 mean for that month, °C, Pakistan, ERA5-Land. Gold shading is warmer than the base, green cooler. The window plays from 1981–95 to 2011–25 when the figure comes into view; drag the slider to choose another window. Source: ERA5-Land (Copernicus C3S) via Open-Meteo; Adaad calculation.

What the weather stations show

A daily maximum or minimum in the archive is either reported by the station or worked out by the archive from the station's hourly or synoptic observations, which can miss the hottest and coolest moments of the day. From 2013 to 2021, almost every Pakistani station-day in the archive was worked out in this way, against about two in five minima and one in four maxima in 1981–2010. At the same station and in the same month, worked-out minima were about 0.9 °C warmer, and worked-out maxima about 1 to 1.5 °C cooler, than reported ones, measured against the reanalysis on the same dates. A comparison that mixes the two kinds of value is therefore contaminated by the change in reporting, and an earlier version of this analysis, which did so, overstated the gap between nights and days. Excluding worked-out values also changes which years can be compared, so the size of the distortion cannot be measured exactly.

Using only reported maxima and minima, 19 stations have at least five qualifying summers in 1981–2010, mostly in 2003–10, and at least three in 2011–24, all in 2011–12 and 2022–24. Figure 3 compares each with ERA5-Land at the station's location on the same dates. Across the 19 stations, the median change was a rise of 0.36 °C in nights and a fall of 0.01 °C in days. The reanalysis on the same dates shows medians of a 0.18 °C rise and a 0.16 °C fall. The two records differ on the general level of change in these short windows, but the median gap between night and day changes is about a third of a degree in both.

Figure 3 · Stations and reanalysis
Nights rose more than days at 15 of 19 weather stations
Nights · May–September minimum −1 0 +1 +2 Jiwani (Balochistan), base 1981-2010, recent 2011-2024. Nights: station −1.45 °C, ERA5-Land +0.20 °C Jiwani Lahore (Punjab), base 2003-2010, recent 2011-2024. Nights: station −0.32 °C, ERA5-Land +0.18 °C Lahore Bahawalnagar (Punjab), base 2003-2010, recent 2011-2024. Nights: station +0.02 °C, ERA5-Land −0.04 °C Bahawalnagar Rohri (Sindh), base 2003-2010, recent 2011-2024. Nights: station +0.08 °C, ERA5-Land −0.16 °C Rohri Chhor (Sindh), base 2003-2010, recent 2011-2024. Nights: station +0.10 °C, ERA5-Land +0.11 °C Chhor Hyderabad (Sindh), base 2003-2010, recent 2011-2024. Nights: station +0.19 °C, ERA5-Land +0.18 °C Hyderabad Khanpur (Punjab), base 2003-2010, recent 2011-2024. Nights: station +0.20 °C, ERA5-Land −0.23 °C Khanpur Jhelum (Punjab), base 2006-2010, recent 2011-2024. Nights: station +0.24 °C, ERA5-Land +0.25 °C Jhelum Padidan (Sindh), base 2003-2010, recent 2011-2024. Nights: station +0.28 °C, ERA5-Land −0.17 °C Padidan Multan (Punjab), base 2006-2010, recent 2011-2024. Nights: station +0.36 °C, ERA5-Land −0.08 °C Multan Kakul (Khyber Pakhtunkhwa), base 2003-2010, recent 2011-2024. Nights: station +0.47 °C, ERA5-Land +0.62 °C Kakul Panjgur (Balochistan), base 2003-2010, recent 2011-2024. Nights: station +0.54 °C, ERA5-Land +0.31 °C Panjgur Balakot (Khyber Pakhtunkhwa), base 2003-2010, recent 2011-2024. Nights: station +0.64 °C, ERA5-Land +0.56 °C Balakot Karachi (Sindh), base 2003-2010, recent 2022-2024. Nights: station +0.66 °C, ERA5-Land +0.36 °C Karachi Nokkundi (Balochistan), base 2006-2010, recent 2011-2024. Nights: station +0.77 °C, ERA5-Land +0.39 °C Nokkundi Drosh (Khyber Pakhtunkhwa), base 2006-2010, recent 2011-2024. Nights: station +0.90 °C, ERA5-Land −0.31 °C Drosh D.I. Khan (Khyber Pakhtunkhwa), base 2006-2010, recent 2011-2024. Nights: station +1.20 °C, ERA5-Land +0.63 °C D.I. Khan Nawabshah (Sindh), base 2003-2010, recent 2012-2024. Nights: station +1.81 °C, ERA5-Land +0.19 °C Nawabshah Zhob (Balochistan), base 2003-2010, recent 2011-2024. Nights: station +2.34 °C, ERA5-Land −0.18 °C Zhob Days · May–September maximum −1 0 +1 +2 Jiwani (Balochistan), base 1981-2010, recent 2011-2024. Days: station +0.06 °C, ERA5-Land −0.16 °C Jiwani Lahore (Punjab), base 2003-2010, recent 2011-2024. Days: station +0.19 °C, ERA5-Land +0.01 °C Lahore Bahawalnagar (Punjab), base 2003-2010, recent 2011-2024. Days: station −0.07 °C, ERA5-Land −0.34 °C Bahawalnagar Rohri (Sindh), base 2003-2010, recent 2011-2024. Days: station −0.24 °C, ERA5-Land −0.73 °C Rohri Chhor (Sindh), base 2003-2010, recent 2011-2024. Days: station −0.41 °C, ERA5-Land −0.33 °C Chhor Hyderabad (Sindh), base 2003-2010, recent 2011-2024. Days: station −0.21 °C, ERA5-Land −0.15 °C Hyderabad Khanpur (Punjab), base 2003-2010, recent 2011-2024. Days: station +0.05 °C, ERA5-Land −0.78 °C Khanpur Jhelum (Punjab), base 2006-2010, recent 2011-2024. Days: station +0.65 °C, ERA5-Land +0.18 °C Jhelum Padidan (Sindh), base 2003-2010, recent 2011-2024. Days: station −1.00 °C, ERA5-Land −0.94 °C Padidan Multan (Punjab), base 2006-2010, recent 2011-2024. Days: station −0.31 °C, ERA5-Land −0.60 °C Multan Kakul (Khyber Pakhtunkhwa), base 2003-2010, recent 2011-2024. Days: station +0.17 °C, ERA5-Land +0.26 °C Kakul Panjgur (Balochistan), base 2003-2010, recent 2011-2024. Days: station −0.01 °C, ERA5-Land −0.08 °C Panjgur Balakot (Khyber Pakhtunkhwa), base 2003-2010, recent 2011-2024. Days: station +0.91 °C, ERA5-Land +0.41 °C Balakot Karachi (Sindh), base 2003-2010, recent 2022-2024. Days: station +0.32 °C, ERA5-Land +0.04 °C Karachi Nokkundi (Balochistan), base 2006-2010, recent 2011-2024. Days: station +0.66 °C, ERA5-Land +0.47 °C Nokkundi Drosh (Khyber Pakhtunkhwa), base 2006-2010, recent 2011-2024. Days: station −0.01 °C, ERA5-Land −0.56 °C Drosh D.I. Khan (Khyber Pakhtunkhwa), base 2006-2010, recent 2011-2024. Days: station +0.59 °C, ERA5-Land +0.52 °C D.I. Khan Nawabshah (Sindh), base 2003-2010, recent 2012-2024. Days: station −0.51 °C, ERA5-Land −0.22 °C Nawabshah Zhob (Balochistan), base 2003-2010, recent 2011-2024. Days: station −0.06 °C, ERA5-Land −0.52 °C Zhob
station ERA5-Land at the station, same dates °C, change between periods
Notes. Change in mean May–September night-time minimum and daytime maximum at 19 synoptic stations (filled) and in ERA5-Land at the station's location (open), using only days on which the station reported both values, and the same dates for both records. Base seasons are mostly 2003–10 (Jiwani's are 1981 and 2007–10) and recent seasons 2011–12 and 2022–24. Stations are ordered by their night-time change. Kakul is in Abbottabad, Balakot in Mansehra, Khanpur in Rahim Yar Khan, Rohri in Sukkur, Padidan in Naushahro Feroze, Chhor in Umerkot and Jiwani in Gwadar. Source: NOAA NCEI Global Summary of the Day; ERA5-Land (Copernicus C3S) via Open-Meteo; Adaad calculation.

These are short windows, and the stations have not been adjusted for moves or instrument changes, so the comparison supports the ordering of night and day changes in this sample rather than a precise national trend. Under every coverage rule tried, nights rose more than days at most of the stations retained, although the number of stations changes from 5 to 28 (see method note).

The records largely agree on where days have warmed or cooled. Across the 19 stations, the station and reanalysis changes in daytime maxima are closely related, with a correlation of 0.8. They do not agree on where nights have warmed. Station night-time changes range from a fall of 1.5 °C at Jiwani to a rise of 2.3 °C at Zhob, unrelated to the reanalysis's changes at the same places. Night-time temperatures at a single site respond to local land cover, building, irrigation and siting, as well as to regional climate, and these comparisons cannot separate those influences. The reanalysis's map of night-time warming is not validated by this comparison and should not be read as a ranking of where Pakistan's nights have warmed most.

For May to September, the national and provincial series behind Figures 1, 2 and 4 can be checked the same way, against the stations inside each area. Year to year, they track those stations closely for days, with correlations of 0.8 to 0.9, and less closely for nights: 0.6 nationally and from 0.3 in Balochistan to 0.8 in Sindh. Over the comparison years they also show less night-time warming than the stations do, 0.2 °C against 0.5 °C on average across Pakistan, with a much smaller difference in Punjab. The difference could reflect local conditions around the stations, changes in their records, sampling, or limitations of the reanalysis, and this comparison cannot distinguish their contributions. No station tests the Gilgit-Baltistan series.

Some long station records also show unexplained shifts against the reanalysis. At Peshawar, the station's summer maxima fall by about 2 °C relative to the reanalysis around 2001–02. At Chaklala, Rawalpindi's airport, a fall of about 1 °C comes around 2020. Either record could have changed. We have not obtained the station histories needed to investigate these shifts.

The north: a signal not yet tested against stations

In the reanalysis, Gilgit-Baltistan has the largest seasonal warming of any province. September and October nights were about 1.4 °C warmer over 2011–25 than over 1981–2010, and days about 1.2 °C warmer (Figure 4, choosing Gilgit-Baltistan). Both months remain warmer when the base period starts in 1991 and when 2024 and 2025 are left out. Winter is mixed: December and January nights were cooler, by 1.0 and 0.5 °C, and February nights 0.3 °C warmer.

These checks show that the signal persists within the reanalysis, not that observations confirm it. The NOAA summaries assembled here cannot provide a comparable long-term autumn test. Published studies of the Pakistan Meteorological Department's station records, one running to 2012 and another to 2013, cover different periods and give mixed results. A cooling trend within an earlier interval, such as 1995–2012, can coexist with warming between our two period averages. The northern result therefore remains unvalidated.

Figure 4 · Provinces · interactive
In the reanalysis, October nights have warmed most in every province but Sindh
2011–2025
As Figure 2, 2011–25, ERA5-Land. The figure opens on Pakistan; choose Gilgit-Baltistan or another province to compare. The method note sets out how the national and provincial series compare with weather stations. This is a reanalysis result. The NOAA station summaries assembled here are too sparse for a comparable autumn test, and published station studies cover different periods with mixed results. Source: ERA5-Land (Copernicus C3S) via Open-Meteo; Adaad calculation.

If the pattern holds, it matters for the glaciers. Many Karakoram glaciers, unlike most in the Hindu Kush and the Himalaya, were close to balance or gaining mass in the early 21st century, a pattern known as the Karakoram anomaly, for reasons glaciologists still debate. Air temperatures averaged over large areas cannot show whether that is changing. The test is the measured change in glacier mass, basin by basin, which satellite elevation surveys now provide from 2000.

What a warmer north could mean downstream

None of what follows is measured here. The Indus above Tarbela draws much of its flow from snow and glacier melt, and in 2005 the World Bank estimated that Pakistan's reservoirs could hold only about 30 days of its average flow. Shrinking glaciers tend to release more water for a time before releasing less, a turning point hydrologists call peak water, and a widely cited 2014 set of projections put upper Indus runoff mostly higher in the 2040s. Changes in that flow would reach cotton planting, wheat's reliance on water carried over from summer, and hydropower on the Indus. Warmer mountains also bear on hazards: in a December 2022 project brief, UNDP counted 33 glacial lakes in Gilgit-Baltistan and Khyber Pakhtunkhwa prone to outburst floods, and in May 2022 an outburst from a lake dammed by the Shishper glacier destroyed the Karakoram Highway bridge at Hassanabad. That flood is an example of the hazard, not a consequence of the anomalies reported here.

What the numbers do and do not support

The reanalysis and the station records agree that Pakistan's summer nights have warmed more than its summer days. In the station sample, nights rose more than days at most stations, with a median gap of about a third of a degree in both records over the short periods the stations can test. The reanalysis puts the full change since 1981–2010 at about 0.4 °C for nights and almost nothing for days. The station comparison does not validate the reanalysis map of night-time warming. The northern autumn signal remains a reanalysis finding. Existing station studies cover different periods and give mixed results, and they do not establish whether its recent magnitude or geographic extent is correct.

The analysis does not evaluate heat risk, which depends on humidity, exposure and health as well as temperature. It does suggest that locally calibrated heat warnings should consider night-time minima alongside daytime maxima. And it shows what is needed next. For the north, the first need is monthly averages of daily minimum and maximum temperature since 1981 at Gilgit, Skardu, Astore, Bunji, Chilas and Gupis, with valid-day counts, flags for estimated values and station histories, plus records from high-altitude stations to test whether valley results extend to snow and glacier terrain. Beyond that, glacier mass change by basin would show whether the glaciers are responding, and a consolidated daily archive of Tarbela's inflow, outflow and level would show whether the river's timing has shifted.

Method note

Reanalysis. ERA5-Land hourly 2 m temperature via the Open-Meteo open-data mirror, reduced to daily maximum, minimum and mean on the Asia/Karachi day for every 0.1° cell and averaged over the cells in each of the 160 administrative areas of the Census 2023 frame. Anomalies are monthly means minus the 1981–2010 mean for that calendar month. Provincial and national series are area-weighted means of those areas' anomalies, and seasonal means weight months by days. ERA5-Land has no urban land cover. National threshold counts (nights at or above 25 °C, days at or above 40 °C, days above the ETCCDI TX90p calendar-day 90th percentile with a five-day window) are counted first for each area, from that area's average daily temperature, and then averaged across the country weighted by area. They are not counts of days on which a national average crossed the threshold, nor of grid cells or people exposed. The 25 °C and 40 °C thresholds are chosen for this analysis.

Stations. Every Pakistani station in NOAA's station history file (103 records) was downloaded from the Global Summary of the Day for 1981–2025, May to October. The archive's 2025 files end on 24 August, so the last complete summer is 2024. Values are converted from Fahrenheit. A maximum or minimum flagged in the archive as derived from hourly or synoptic observations is excluded, so only days with both values reported by the station are used. Unflagged values are not fully standardised: NOAA notes that reported extremes may include part of the previous day, and the stations were not homogenised for moves or instrument changes. A month counts if it has at least 15 such days, and a summer if all five months count. Its value is the day-weighted mean of the monthly means. Nineteen stations have at least five qualifying summers in 1981–2010 and three in 2011–24. The base seasons are 2003 and 2006–10, except Jiwani's, which are 1981 and 2007–10, and the recent seasons are 2011–12 and 2022–24. Each station is compared with ERA5-Land on the same dates, both at the station's coordinates (Open-Meteo archive API) and averaged over the administrative area that contains it.

Reporting change. In 1981–2010 about 44 per cent of minima and 27 per cent of maxima in the archive were derived; in 2013–21, 88 to 97 per cent. Within the same station, year and month (cells with at least three days of each kind), derived minima were a median 0.83 °C warmer, and derived maxima 1.44 °C cooler, than reported ones, relative to ERA5-Land. For the same 19 stations, the median gap between night and day changes is 1.02 °C using all days and all years, 0.45 °C using all days in the station-years kept by the main analysis, and 0.33 °C using reported values only. The matched reanalysis gap is 0.26, 0.35 and 0.34 °C. The first step changes both the reporting mix and the years compared, so it does not isolate the reporting effect.

Sensitivity. The main rule (at least 15 reported days a month, five base and three recent summers) keeps 19 stations, 15 with nights rising more than days, with median changes of +0.36 °C for nights and −0.01 °C for days. Requiring five recent summers keeps 15 stations (12; +0.28 and −0.01 °C). Requiring six base and four recent summers keeps 6 (5; +0.19 and −0.04 °C). Requiring 10 days a month keeps 28 (21; +0.38 and +0.02 °C), and requiring 20 days a month keeps 5 (5; +0.44 and −0.02 °C).

The station composition changes between rules, and the cross-station correlations are unstable in the smaller samples. Leaving out one station at a time gives median gaps of 0.33 to 0.37 °C for the stations and 0.33 to 0.35 °C for the reanalysis. Using every May–September date in the retained station-years, rather than only the reported dates, gives a reanalysis gap of 0.35 °C. Defining the reanalysis day on Greenwich time instead of Pakistan time changes point results by 0.07 °C or less at the five stations tested. ERA5-Land at the station's location and over its surrounding area give median changes within 0.04 °C of each other.

Regional series against stations. Each station's qualifying May–September seasons were compared with the area-weighted ERA5-Land series for its province and for Pakistan, using every date in the same years. Year to year, the correlation between the regional series and the average of its stations' anomalies is 0.81 for days and 0.60 for nights across Pakistan (19 stations, 12 years). By province, for days and nights, it is 0.91 and 0.83 in Sindh (six stations), 0.87 and 0.73 in Punjab (five), 0.88 and 0.68 in Khyber Pakhtunkhwa (four, three of them hill stations), and 0.83 and 0.33 in Balochistan (four, including coastal Jiwani). Averaged across stations, the change between their base and recent seasons is +0.47 °C for nights and +0.04 °C for days at the stations, against +0.19 and −0.10 °C in the national series on the same years. By province the night-time changes are +0.52 against +0.08 °C in Sindh, +0.80 against +0.40 °C in Khyber Pakhtunkhwa, +0.55 against +0.16 °C in Balochistan and +0.10 against +0.06 °C in Punjab. The gap between night and day changes is positive in both records nationally (0.44 °C at stations, 0.29 °C in the series) and in every province except Punjab, where the stations show none. These figures are means, so they differ from the medians given for Figure 3. The caveats are large. Four to six stations cannot represent a province, and these are mostly lowland airports. The comparison windows are short, mostly 2003–10 against 2011–12 and 2022–24, and correlations over 11 or 12 years are imprecise. No station in the main sample falls in Gilgit-Baltistan, Azad Kashmir or Islamabad. The comparison describes agreement with these stations, not the accuracy of either record.

Peshawar and Chaklala. A single-break scan of the station-minus-reanalysis series for May–September maxima puts the largest shift at 2002 for Peshawar (−1.9 °C) and 2021 for Chaklala (−1.5 °C).

Gilgit-Baltistan. In the Global Summary of the Day, the September–October records for Gilgit, Skardu, Astore, Bunji, Chillas and Gupis begin in 2001 at the earliest and, from 2005, are mostly confined to 1–15 October, with a few September months (2005 at Gilgit and Chillas; 2003 and 2005 at Skardu; Gupis in 2021–24). This pattern is in NOAA's source files. In early October, 82 to 100 per cent of extrema are derived. An exploratory comparison of 1–15 October in 2005–10 (Skardu from 2001) with 2011–21 shows station maxima falling by 1.6 to 3.5 °C and minima changing by −0.3 to +1.5 °C, while on the same dates reanalysis maxima rise by 0.3 to 1.2 °C and minima by 0.1 to 2.0 °C. With short windows and mostly derived values, it is not used as evidence either way. NOAA's successor product, the Synoptic Summary of the Day, holds no Gilgit or Skardu record, and its Astore and Bunji records begin in 2004–05 with the same early-October pattern. The Meteorological Department's regional centre in Gilgit publishes the latest day's minimum and maximum for nine stations (Gilgit, Skardu, Astore, Bunji, Chilas, Gupis, Hunza, Babusar and Bagrote) but no historical series. Reanalysis figures for the province are area-weighted means over the province. As robustness checks, September minima and maxima change by +1.26 and +1.05 °C with a 1991–2010 base and by +1.07 and +0.82 °C with a recent period ending in 2023. The corresponding October changes are +1.26 and +1.04 °C, and +1.35 and +1.19 °C. These are not confidence intervals. Contrasts between preselected period means are diagnostics, not detected steps or an additive decomposition: October maxima and minima differ by +1.74 and +1.40 °C between 1981–94 and 1995–2012, and September maxima and minima by +1.17 and +1.24 °C between 1995–2012 and 2013–25. Trends within 1995–2012 depend on the estimator. For provincial October maxima, ordinary least squares gives +0.38 °C a decade and a raw Sen's slope −0.06 °C a decade, so the sign of a trend over that period is not robust. Hasson and others (2017) used Sen's slope with an autocorrelation correction, which was not reproduced here. Ullah and others (2024) use six Gilgit-Baltistan stations over 1991–2021 but fill missing temperatures with ERA5, so their completed series cannot independently test it.

Tarbela. Whether the Indus's timing has already shifted cannot be read from the records assembled so far. Satellite measurements of Tarbela's surface area (Global Water Watch, Deltares/WRI) are too sparse before 2016, with only 2016, 2020 and 2023–25 observed within ten days of the seasonal rise past 90 per cent of maximum area, and we have not assembled a sufficiently complete historical daily series of the reservoir's inflow, outflow and level.

Files. The station comparison behind Figure 3, the Gilgit-Baltistan monthly contrasts, the regional comparison with stations and the numbers behind Figures 1, 2 and 4 are downloadable below.

Sources

Arab News (2022). Glacial outburst destroys strategic bridge connecting Pakistan with China. 8 May 2022.

Farinotti, D. and others (2020). Manifestations and mechanisms of the Karakoram glacier anomaly. Nature Geoscience 13, 8–16. doi:10.1038/s41561-019-0513-5

Hasson, S., Böhner, J. and Lucarini, V. (2017). Prevailing climatic trends and runoff response from Hindukush–Karakoram–Himalaya, upper Indus Basin. Earth System Dynamics 8, 337–355. doi:10.5194/esd-8-337-2017

Hugonnet, R. and others (2021). Accelerated global glacier mass loss in the early twenty-first century. Nature 592, 726–731. doi:10.1038/s41586-021-03436-z

Huss, M. and Hock, R. (2018). Global-scale hydrological response to future glacier mass loss. Nature Climate Change 8, 135–140. doi:10.1038/s41558-017-0049-x

Latif, Y. and others (2020). Spatial analysis of temperature time series over the Upper Indus Basin (UIB) Pakistan. Theoretical and Applied Climatology 139, 741–758. doi:10.1007/s00704-019-02993-8

Lovell, H. and Muhammad, S. (2025). Multiple phases of ice-dammed lake formation and drainage associated with a surge of Shisper Glacier, western Karakoram. Journal of Glaciology 71, e10. doi:10.1017/jog.2024.80

Lutz, A. F. and others (2014). Consistent increase in High Asia's runoff due to increasing glacier melt and precipitation. Nature Climate Change 4, 587–592. doi:10.1038/nclimate2237

Muñoz-Sabater, J. and others (2021). ERA5-Land: a state-of-the-art global reanalysis dataset for land applications. Earth System Science Data 13, 4349–4383. doi:10.5194/essd-13-4349-2021

NOAA National Centers for Environmental Information. Global Summary of the Day and Synoptic Summary of the Day.

Ullah, S. and others (2024). Insights into climate change dynamics: a tourism climate index-based evaluation of Gilgit-Baltistan, Pakistan. Heliyon 10(15), e35315. doi:10.1016/j.heliyon.2024.e35315

UNDP (2022). Scaling-up of GLOF risk reduction in northern Pakistan (GLOF-II), project brief. December 2022.

World Bank (2005). Pakistan Country Water Resources Assistance Strategy. Water Economy: Running Dry. Report 34081-PK, 22 November 2005, p. xii (Figure S10) and pp. 60–61.

Data and methods downloads