# The Danube at Low Water: how one river constrains power, freight and imports

> Low river flow can constrain hydropower, cooling, navigation, imports and peak-demand balance at the same time. The nuclear emergency is one visible point in a wider water–energy system.

_Research cutoff: 2026-08-14T13:15:00+03:00 · all values are labeled snapshots_

## A river at its all-time low. An entire network feels it.

On 6 August 2026 the Danube entered Romania at 1,400 m³/s, exactly the all-time low measured at Baziaș in 1985. The figure matched the record. Flow was under 30% of the July multiannual average (4,700 m³/s). One nuclear unit was offline at Cernavodă, Paks was five centimetres from its first-ever full shutdown, and barges carried 30% of their usual load. On 13 August, the Associated Press reported that U2 had also entered a controlled shutdown. The common factor: water.

In three weeks, flow measured at Baziaș, where the river enters Romania, fell from 1,750 m³/s on 17 July to 1,400 m³/s on 6 August. The INHGA bulletin of 14 August recorded 1,400 m³/s for the 13–14 August window and forecast 1,300 m³/s by 21 August. The 1985 record remains matched in the observations available.

The number matters because the Danube supplies cooling water to three nuclear sites, carries grain and ore, and feeds hydropower plants. These roles also affect the evening electrical balance of Romania and Hungary. When flow falls, several functions can become constrained at the same time.

> Deep dive: Every figure carries an evidence status, and the complete source ledger is public, including JSON. The public record still lacks Cernavodă's 2026 numeric intake limits and credential-gated ENTSO-E series. Those gaps remain visible.

## 2,850 kilometres: from source to sea, the same water

Low water appears along the river, from Germany to Romania. Pfelling 216 cm, Budapest at an all-time low of 23 cm, Ruse −94 cm, and Cernavodă −209 cm are each read against their local gauge zero and have a source.

The atlas pins each station to a verified river kilometre, following the Danube Commission convention with km 0 at Sulina and values increasing upstream. Hydrometry has a simple rule: centimetre levels are read against each gauge's local zero and cannot be compared directly across stations. The values −209 cm at Cernavodă and 216 cm at Pfelling indicate unusually low water only in their local context. Discharge is a separate measurement, expressed in m³/s.

The atlas covers the course from Donaueschingen to Sulina. Measurements appear only where verifiable observations were available. Regensburg is the first measured station in the dataset. On 9 August, WSV classified Regensburg, Pfelling, Hofkirchen and Passau as low. The Inn influences Passau, which was in the normal class on 7 August.

> Deep dive: The six German and Austrian stations come from the public PEGELONLINE API, re-retrieved on 09.08.2026 at 11:15–11:30 CEST. Hydroinfo listed Paks at −121 cm on the morning of 14 August. The Romanian gauges at Giurgiu, Călărași, Brăila, and Tulcea remain a historical 6 August snapshot and are labeled accordingly. Every observation can be traced to its URL and retrieval time.

## 1,750 → 1,400 m³/s in twenty days

Baziaș flow: 1,750 m³/s on 17 July, 1,400 on 6 August. July multiannual average: 4,700. August average: 3,900. The river ran at 29.8% of the July average and 35.9% of the August average.

INHGA's daily series shows the descent: 1,750 (17.07) → 1,600 (29–31.07) → 1,500 (03.08) → 1,450 (04–05.08) → 1,400 (06–08.08). The 14.08 bulletin records 1,400 m³/s again for the 13–14 August window and forecasts 1,300 m³/s by 21 August. Cernavodă estimates remain indicative during the Bala works. Water temperature was 27 °C at Cernavodă on 27.07 and 27.9 °C at Paks on 05.08. For cooling, temperature is read alongside discharge and level.

Precipitation and neighbouring-river data point to a regional deficit: Hungary reported 80–130 mm less rain over 90 days (OMSZ), while the Tisza reached −214 cm at Szeged, a record low. These observations place the Danube's fall in a wider context than one reach.

> Deep dive: On 31.07, one outlet headlined “1,400 m³/s”, a week before INHGA officially measured that value on 06.08. At headline time, the primary series read 1,600. The conflict is recorded in the claim register (C20), and the product uses the primary series. On 06.08, the value became a documented fact. The rule is simple: use the primary series, date the observation and show the evidence status.

## The same river, three cooling problems

Cernavodă shut one unit because its cooling basin is fed from a branch that receives only ~15% of the flow at low water. Paks came within five centimetres of total shutdown because the level fell toward its pump suction nozzles. The verified Kozloduy source reported no power restriction. On 13 August, Nuclearelectrica officially announced in a statement addressed to the Bucharest Stock Exchange that Cernavodă U2 also entered a controlled shutdown; the Associated Press quoted the statement.

Cernavodă (2 × ~700 MWe, CANDU, km 300) draws ~90 m³/s for two units from a basin fed by the Danube–Black Sea Canal. At low water, ~85% of the current takes the Bala branch at the km 346–345 bifurcation and only ~15% reaches the plant branch. Unit 1 was shut down in a controlled manner on 28 July; Apele Române reported a ~10 cm level gain for Unit 2. On 09.08, Transelectrica showed 684 MW of nuclear output, consistent with one Cernavodă unit operating. The 2003 precedent describes the mechanism; current numeric thresholds are not public.

Paks (4 × 500 MWe, VVER) cools directly from the river, ~100 m³/s at full power. The operator describes a suction problem: water volume can be sufficient while the surface falls below the pump mouths. Overnight on 3–4 August, the river came within five centimetres of the threshold for the final turbine. Full shutdown did not occur. Hydroinfo listed −121 cm on the morning of 14 August and forecast −134 cm for 17 August; the latest official power figure included here remains the 09.08 OAH snapshot, with Unit 2 at 228 MW and the other units at 0 MW.

Kozloduy (2 × 1,000 MWe) takes its water with a pump station in a deep bay, with the intake below 17 m and designed for a 1% annual-probability event. Both units stayed at planned load, and Bulgaria kept exporting to Romania and Serbia. The design difference matters: intake placement determines part of a plant's ability to withstand low water.

> Deep dive: The thermal rule adds a second constraint at Paks: discharged water may not exceed 30 °C at 500 m downstream (KöM 15/2001), with an internal limit of 29.5 °C and an exceptional regime at 31.5 °C. On 18 July the first cut (−160 MW) came from temperature; the later low-water constraint came from level. An investigative outlet reported exceedances of the 30 °C rule, and the operator disputes that report. The claim is marked disputed (C21).

> Deep dive: Hydropower draws on the same limited flow: Iron Gates I (1,166.4 MW on the Romanian side, 27.17 m nominal head) produces ≈10% of national energy and provides ≈half of system services. At 1,400 m³/s, the same water serves turbines, the cooling canal and navigation.

## Three weeks, event by event

The timeline follows the first Paks cut (18 July), the Cernavodă U1 shutdown (28 July), the Viking Ullur grounding (28 July), the critical night at Paks on 4 August, the controlled Pârjoaia blast (3 August), the matched Baziaș record (6 August), and the 13–14 August update. Every entry has a source.

The timeline below is built exclusively from dated sources: INHGA bulletins, Nuclearelectrica statements (via Agerpres), MVM/Paks, OVF, AFDJ and named national press. Where only the press reported an event, the entry carries “reported” status.

> Deep dive: Social-media rumours about “water held by the Austrians” are handled separately in the mass-balance audit. The available figures do not support that explanation.

- **17.07**: Baziaș: 1,750 m³/s, the descent series begins _[documented fact · S01]_
- **18.07**: Paks cuts 160 MW over water temperature _[documented fact · S07]_
- **27.07**: Paks: record −106 cm; U1 −254 MW. At Cernavodă the water is 27 °C _[documented fact · S06]_
- **28.07**: 11:00: Cernavodă U1 (~700 MWe) shuts in a controlled manner under severe-drought procedures _[documented fact · S03]_
- **28.07**: Viking Ullur grounds at ~km 817 near Vidin; 186 passengers evacuated _[reported · S13]_
- **29.07**: Budapest: 23 cm, all-time low. Paks takes U3 offline _[documented fact · S13]_
- **30.07**: MVM: full Paks shutdown is “inevitable within 24–72 h”; level is the constraint. Nuclearelectrica reverses the U2 shutdown decision _[documented fact · S47]_
- **31.07**: Paks U4 at −50% _[documented fact · S08]_
- **01.08**: 05:45: Paks U1 offline. Cernavodă reports −209 cm _[documented fact · S09]_
- **02.08**: 01:30: Paks U4's last turbine begins shutdown; only U2 left (~50%) _[documented fact · S48]_
- **03.08**: Second Pârjoaia blast succeeds (the first, on 02.08, had failed): ~4 m of rock removed. RO: est. imports ~1,700 MW at a 7,300 MW peak _[reported · S20]_
- **04.08**: 00:30: Paks reaches within millimetres of its first-ever full shutdown (83.68 vs 83.63 mBf); then the level rises _[documented fact · S10]_
- **04.08**: Romania announces an August alert state and 7M lei; Viking Ullur refloated; Ruse −94 cm _[reported · S15]_
- **05.08**: Paks: −134 cm at 07:00, the event low; output ~230 MW; MAVIR peak 6,750 MW (19:45). Baziaș: 1,450 m³/s, stationary _[documented fact · S49]_
- **06.08**: Baziaș: 1,400 m³/s, the 1985 absolute record matched. INHGA forecast: stationary through 13.08, 07:00 _[documented fact · S01]_
- **06.08**: Paks: level +4 cm/24 h (≈−130 cm); water 27.9 °C. The full shutdown did not occur _[documented fact · S49]_
- **08.08**: Baziaș remains at 1,400 m³/s; forecast indicates a slow rise toward 1,500 by 15.08 _[documented fact · S58]_
- **09.08**: Paks −128 cm, Unit 2 at 228 MW; Transelectrica: 684 MW nuclear in Romania _[documented fact · S59 · S60 · S62]_
- **13.08**: Nuclearelectrica officially announces in a statement addressed to the Bucharest Stock Exchange that Cernavodă U2 also enters a controlled shutdown; the Associated Press quotes the statement _[documented fact · S63]_
- **14.08**: INHGA: Baziaș 1,400 m³/s; forecast declines to 1,300 m³/s by 21.08 _[documented fact · S64]_
- **14.08**: Hydroinfo: Paks −121 cm on the morning of 14.08; forecast −134 cm on 17.08 _[documented fact · S65]_

## What happens if the water falls further?

Four sliders, Baziaș flow, water temperature, Paks level and evening peak, feed one transparent static balance with four labeled states: unconstrained / watch / constrained / unavailable. The result is a scenario with declared rules and no invented resilience score.

The model uses simple formulas, shown in full: available = base internal production (4,000 MW) + imports (1,700 MW) − hydro loss − Cernavodă loss. Hydro loss = 30% × 4,000 MW × 40% × min(1, max(0, (1,600 − flow)/300)); Cernavodă loss = 700 MW if flow < 1,500 m³/s. Each rule is an assumption and appears in scenario-parameters.json. The sourced Paks thresholds are −107 / −134 cm and the 29.5 °C internal limit.

The instrument separates the two balances: it calculates Romania's evening peak, while the Paks constraint belongs to Hungary's system. Paks can dominate the regional picture, at ~2,000 MWe on normal days and ~230 MW on 05–06.08, without being subtracted from Romania's balance. The defaults use the 06.08 observations: 1,400 m³/s, 27.9 °C, −130 cm and a 7,300 MW peak. The reset button restores them.

> Deep dive: The ENTSO-E series are absent because the API returned HTTP 401 and required credentials at cutoff. The block is documented; available figures come from Transelectrica and ministry statements with reported status. The instrument tests sensitivities; it does not provide dispatcher precision. Without JavaScript, the scene remains readable as an explanation using the 06.08 values.

## “Are upstream countries holding water?” The arithmetic does not support the claim

The deficit versus the July average is 3,300 m³/s, or 285.1 million m³ per day. Holding that water for two weeks would require ~3.99 billion m³, nearly twice the Iron Gates I reservoir. The available evidence does not support this explanation.

The arithmetic, step by step, with the formulas on the table: the July multiannual average at Baziaș is 4,700 m³/s; the 06.08 measurement is 1,400 m³/s; deficit = 4,700 − 1,400 = 3,300 m³/s. Missing volume per day = 3,300 m³/s × 86,400 s = 285.1 Mm³/day. Over 14 days: 285.1 × 14 ≈ 3,991 Mm³ ≈ 3.99 billion m³. Against the August average of 3,900 m³/s, the deficit is 2,500 m³/s and 216 Mm³/day. Discharge measures flow, volume measures stored water, and transit time measures movement.

The only reservoir of comparable volume on the stem is Iron Gates I: over 2,000 Mm³ over 120 km, Romanian–Serbian and jointly operated. Sources describe Serbia as a partner in managing the crisis. Austrian and German plants are mainly run-of-river; their output follows instantaneous flow and does not provide retention at the required scale. An extra release of 500 m³/s for five days would mean 216 Mm³, under 11% of the Iron Gates I reservoir. That could smooth variations for several days; it would not explain a sustained deficit.

Water leaving Austria today reaches Cernavodă in about two weeks (Apele Române: 7+5 days across reaches). Retention in July should have left a detectable variation downstream. The Baziaș descent is gradual and consistent with the regional precipitation deficit, including Hungary's −80 to −130 mm over 90 days, and upstream snapshots such as Hofkirchen at 182 m³/s, about 30% of its norm. This conclusion is an analytical inference from documented figures (C16).

> Deep dive: The rumour and its variants are documented in the claim register. The mass-balance test is reproducible: every figure appears in claims.json and the source ledger. Multiannual averages have their own ranges; we use INHGA's rounded values and keep that uncertainty in the interpretation.

## Barges at 30%: cargo left at the quay

Hungarian cargo transport near standstill; Austria officially reports cargo shipping “largely at a standstill”, with a 2.0 m fairway in the Wachau and 1.8 m east of Vienna; Bulgaria limits draft to 160–180 cm by sector; Romanian barges load at ~30% versus the usual 80–90%; the Viking Ullur, with 186 passengers, grounded near Vidin and was refloated only on 4 August.

At low water, the draft limit turns directly into cargo left behind. With limits of 165 cm at Batin and 160 cm at Belene, convoys split their loads or wait. The Viking Ullur grounding at ~km 817 (28 July) also showed the safety risk: 186 passengers evacuated, days of blockage on an international sector.

The economic effect propagates from the river: grain and ore that cannot board barges move by truck or rail, at higher cost and lower speed, or remain at origin. In 2018, a similar low-water year, the volume carried on the Austrian Danube fell 25.1% (9.6 → 7.2 million tonnes), and in 2026 low water arrived unusually early. Operators report the association between low level and the 30% load factor; a public barge freight-rate series was unavailable at cutoff and is marked unknown.

> Deep dive: Budapest reached 23 cm on 29 July, an all-time low, and river traffic was further interrupted by the closure of the Margaret Bridge zone for works on a Second World War explosive device. At Novi Sad, boats were left on the sand. Austria is piloting “flexible infrastructure”, with gravel barges that temporarily deepen the fords. Low water produces both commercial costs and risks for river cities.

## What can be fixed today. What must be designed for tomorrow.

Short term: the Pârjoaia rock blasted on 3 August, about 4 m of rock removed, 4 barges × 500 t of stone at the Bala bifurcation (km 346–345), AFDJ dredging, 7 million lei, a national alert state for August, and support requested from Ukraine. Estimated intake effect: +6–10 cm. On 13 August, the Associated Press reported a controlled shutdown of Cernavodă U2. Long term, the Kozloduy lesson is to design an intake that tolerates low water.

The emergency measures target the bifurcation at km 346–345 near Izvoarele/Lipnița, where the Bala branch takes ~85% of the flow. The 2 August blast (30 kg) failed; the 3 August blast (100–180 kg) removed about 4 m of rock at Pârjoaia. The plan continues with sinking stone barges (70×10×3 m) and works at both ends of the branch. Apele Române estimates +6–10 cm in the suction area; the initial variant said 10–15 cm. The gain may help temporarily, but it does not remove the structural water dependency. INHGA says Cernavodă-reach estimates remain indicative while the works continue.

System-wide, Romania entered a national alert state for August, allocated 7M lei for the works and requested emergency support from Ukraine. Evening imports were estimated at ~1,700 MW; the reported scenario exceeded 3,000 MW if U2 also stopped. Hungary saved 700 MW through demand response on 2 August, including 240 MW of voluntary industrial response. MAVIR peaks were 7,488 MW (29.06), 7,282 MW (03.08), 6,750 MW (05.08, 19:45, below the 7,000 estimate with a Dunamenti unit outage) and an estimated 6,700 MW on 06.08. Bulgaria, a net exporter, supplied neighbouring systems for balancing.

> Deep dive: One minister said the structural Bala-bifurcation rebalancing project has been stalled “since April”; another statement by the same minister says “since 2024”. The claim keeps its single attribution and records the conflicting variants (C24). Separately, the energy ministry said the Olt and Argeș reservoirs release ~100 m³/s and 40–45 m³/s respectively for navigation depths. That water supports navigation and does not reach Cernavodă.

## Every figure has a source. Start here.

The source ledger, claim register and normalized datasets can be downloaded below. The canonical cutoff is 14.08.2026, 13:15 EEST (UTC+3), or 10:15 UTC.

What remains unknown at cutoff: Cernavodă's 2026 numeric intake limits, credential-gated ENTSO-E series, a river-freight tariff series, cooling constraints for Romanian coal plants, and Paks output after the 09.08 OAH snapshot. The Paks level on 02.08 was removed because its source could not be verified. Missing values stay missing and are not interpolated. The gaps are listed in manifest.json.

PEGELONLINE data can be re-retrieved from the public endpoints listed in stations.json. The six German and Austrian gauges remain snapshots from 09.08.2026 at 11:15–11:30 CEST. Hydroinfo published −121 cm for Paks on the morning of 14 August and forecast −134 cm for 17 August; the page does not publish an exact time for the morning reading. Unit output remains the OAH snapshot from 09.08 at 11:34 CEST.

> Deep dive: The conclusion is simple: the river constrains a network, not one plant. For the 10-second read, remember 1,400 m³/s, the 1985 record matched on 6 August 2026, and the INHGA forecast of 1,300 m³/s by 21 August. For the 2-minute read, follow the chain: flow → level → suction/temperature → MW → imports. For the deep dive, open the claim register and source ledger.

- **S01** · INHGA: Daily hydrological bulletins (hidro.ro) _[documented fact]_
- **S03** · Nuclearelectrica / Agerpres: Controlled U1 shutdown, 28.07 (agerpres.ro) _[documented fact]_
- **S06** · MVM Paks Zrt.: −106 cm low, KöM 15/2001 and operating thresholds (mvm.hu) _[documented fact]_
- **S10** · Portfolio.hu: The night of 04.08 at Paks (portfolio.hu) _[reported]_
- **S14** · Digi24 / Apele Române: Two-week transit, imports, 10–15 cm (digi24.ro) _[reported]_
- **S20** · Romanian press cluster: Pârjoaia blasting and the 85/15 Bala split (multiple) _[reported]_
- **S21** · CNCAN: Cernavodă stress test: ~90 m³/s (PDF) (cncan.ro) _[documented fact]_
- **S23** · IAPPD / Bulgarian press: Ruse −94 cm and 160–180 cm draft limits (appd-ruse.org) _[reported]_
- **S24** · Hidroelectrica: Iron Gates I: 1,166.4 MW and >2,000 Mm³ (hidroelectrica.ro) _[documented fact]_
- **S58** · INHGA: Baziaș 08.08, historical snapshot (hidro.ro) _[documented fact]_
- **S59** · OVF Hydroinfo: Paks −128 cm on 09.08, 10:00 CEST (hydroinfo.hu) _[documented fact]_
- **S60** · OAH / HAEA: Paks U2 at 228 MW; U1/U3/U4 at 0 MW (haea.hu) _[documented fact]_
- **S61** · WSV PEGELONLINE: Six German and Austrian gauges, re-retrieved on 09.08 (pegelonline.wsv.de) _[documented fact]_
- **S62** · Transelectrica: Romanian nuclear output of 684 MW on 09.08, 12:43 EEST (transelectrica.ro) _[documented fact]_
- **S63** · Nuclearelectrica, quoted by Associated Press: Official U2 controlled-shutdown announcement, 13.08 (apnews.com) _[documented fact]_
- **S64** · INHGA: Baziaș 14.08 and forecast through 21.08 (hidro.ro) _[documented fact]_
- **S65** · OVF Hydroinfo: Paks −121 cm on 14.08 and forecast (hydroinfo.hu) _[documented fact]_
- **S52** · Danube Commission: Official chainage: Sulina 0 … Donaueschingen ≈2,850 (danubecommission.org) _[documented fact]_

## Reading test: can you answer these seven questions?

**What happened to the Danube in July–August 2026?** Flow at entry into Romania fell to 1,400 m³/s on 06.08, matching the 1985 record and remaining under 30% of the July average. On 14.08, INHGA still recorded 1,400 m³/s and forecast 1,300 m³/s by 21.08.

**Why did Cernavodă shut a unit?** Its cooling basin is fed from a branch that gets ~15% of the flow at low water; U1 shut in a controlled manner on 28.07. On 13 August, Associated Press reported a controlled U2 shutdown, marked reported here.

**Why was Paks close to total shutdown?** The constraint is water level at the pump suction mouths. The generation threshold is −134 cm; Hydroinfo listed −121 cm on the morning of 14 August and forecast −134 cm for 17 August. Later unit output is not included without an official source.

**Why had Kozloduy reported no power restriction?** Its intake sits in a deep bay, below 17 m, designed for a 100-year event; at the source date, both units were at planned load.

**Are upstream countries holding water?** The available evidence does not support that explanation: the 3,300 m³/s deficit would mean ~3.99 bn m³ hidden in 2 weeks: nearly 2× the Iron Gates I reservoir.

**What does it mean for freight?** HU and AT near-standstill, 160–180 cm draft limits in BG, RO barges at ~30% load, a passenger ship grounded for 7 days.

**What comes next?** Emergency: works at the Bala bifurcation (km 346–345), with an estimated +6–10 cm effect. The 14.08 INHGA forecast declines to 1,300 m³/s by 21.08. Resilience: deeper intakes and a grid with alternatives.

---
Static page, no tracking. Values are labeled snapshots at cutoff 14.08.2026, 13:15 EEST (UTC+3). Non-governmental content compiled from fully listed public sources.

Text: CC BY 4.0 · Data: CC0 (subject to primary-source licences) · Fonts: SIL OFL 1.1 (served locally)