The Bangladesh Power Development Board publish daily totals of electricity demand for each of the country's nine power zones (Fig. a). These are archived at URLs of the form https://misc.bpdb.gov.bd/area-wise-demand?date=DD-MM-YYYY (last access: 30 April 2026), from which they can be readily scraped. This gives us daily data, starting in December 2015 and running through to the present. Some dates are missing from the archive, and some zone entries appear as zeroes, which we treat as missing data (rather than true demand). For national totals and all subsequent composite analyses, we conservatively exclude any day where at least one of the nine zones is missing (i.e., reporting zero demand), so that the national total is always based on complete zone coverage. This leaves us with approximately 99.7 % coverage.

To identify and analyse dips in demand met, we use a centred 60 d running mean (requiring at least 30 valid days to compute the baseline). We then express anomalies as fractional deviations from this mean and define “dip days” as those where the anomaly is less than 80 % of the running mean.

We have made the scraped data available in machine-friendly format at 10.5281/zenodo.17175212 (). The code to download the data and recreate all figures and analysis in this paper is available at https://github.com/kieranmrhunt/bangladesh-renewable (last access: 30 April 2026). This is part of a wider research group initiative to make energy data open and accessible to all, e.g., for training large-sample demand models.

Daily demand met data for West Bengal are taken from the Indian dataset published by . They extracted the data from daily reports issued by the Grid Controller of India (Grid-India), known before November 2022 as the Power System Operation Corporation (POSOCO). The reports are in PDF format and contain the total energy supplied by the national grid to each state on each day. The final version of the dataset covers all Indian states from 2014 to present, and is available at 10.5281/zenodo.14983362 .

ERA5 is the fifth generation atmospheric reanalysis of global climate produced by the Copernicus Climate Change Service at the European Centre for Medium-Range Weather Forecasts . Data from ERA5 have global coverage at a horizontal resolution of ∼ 25 km and resolve the atmosphere on 137 levels from the ground up to ∼ 80 km in altitude. It covers from January 1940 until present at hourly frequency. We use single-level data (10.24381/cds.adbb2d47, ) to show the near-surface instantaneous 10 m vector wind fields associated with blackout case studies, as well as “significant height of combined wind wave and swell” as a proxy for the magnitude of storm surges. There are global biases in ERA5's under-representation of the magnitude and intensity of surface wind gusts, particularly over complex terrain , but here we use it qualitatively to contrast the relative intensities and structures of the cyclones.

Global Precipitation Mission Integrated Multi-satellitE Retrievals for (GPM-IMERG) is a global satellite-based precipitation product at a half-hourly, 0.1° resolution, starting in June 2000 and continuing to the present day. Over the tropics, IMERG primarily ingests retrievals from (for 2000–2014) the now-defunct Tropical Rainfall Measuring Mission 13.8 GHz precipitation radar and microwave imager and (for 2014 onwards) the Global Precipitation Measurement Ka/Ku-band dual-frequency precipitation radar. We use IMERG precipitation data (0.1°/30 min) to quantify rainfall in our case studies because its finer resolution than ERA5 (0.25°/hourly) and demonstrated skill in reproducing the spatial and diurnal structure of precipitation over Bangladesh make it preferable for resolving small-scale extremes; more broadly, many evaluations report IMERG outperforming reanalyses such as ERA5 in convective regimes e.g.,.

The Emergency Events Database e.g., maintained by the Centre for Research on the Epidemiology of Disasters provides a homogenised catalogue of natural hazards and their reported impacts. We queried EM-DAT for the period December 2015–May 2025 and retained all events whose affected country list contains Bangladesh. We then subset to hydrometeorological classes potentially relevant to the power system: lightning, floods, coldwaves, heatwaves, and wildfires. For tropical cyclones and depressions, we additionally verified the reported event dates against the India Meteorological Department (IMD) Regional Specialized Meteorological Centre metadata (https://rsmcnewdelhi.imd.gov.in/report.php?internal_menu=MzM=, last access: 30 April 2026) to ensure that the disturbance made landfall over, or had a clearly documented direct impact on, Bangladesh. The EM-DAT catalogue is used in two ways: (i) to tag and interpret sharp demand shortfalls that are not associated with a tropical cyclone or depression (Fig. ); and (ii) to ensure that our composites of “other” hazards do not mix qualitatively different drivers (e.g., major river flooding versus pre-monsoon mesoscale convective systems).

Building on these composite statistics, we summarise key impacts metrics and the maximum national and zonal electricity shortfalls for each landfalling tropical cyclone in Table .

Across the 14 landfalling cyclones in Table , the magnitude of the national electricity deficit (“national dip”) is strongly related to the severity of the worst-affected zone (“zone dip”). Within ± 2 d of landfall, the maximum national dip is highly correlated with the corresponding maximum zonal dip (r=0.80, p<0.001). This suggests that national-scale shortfalls are not distributed evenly, but are driven by near-total collapse in the most exposed zone, preventing load balancing.

Correlations with human impacts are weaker, as these metrics are often missing from EM-DAT (10 of 14 cyclones have reported deaths; 6 of 14 have reported damage), but available reported deaths correlate more strongly with the worst-zone deficit (r = -0.80, p=0.006, n=10) than with the national deficit (r = -0.56, p=0.09, n=10), suggesting that mortality is more closely tied to localised, extreme disruption than to the national average. This provides evidence that mortality is driven by local infrastructure collapse and inability to access services, rather than broader national energy shortages.

Our storm-surge proxy is moderately correlated with cyclone intensity, increasing for lower minimum central pressures (r = -0.55, p=0.04), but the apparent correlation between reported economic damage and surge is insignificant due to small sample size (r=0.64, p=0.17, n=6). Having established the national-scale picture, we now examine how these impacts vary regionally.

We now extract 4 cases from among these large dip events. These allow us to better examine, compare, and contrast the regional structure of their causes and impacts. These cases comprise two cyclones (Remal and Sitrang), which led to the largest relative reductions in demand met; the deep depression with the largest reduction in demand met (the landfalling deep depression of October 2017); and a major blackout in March 2019 that did not coincide with any significant synoptic-scale circulation, but is logged in EM-DAT as a day with an intense thunderstorm.

We start by investigating the change in regional impacts on demand met across Bangladesh's nine power zones (Fig. ). For Cyclone Remal, we are missing data from day -2. However, on day -1, we can already see a significant impact in coastal zones, with Barisal reporting a fall of 72 % in demand met and Khulna reporting a fall of 35 %. On the day of landfall itself (28 May), all zones reported a large deficit. Even the least affected zone – Chittagong – still had demand met below its rolling 60 d mean. After the dissipation of Remal, most zones recovered quickly, and within 2 d were returned to normal power. Barisal was the exception here, having been heavily damaged by Remal, even 3 d later (31 May), the demand met was still 17 % less than the rolling mean.

The profile for Sitrang is quite different. This time there is no noticeable impact on day -1. As in Remal, on the landfall day itself (24 October) all zones reported a large deficit (from 40 % in Rajshahi to 91 % in Barisal). However, the recovery was considerably slower than for Remal, such that even 4 d later (28 October), all zones reported a deficit, with all but one reporting a deficit of more than 15 %. Three of the coastal zones (Barisal, Khulna, Comilla) were disproportionately affected.

The impact of the worst deep depression (Fig. c; landfall on 21 October 2017) was considerably less than either of these tropical cyclones, and the recovery was thus quicker, with all zones returning to normal just 2 d after landfall. As we will see in the next section, depressions often produce heavy rain much further inland than tropical cyclones. This leads to the inland zones being more affected on day 0, and the coastal zones being more affected the day before. Despite the relatively weak synoptic signature of such depressions, they are still capable of significant disruption – Mymensingh and Sylhet reported deficits of 72 % and 67 %, respectively on the day of landfall.

The major blackout (Fig. d; 31 March 2019) differs from the cyclones and depression in that on day 0, there was no significant regional pattern, with neither coastal nor inland states were more affected than the other. Recovery was almost immediate (into day +1), but then a second event hit the grid (which did not coincide with any event in the EM-DAT catalogue) on day +2, again affecting a mix of coastal and inland states. This lack of a clear regional pattern suggests a non-meteorological cause, supported by the absence of strong synoptic forcing, as we show in the next section.

We now present the same 4 cases as synoptic weather charts on the day of maximum dip (Fig. ). Synoptic-scale storms damage electricity infrastructure through three principal hazard pathways: (i) heavy precipitation and the ensuing pluvial and fluvial flooding; (ii) storm surge and associated coastal inundation; and (iii) destructive winds. In practice these hazards co-occur, making it difficult to cleanly attribute observed power-system impacts to any single component. Bangladesh's small geographic extent relative to the typical precipitation footprint of Bay of Bengal systems further complicates attribution since most events affect the majority of the country more or less simultaneously.

Nevertheless, the 3 case studies in Fig.  do allow a degree of contrast. Cyclone Remal (Fig. a) produced comparatively modest rainfall totals over Bangladesh – the heaviest precipitation fell over north-eastern India – but generated a large storm surge, with reports of water levels exceeding three metres along parts of the coast . This is consistent with ERA5 significant wave heights exceeding 6 m adjacent to Bangladesh's coastline (Fig. e), and with widespread, surge- and wind-driven damage to coastal substations and distribution infrastructure. Cyclone Sitrang (Fig. b) had weaker winds and surge, yet produced substantially greater rainfall over Bangladesh than Remal. The deep depression (Fig. c) was weaker still in terms of winds and surge, but produced even heavier rain totals across the country. This behaviour is characteristic of monsoon depressions over the Indian subcontinent, which often propagate far inland under favourable environmental conditions and are a dominant source of extreme precipitation for much of India . Finally, synoptic charts for the March 2019 blackout (Fig. d) confirm that this was unlikely to have been caused by weather, despite thunderstorm reports on the day.

Together, these events suggest a general pattern. As precipitation increases and wind and surge severity decreases, the aggregate impact on the power system tends to be slightly smaller (but still very large in absolute terms). All three pathways can produce substantial, system-wide losses on their own, and they frequently arrive as a compound risk. Therefore, any credible risk model for Bangladesh's national critical infrastructure must treat these hazards jointly. These regional disparities raise critical questions about Bangladesh's import strategy, which we will now discuss.

The new high-voltage exchange between Bangladesh (BD) and West Bengal (WB) is intended, in part, to provide resilience when production or transmission assets in Bangladesh are damaged. Therefore, we now ask: how often are Bangladesh's deepest supply shortfalls not mirrored across the border, such that imports could possibly help?

As neighbouring regions, Bangladesh and West Bengal experience very similar weather, and consequently very similar seasonal demand profiles (Fig. a). This is bad for seasonal-scale complementarity: when one system's demand is high (or low), so is the other's, limiting arbitrage value. Extreme, short-lived events are more promising. A closer inspection of the joint distribution of daily demand met (Fig. b) shows that large single- or few-day dips in one system are often not present in the other.

There are really only two clear episodes with very low West Bengal demand that are not matched by similarly low Bangladesh demand. The first is Cyclone Amphan (20–22 May 2020), where landfall over West Bengal led to much larger West Bengal impacts than in Bangladesh. The second is Cyclone Yaas (26–27 May 2021), where landfall over Odisha, close to West Bengal, produced a stronger West Bengal response; the system subsequently weakened before significantly affecting Bangladesh.

In contrast, there are many cases where Bangladesh demand met is markedly reduced while West Bengal is comparatively unaffected. Firstly, pre-monsoon thunderstorm activity in Bangladesh (16 June 2018, 19 June 2018, 6 May 2024, 23 May 2023), with or without an accompanying synoptic low, depressed Bangladesh demand met disproportionately. Secondly, thunderstorms over Bangladesh associated with the monsoon onset (15–20 June 2024) – during which West Bengal demand remained high, consistent with pre-monsoon heat stress there. Thirdly, there are cases with no obvious synoptic-scale driver (16–17 April 2024, 21 April 2023, 27 April 2023), where Bangladesh demand stayed high in absolute terms, suggesting temperature differences or local operational issues rather than damage.

There are also more extreme Bangladesh-only cases where weather clearly dominated, namely Cyclone Maarutha (19 April 2017), Cyclone Biparjoy (16 June 2023), and Cyclone Remal (28–29 May 2024); and several that were likely not weather-related (30 March 2018 and 31 March 2019).

Days on which both regions reported very low demand met occur during seasonal troughs, possibly linked to public holidays, as 25 December 2015, 1 January 2016, and 16 December 2019 all occurred during the climatologically coolest period of the year and are not associated with strong synoptic forcing. The notable weather-driven exception was on 10 November 2019 (Cyclone Bulbul), when both Bangladesh and West Bengal were simultaneously stressed.

We now quantify correlated stress across the border. As before, we define “dip” days as those when demand met falls below 80 % of a centred 60 d running mean. From 2015–2025 (a total of 3393 d), Bangladesh experienced 82 dip days and West Bengal experienced 46. Only 9 d (0.27 % of all days; 11 % of Bangladesh dips) are synchronised (occurring on the same day) and 16 out of 82 (19.5 %) Bangladesh dips coincided with a West Bengal dip within ± 1 d. During synchronised dips, the mean deficits are large in both systems (Bangladesh averaging 3.1 GW; West Bengal averaging 1.9 GW). While synchronised dips are rare (0.27 % of all days), they are strongly associated with major cyclones, with 5 out of 9 occurring within ± 1 d of a Bangladesh landfalling cyclone. In other words, synchronicity is conditionally high during exactly those events when cross-border import is most needed. Conditional on a Bangladesh deep dip occurring within ± 1 d of cyclone landfall, the probability of a same-day West Bengal deep dip rises to ∼ 45 %, compared with ∼ 6 % for Bangladesh deep dips not occurring with a day of cyclone landfall. No synchronised dips coincided with depression landfalls.