Onde está a água?

Start with the resource, not the tap. Portugal has a large amount of renewable water and controls rather less than half of it.

Two intergovernmental bodies measure the same country and reach numbers that differ by four percentage points. Eurostat and the OECD put Portugal's renewable freshwater at 73,593 million m³ a year, of which 35,000 flows in from Spain. FAO's AQUASTAT puts the total at 77,400 with 39,400 external. Neither is wrong; they account for border rivers differently.

What matters is that every credible source lands the dependency ratio between 44.7% and 57.7%. The often-quoted "two-thirds" is not the national figure — it is the Tejo specifically, and even there the measured 2024/25 share was 62.4%.

Excluded, deliberatelyA seventh calculation exists and is not shown in the range above. Pooling Eurostat's twelve reported annual observations for 1995–2018 gives 62.3%. It is arithmetically correct and methodologically incomparable: it sets individual wet and dry years against a long-term-average denominator. Quoting it beside the others would inflate the range by five points on a definitional artefact.

The fork this study had to decide

Portugal's water stress has two authoritative answers that point in opposite directions, and both verify in their own source.

Eurostat's Water Exploitation Index (WEI+) ranks Portugal more stressed than Spain — 7.27% against 7.12%, with the EU27 aggregate at 5.15%. FAO's national water-stress indicator ranks Portugal far less stressed than Spain — 12.3% against 43.3%.

This study leads with Eurostat WEI+ and shows FAO alongside. The reason is the denominator. FAO divides abstraction by total renewable resources including the full external inflow, so a country that depends heavily on a neighbour looks comfortable precisely because the neighbour's water is counted as its own. WEI+ nets out returns and is computed on renewable resources at a scale that can be disaggregated to river basins — which is where Portuguese scarcity actually lives. Neither is discarded; both are on the page.

The national annual figure hides the thing that matters twice over — once in space, once in time. Below is the time dimension, and it is severe: the same country that abstracts 5% of available water in January abstracts 84% of it in August.

Provenance, not silenceFive of the eight basin WEI+ values are not published by APA. APA publishes three — Sado e Mira 74%, Ribeiras do Algarve 66%, Guadiana 51%. The other five trace to a study commissioned by APA in 2021 from Bluefocus / Nemus / Hidromod, "Avaliação das disponibilidades hídricas por massa de água e aplicação do Índice de escassez WEI+". Three of those five — Tejo 48%, Cávado 43%, Vouga/Mondego/Lis 42% — exceed the severe-scarcity threshold, and they are precisely the hydro-cascade basins where the multiple-counting trap in section 04 lives. Whether that study computed WEI+ on gross or on consumptive abstraction decides whether those five values are usable. They are shown, marked, and not treated as equivalent to the three APA publishes.

Spain. Under a treaty signed in 1998, revised in 2008, whose obligations are a floor rather than a share — and whose floor moves.

The Convenção de Albufeira sets minimum volumes that must cross specified control sections in each hydrological year, each quarter, and — since the 2008 revision — each week. It is a real, enforceable, jointly monitored instrument, and it is unusual among transboundary water agreements in having twenty years of published compliance reporting behind it.

It is also much smaller than it sounds.

Derived — arithmetic shown The guarantee is about 31% of the average inflow.

And it erodes The exception thresholds are set as a percentage of a reference precipitation computed over 1945/46–2006/07 and updated every five years. As the climate dries, the reference mean falls, so the precipitation level that triggers suspension of the minimum flow falls with it. The obligation is defined relative to a baseline that moves in the same direction as the problem.

SourcedSpain has breached, outside any exception. In 2021/22 two stations fell short with no exception available: Douro at Saucelle–Águeda, 91% of the annual minimum — a shortfall of 347 hm³, and Tejo at Cedillo, 86% — a shortfall of 383 hm³. The exception clause was not available at either: accumulated precipitation was 67.8% of reference against a 65% trigger in the Douro, and 90.6% against a 60% trigger in the Tejo. In 2016/17 the treaty stations Miranda (3,199 hm³) and Bemposta (3,107 hm³) delivered 91% and 89% of the 3,500 hm³ minimum, also under the normal regime — the Douro precipitation test came in at 66% against a 65% trigger, one point above the threshold that would have excused it. Salto de Castro was also short at 89%, but it carries no treaty obligation: Article 3 lists the Douro sections exhaustively and Castro is absent. It is a Spanish hydroelectric dam used as a measurement proxy for Miranda, so citing it as the breach mislocates it.

Derived — arithmetic shown The shortfall volumes below are the difference between the obligation and what was delivered, from the same CADC table.

Portugal, in the same year, met every one of its own obligations while qualifying for drought relief. Ponte de Muge was under a declared exception for the Portuguese sub-basin and still delivered 100% of the unreduced minimum; Crestuma delivered 130%.

Read the year, not the verdict 2024/25 was fully compliant, and it was wet. Minho at Frieira delivered 244% of its annual minimum, Douro at Saucelle 243%, Tejo at Cedillo 365%, Crestuma 327%, Ponte de Muge 457% and Guadiana at Badajoz 219%. Sixty-seven of seventy-eight monitored Portuguese reservoirs were above 80% of capacity at 31 May 2026. Compliance in a wet year says nothing whatever about a dry one — and the chart below shows how tightly the delivered volumes track the weather rather than the obligation.

Corrected against the primary source Three 2024/25 cells in the grid above were re-read from the CADC report itself and changed. The intermediate extraction this page was built from put Crestuma's 327% on the Cedillo row, and carried no figure at all for Crestuma or for Badajoz. The report is unambiguous: "Em Cedillo o volume, neste ano hidrológico, foi de 9.853 hm³, correspondente a 365% do volume anual mínimo"; "Em Crestuma o volume total registado … foi de 16.375 hm³, que corresponde a 327%"; "No Açude de Badajoz o volume total observado foi de 1.096 hm³, que corresponde a 219%". The corrected cells carry a blue underline in the grid. Worth knowing about that last one: the Spanish column of the same paragraph misprints the volume as 1.906 hm³ while the Portuguese column says 1.096 — and 1,096 ÷ 500 = 219% settles which is right.

And Portugal largely cannot verify any of it

The Convention's control sections are monitored, but the monitoring is not symmetrical. On the river where Portugal's dependence is highest, there is no Portuguese measurement at all.

Not measuredThere is no active Portuguese gauge on the Minho mainstem. Every Minho-basin entry in the register — Mazedo, Foz do Mouro, Segude, Casais, Messegães — is on a tributary. Verification of the 3,700 hm³ owed at Frieira rests on Spain's own data from the Frieira dam. A blind re-derivation on 2026-07-29 confirmed the absence within the CADC compliance reports and went further: Portugal independently gauges only two of the six stations Spain is obliged at — Miranda and Bemposta, both Portuguese dams. Frieira, Saucelle+Águeda, Cedillo and Badajoz are measured and reported by Spanish basin authorities, and every precipitation reading that decides whether Spain owes anything comes from AEMET, the Spanish meteorological agency. This is an inference from station names in a scrape, not a proven absence: SNIRH returned 403 and the register could not be re-queried directly.

Estimated, not measuredGuadiana at Pomarão is Portugal's own obligation, and its daily values are estimated from Pulo do Lobo upstream rather than measured in situ. The 2024/25 annual report states only that the daily mean stayed above the 2 m³/s minimum. No annual volume is published for the station at all.

DecommissionedBarca de Alva, the historic Douro border gauge, is EXTINTA. So is Vila Velha de Ródão on the Tejo. And so, according to the same register, is Ponte de Muge — which CADC reports through December 2025 still cite as a live Convention control station. That contradiction is unresolved and is recorded rather than smoothed over.

The escape hatches, and where the ground shifts

Each basin has its own suspension test, and they are not alike. The Minho suspends below 70% of reference precipitation; the Douro below 65%; the Tejo below 60%, with a second two-condition trigger at 70%. The Guadiana does not use precipitation alone at all — it uses the volume stored in six Spanish reference reservoirs on 1 March, crossed with precipitation, in a table whose bottom rows read simply EXCEPÇÃO: no minimum owed.

Live nowThe regime is being amended. The 4th Conference of the Parties, Faro, 23 October 2024, agreed a new monthly regime for the Guadiana at Pomarão — the first time that section has had anything beyond a daily mean. Proposta de Resolução 18/XVII, tabled 23 June 2026, approves the Second Revision Protocol carrying it. The new regime is monthly, keyed to a hydrological class, and includes an explicit socioeconomic volume on top of the ecological flow.

Carried, not re-extractedSpain's own Douro basin authority is reported to project its non-compliance rate rising from 11% of years now to 21% by 2039. This figure reached the research record through a delegated agent and was not personally re-extracted from the Spanish plan in this build. It is carried here with that caveat attached rather than dropped or laundered.

Portugal has no official answer to "how much water can we store". It has three unofficial ones, and they disagree by 1,600 hm³.

Each of the figures below is real, verbatim, and describes a specific named population of dams. Correction, 29 July 2026: an earlier version said none of these is a government statement of a national total. One is — the PGRH third-cycle national summary, Quadro 4, states 263 large dams and 14,804.71 hm³, and summing the eight regional documents independently reproduces it to the last digit. The disagreement between populations is real; the absence of any official total was not. The counter-intuitive part is that the nominally broadest registry produces the smallest number.

Unresolved — history belowThis page has reported that APA's weekly reservoir bulletin contradicts itself: 13,185 hm³ of total capacity on the front page against 13,299.2 hm³ summed line by line from its own per-reservoir appendix — a difference of 114 hm³, 0.87%. That reading has changed status three times. It is carried here as reported, but not independently re-tested — which is a weaker position than either earlier version of this page took.
29 July 2026 — withdrawn. The bulletin could not be retrieved: the cited path returned HTTP 404 with an HTML error body, the SNIRH copy returned 403, and no Internet Archive snapshot was found. A known-positive PDF on the same host returned 200, so the fetch route was working.
1 August 2026 — reconfirmed. The published path had shifted. The bulletin was reachable again, and both figures were reconfirmed by direct reading of the 27 July 2026 edition — front page 13,185 hm³, appendix sum 13,299.2 hm³, two independent derivations. The internal inconsistency remained unresolved in the source.
12 August 2026 — could not be re-tested. Reconciling the two language editions, neither figure could be checked again: apambiente.pt and snirh.apambiente.pt refused the connection outright from the checking environment, the Internet Archive was not reachable to that checker either, and its web-search quota was exhausted. Control fetches to unrelated public sites returned 200, so the checker's own route worked — but a restriction at the checking end is not evidence about what the bulletin does or does not say. Neither confirmed nor refuted.

Retracted hereAn earlier pass of this research reported that the CNPGB registry holds 136 dams and omits Alto Rabagão. A direct recount across all nine pages gives 171, and Alto Rabagão, Alqueva, Castelo do Bode, Vilarinho das Furnas, Aguieira and Fagilde are all present. The registry was not incomplete; the scrape was. Its 12,270 hm³ total was summed from 131 of 171 records and should not be used. It is shown above as an explicit null rather than deleted, because the number is in circulation.

Not published, nationallyUseful capacity does not exist as a national figure. The weekly bulletin publishes only capacidade total; none of the eight PGRH regional inventories carries a volume útil column at any level of aggregation. Grepping for útil against a working control returns zero hits. Total capacity includes water that cannot be abstracted.

Derived — arithmetic shown One reservoir is more than a quarter of the country's storage.

The current position is comfortable and recent, read from the same un-retested bulletin: at 27 July 2026 the monitored reservoirs held 10,762 hm³, 81.6% of the capacity the bulletin states [verify] — inheriting the status of the note above, because it comes from that same document. Three years before that, several of the same reservoirs were at a tenth of theirs.

[verify] — the two editions disagreedThe most recent sourceable national end-of-hydrological-year figure is in conflict, and the conflict is not resolved. This edition carried 30 September 2025, 10,047 hm³, 76% of capacity, attributed to APA's hydrological-year report. The Portuguese edition carried 29 September 2025, 10,067 hm³, 76% of capacity, attributed to APA's weekly bulletin via an Internet Archive snapshot of 5 October 2025. The same nominal datapoint, on a different date and 20 hm³ apart; the two agree only on the 76%. The prior-year comparatives differed too — this edition gave "70% a year earlier and 68% in October 2023", the Portuguese edition gave 9,314 hm³ at 30 September 2024 from the same archived series. On 12 August 2026 an attempt to settle this against the primary source failed: APA and SNIRH were unreachable from the checking environment and the Internet Archive route was unavailable. Neither figure is adopted here. Both are shown, marked, and identical in both language editions until the source can be read again.

Why not something more recentAPA's latest monitoring report, 31 May 2026, gives only counts — 78 reservoirs monitored, 67 above 80% of capacity, none below 40% — and no national volume. The percentages also describe the monitored set, whose capacity is never stated and works out at roughly 13,250 hm³, not the 14,805 hm³ of the large-dam inventory. Dividing one by the other gives 68%, not 76%: they are different populations.

The volume mortoBarragem de Santa Clara, on the Mira. It sat below its minimum operating level continuously from 2019 to March 2025 — water in the reservoir, none of it usable. At 31 May 2023 it was at 34.9%; at 31 May 2026, 97.7%. Its scheme has a potential irrigable area of 12,084 ha, of which 5,586 ha — 46.2% — was actually irrigated in 2024. Four other candidate reservoirs were ruled out on their own area figures.

What is planned, and what is funded

The national water strategy Água que Une costs the storage programme precisely. Raising twelve existing dams would add 47 hm³ for €240 million. Building fourteen new ones would add 508 hm³ for €1,183 million.

Two items have a funding source Of everything in that programme, exactly two entries name a financing instrument — the Marechal Carmona/Idanha raising (€8m, PT2030) and the Foupana dam in the Algarve (PT2030, no figure given). Every other line reads a determinar. Every new dam is at study or feasibility stage. And a search of the PRR's 391,000-project register found zero dam-construction line items. A costed programme is not a funded one.

Do not add these upThe same deck's summary table gives 919 hm³ for "build dams, interconnections and new systems" against the 508 hm³ named for new dams alone. The difference is interconnections and other systems, not extra reservoirs. The 47 hm³ reinforcement figure matches exactly across both slides.

Agriculture, overwhelmingly — but the number depends entirely on whether you count the water that passes through turbines, and that single choice changes every percentage by a factor of twenty-two.

The reconstruction totals 121,668 hm³ for 2018 on the gross measure and 5,417 hm³ of fresh water for uses that actually consume it. Both are correct. The difference is 116,252 hm³ of energy water — 95.5% of the gross figure — which is returned to the river in full and is counted more than once, because the Douro cascade records the same water at each successive dam. Douro basin "abstraction" exceeds the river's entire annual flow.

Use the toggle. Watch what happens to every share.

Derived — arithmetic shown The industry line excludes seawater, and the exclusion is APA's own rule. APA removes seawater from its published chart and states the mechanism verbatim for energy — "a origem de água para este fim é captada no mar". Applying the identical test to industry removes three lines.

The reconstruction then reconciles to APA's own published chart to the second decimal: agriculture 4,164.60 against 4,164.60; urban 794.48 against 794.30. Industry is 308.9 hm³, not the 609.1 hm³ that counting seawater produces, and agriculture is correspondingly 77.8% rather than 73.7%.

Two consequences of that correction First, the ratio between the gross and consumptive totals is 22.5×, not the 21× that circulates — the older ratio was computed before the seawater came out of industry, and both numerators and denominator moved. 121,668 ÷ 5,417 = 22.5. Second, the gross figure is still not a single basis: the energy line itself contains 1,098 hm³ of seawater cooling at Sines, which APA states verbatim is drawn from the sea. Removing it too would give 120,570 hm³. That third figure is not used on this page, but it exists, and anyone recomputing this should know why three defensible totals are possible.

A trap with a date on itThere is no 2024 sector balance. A card in circulation reads "VOLUME DE ÁGUA CAPTADO POR SETOR (2024) — 70% Agricultura · 13% Urbano · 5% Industrial · 12% Outros. Fonte: APA". It renders the PGRH 2018 dataset: the figures were extracted verbatim from the page's own inlined chart data and cross-checked against this reconstruction — agriculture 4,164.60 against 4,164.60, exact. On REA, "Fonte: APA, YYYY" is the year APA supplied the figure, never the data year. The same page family carries Fonte: APA, 2026 over 2024/25 hydrological-year data.

Not measuredAgriculture's number is a model, not a measurement. The PGRH computes it as irrigated area × crop coefficient ÷ conveyance efficiency. 71% of Portuguese farmers have no water meter and 61% do not pay for water directly. The sectors that are metered — urban and industry, from tax declarations — agree across sources to within 2.5%.

How much of it comes from underground

The 55%-versus-70% conflict over groundwater has a clean resolution: they are answers to different questions, and the figure most often quoted matches neither.

Groundwater's share of irrigation water is 54.8% nationally and 82.7% in the Algarve. The genuine 70% is a different statistic entirely — agriculture's share of all captured water, which APA states verbatim. Quoting either 70% in support of the other is the trap.

Ninety-three groundwater bodies are assessed under the Water Framework Directive. Eighty-one are in good quantitative status; twelve are not. The Algarve holds five of the twelve — Querença-Silves, São João da Venda-Quelfes, Chão de Cevada, and both Campina de Faro subsystems.

Sourced, and unquantifiedThirteen of the Algarve's twenty-five groundwater bodies show falling levels. The one that rose did so only because irrigation on it stopped — Luz-Tavira, "deixou de se usar para rega". Saline intrusion is formally coded on exactly one body, Campina de Faro–Vale de Lobo, attributed verbatim to "as extrações existentes do setor agrícola e do golfe". No chloride concentration, no extent and no affected volume is published anywhere.

One eighth of Portugal's water runs through a system that measures its losses carefully. Three quarters runs through systems that do not measure them at all.

The urban distribution network is the measured part: 214 retail operators, each reporting to the regulator, published annually. 26.5% of the water entering those networks is never billed. That indicator — água não faturada — is not a synonym for leakage, and ERSAR is explicit about it. It contains three different things.

So of every hundred litres entering a Portuguese distribution network, about twenty leak into the ground, roughly four are lost to metering error and unauthorised use, and about three are delivered deliberately and never billed. Nationally that is about 81 hm³ a year — the government's own figure.

Watch the denominatorDo not average the operator column. The unweighted mean across operators is 36.6%; the volume-weighted national figure is 26.4%. Small operators with terrible ratios are numerous but move little water. Averaging overstates national loss by ten points. Individual operators run from 2.9% to 72.7%.

The pipes are being replaced on a two-century cycle

The national strategy defines exactly the right indicator — "percentagem de rede que excedeu a vida útil esperada", with a target below 10%. Its current value is recorded as s/d: sem dados. What is published is the volume rehabilitated, and that is enough.

Derived — arithmetic shown

Portugal is replacing its water network roughly six times slower than its own assumed design life requires, in a network where 86.9% of mains have already passed ten years and whose infrastructure value index has fallen to 0.36, below the regulator's acceptable floor of 0.40 and still declining.

And then the three quarters that is not measured

The asymmetry, in one tableIn APA's Guadiana basin plan a single indicator table carries both sectors. The urban row has a regulator, a year and a number: physical losses 25.3% regionally, 21.2% nationally, 2018, sourced to ERSAR. The agricultural row in the same table has no regulator, no baseline and no number — only a 2027 target of 25%. Portugal's own planning document quantifies the losses of one eighth of its water and leaves the three quarters blank.

What does exist is per-scheme conveyance efficiency for the 33 largest public irrigation schemes: a simple average of 75.5%, from 8% at Alvor to 98% at Alfândega da Fé, with Alqueva at 92.2%.

Two valid national averagesThe same data supports a ~95% volume-weighted efficiency as well as the 75.5% simple average, because Alqueva alone is about 40% of national volume and far better than the median small scheme. Quoting either alone misrepresents the system.

The circularity trapThe obvious move is to multiply agriculture's 4,212 hm³ by a 75% efficiency to get irrigation losses. That would be wrong. The PGRH abstraction figure is already derived by dividing field consumption by conveyance efficiency. Applying the efficiency twice manufactures a large, confident, entirely invented number. No such figure appears on this page.

Not publishedReservoir evaporation is not quantified by anyone. No figure from APA, REN, EDP or EDIA. Academic estimates run 750–1,230 mm per year over the water surface. In a country whose largest reservoir has a surface of 250 km², that is not a rounding error, and nobody publishes it.

Three options are on the table — desalination, reuse, and pricing. Two are infrastructure and are barely built. The third costs nothing to change and is the largest of the three.

The price of a cubic metre depends on who is buying it

A Portuguese household pays €1.35 per cubic metre for water supply at the tap — ERSAR's 2024 reference bill of €13.48 a month for 10 m³, inclusive of the fixed charge, the water-resources levy and VAT. An Alqueva irrigator pays €0.0382 on the gravity network and €0.0706 pumped, before levy and VAT, plus a per-hectare charge.

Corrected 29 July 2026This page previously said €1.61. That figure is ERSAR's revenue divided by billed volume across domestic and non-domestic users and auxiliary services — not a household tariff. Using it inflated the ratio below. An independent re-derivation caught it.

Derived — arithmetic shown

This ratio is a calculation from two separately sourced official figures. No single source states it. The Alqueva rate also carries a fixed conservation charge — €55.22/ha/year at high pressure, €20.08 at low — which the €/m³ comparison does not include.

Currency warningThe 2017 tariff has been superseded. Despacho 1645/2026, 10 February 2026 sets the 2026 campaign and introduces an over-quota penalty structure — consumption at 100–110% of the authorised volume is charged at base price +150%, and at 110–125% at base price +400%. The 2026 base €/m³ figures were not obtained. The structure holds; the exact indexed rates for 2026 are not on this page because they were not found.

AuditedIrrigation tariffs cover operations. They did not build the scheme. The Tribunal de Contas' 2005 follow-up audit of EDIA assessed the Alqueva project at €2.7 billion, of which more than 70% is irrigation infrastructure, and found that estimated tariff and energy revenue for 2004–2006 of €35 million represented "apenas cerca de 7% do volume de investimento previsto" — about 7% of the capital investment planned for the same period. The same report notes EDIA received no operating subsidy in 2003. Operations were roughly self-funding; the state and EU funds built it. This is the only Tribunal de Contas coverage of EDIA in the full 1999–2025 archive — two audits among 1,577 reports, in 1999 and 2005. There is nothing more recent.

Desalination: contracted, not built

The Algarve plant is a €107.9 million mixed design-build-operate contract, of which €54.0 million is PRR co-financing. Phase 1 is designed for 16 hm³ a year, expandable to 24. A symbolic groundbreaking was held on 8 July 2026; completion is expected late 2027 or early 2028.

Not publishedThe plant's energy consumption is not published anywhere. Not in the contract announcement, not on the project hub page, not in the key-figures page. The ≈3.5 kWh/m³ and €0.51/m³ figures that circulate in earlier work on this topic are Porto Santo's, used as an analogy. They are not this plant's, and they are not used as such here.

One property of the plant does matter for the wider system and is documented: a 25% turndown capability, which makes desalination a schedulable electrical load rather than a fixed one. In a grid with high renewable penetration and reservoirs that can hold the output, that is the interesting part.

Reuse: 1.1% nationally, about 15% in the Algarve

National reuse stands at 8.3 hm³ — 1.1% of treated wastewater — against a target of 116 hm³ across 315 treatment plants by 2040, for €137 million plus €12 million from retail and private operators. That is a fourteen-fold increase required in fourteen years. The Algarve is already at ~6.2 hm³, about 15%, across fifteen users — nine golf courses, three private green-space users, two municipal and one environmental — with protocols signed to take it to 9 hm³. Do not extrapolate the Algarve rate to the country.

The contradiction the strategy does not name

Inference — flagged as such by its source Água que Une projects agricultural consumption rising from 3,175 hm³ to 4,093 hm³ by 2040 — plus 28.9% — in the same document that proposes hundreds of millions in efficiency investment. Its 2040 scenario assumes 120,000 hectares of additional or intensified irrigation: 28,000 ha genuinely new, 21,000 ha completing Alqueva, 20,000 ha of increased use in existing schemes, and the remainder from raising utilisation of already-built public infrastructure to 90%. This is a scenario input to a demand model, not a funded plan, and it must not be reported as "the government plans to add 120,000 hectares". The document never names the tension between the two halves of itself.

What is being planted mattersAlmond needs about 2.4× the water of olive — 6,988 against 2,953 m³/ha in the Alentejo, INE 2023. Efficiency gains and crop change pull in opposite directions, and only one of them is in the investment programme.

Not a hypothetical. Eight drought episodes in twenty-two years, two of them covering the entire country, and a supply system that responds with tanker lorries.

2005 remains the benchmark — the only year in which more than half the territory reached the extreme class. 2022 is second, and in a specific sense worse: at 31 August 2022 every square kilometre of mainland Portugal was in severe or extreme drought, 60.4% severe and 39.6% extreme, with no territory in any lower class at all. 2025 was not a drought year. As of June 2026, 36.2% of the territory is normal and 61.1% in mild drought, with SPI-3 putting Minho, Vouga, Sado, Mira and the Algarve in severe drought.

SourcedFagilde, 2017. The reservoir supplying Viseu fell to 15–20% of capacity; the municipality ran 112 tanker loads a day and approved €4.5 million in emergency works. Fagilde holds 3.84 hm³ and has held 3.84 hm³ since 1984. A replacement was announced in 2024. Nothing has been built.

Traced, and not confirmedThe widely-repeated "135,000 people, 100+ trucks" for Fagilde traces to an unidentified archival television clip. The figure on the ministerial record is 96 trucks, on 18 November 2017, with a historic reservoir minimum of 7.1%. The larger number is not used here.

SourcedAugust 2022: 2,346 tanker supply operations nationally in a single month, +140% against the historic mean. Bragança 685, Viseu 479, Vila Real 272. That is what a national supply system looks like when its margin is gone.

SourcedThe Algarve, 2024: mandated cuts of −15% urban, −25% agriculture and −18% golf against 2023, imposed by Council of Ministers resolution explicitly to avoid "rutura efetiva no abastecimento público" — actual failure of public supply. At the time 84% of the region's groundwater bodies were below the 20th percentile.

The climate claims that do not hold

Several figures in wide circulation about Portuguese precipitation could not be sourced, and one is simply wrong. They are listed rather than quietly dropped, because a reader who has met them deserves to know what happened when they were checked.

Two name-versus-content traps portaldoclima.pt is dead — it 301-redirects to the IPMA homepage, and IPMA's own menu link resolves to a stub whose body is the single word Redireccionamento. And "PNACC 2100" does not exist as a document. The real one is RNA2100, the Roteiro Nacional para a Adaptação; the current plan in force is ENAAC 2020. RNA2100's figures — South −15% under RCP4.5 and −35% under RCP8.5 for 2071–2100 against ~1971–2000, from 45 EURO-CORDEX simulations — are the only primary projections located.

Machine-readable, and undocumentedPDSI and SPI for mainland Portugal are available monthly from 1941 and 1942 respectively through June 2026, on an undocumented OGC WMS server at cs2.ipma.pt — discoverable only by reading the JavaScript in IPMA's own pages. It is the single richest Portuguese climate series that is programmatically reachable, and nothing points at it.

What Portugal controls, what it does not, and what the options actually are once the arithmetic is on the table.

"Water autonomy" is usually posed as a question about supply. On the evidence above it is mostly a question about three things Portugal already owns: what it stores, what it charges, and what it measures.

What Portugal does not control

Roughly 47.6% of the renewable freshwater — 35,000 hm³ a year — arrives from Spain. Of that, the treaty guarantees 10,800 hm³. The remaining 24,200 hm³ arrives because it rains in Spain and Spain does not need it, not because Portugal has any claim on it.

And Spain's ability to hold water back is not symmetrical with Portugal's ability to hold it.

Derived — arithmetic shown Spain holds far more Guadiana water upstream than it owes downstream.

The Guadiana obligation is a sliding scale keyed to Spanish storage on 1 March, running from 600 hm³ in the wettest tier down to 300 hm³ and then to EXCEPÇÃO — no minimum at all — when the six Spanish reference reservoirs fall below 2,650 hm³. The ratio between what Spain can hold and what it must release is between sixteen and thirty-two to one, and it is widest exactly when it matters.

The demand on the Spanish side is not static either. The Duero plan records 3,644 hm³/yr of current gross demand, of which 3,346 hm³ is agriculture, and plans 23,930 ha of new irrigation by 2027 with a further 18,962 ha by 2033. The Guadiana plan raises demand from 2,023 to 2,266 hm³ by 2027. Upstream demand is being planned upward at the same time as the downstream guarantee is being recalculated against a falling baseline.

What Portugal does control

The ordering is uncomfortable but it is what the arithmetic gives.

• Agricultural demand is the only lever of the right size. Agriculture is 4,212 hm³. A 10% reduction is 421 hm³ — twenty-six desalination plants, or 3.6 times the entire 2040 reuse target. Nothing else on the list is in that class.
• Urban leakage is real and second-order. The 81 hm³ lost annually to network losses is worth recovering and is roughly five desalination plants — but it is 1.5% of consumptive abstraction.
• Reuse is worth doing and is not a solution. The full 2040 target of 116 hm³ is 2.1% of consumptive abstraction, and reaching it requires a fourteen-fold increase in fourteen years from a base that grew 6.6 → 8.3 hm³ in the last one.
• Desalination is drought insurance, not supply. 16 hm³/yr is 0.3% of national consumptive abstraction and about 9% of the Algarve's. Its value is that it is rainfall-independent and schedulable, not that it is large.
• Storage is the lever that is costed but not funded. 555 hm³ of proposed additional storage — 47 by raising existing dams, 508 by building new ones — for €1.42 billion, of which two line items name a funding source.
• Pricing is the lever that costs nothing and is not being pulled. Agriculture pays 19–35× less per m³ than households, has recovered none of Alqueva's capital cost — €2.03 bn of water-segment assets are fully impaired in EDIA's own audited accounts — and 71% of farmers have no meter to price against.

The honest conclusionPortugal cannot become water-autonomous in the sense of not needing Spanish water. Half its renewable resource physically originates outside its territory, and no combination of desalination, reuse and storage on the table closes a 35,000 hm³ gap — the entire proposed storage programme is 1.6% of it. What Portugal can do is reduce the consequence of a bad Spanish year: store more, lose less, charge for what it uses, and measure what it charges for. The first is costed and unfunded, the second is measured and slow, the third is politically hard and arithmetically the largest, and the fourth is the precondition for the third.

And one thing autonomy is notEvery figure in this section rests on measurements Portugal either does not take or cannot currently reach. There is no gauge on the Minho mainstem. Pomarão is estimated. SNIRH has been dark to automated access for four months. A country cannot manage what it does not measure, and cannot negotiate over what it cannot verify.

What this page cannot tell you, and what was tried before giving up. These are findings, not apologies — several of them are the most important things on the page.

The access failures are infrastructure findings

SNIRH is darkPortugal's national water information system returns HTTP 403 to every client. curl, WebFetch, proxy services, and a real logged-in Chrome browser on a residential IP. The Internet Archive dates it: a 200 as late as 6 March 2026, intermittent 403s from November 2025, and continuous 403 since 29 March 2026. Precipitation, flow, reservoir and piezometric series — unavailable to automated access for four months. One thing was not tested: a Portuguese domestic IP address.

Three routes around it were found and are what much of this page rests on: bulk station shapefiles at sniambgeoviewer.apambiente.pt (named files return 200; the directory listing returns 403), a working ArcGIS REST API at inspire.apambiente.pt, and reservoir storage through the drought commission's monthly PDFs, continuous from April 2019 to May 2026.

A 200 that is not a success diariodarepublica.pt now returns the same 2,346-byte "JavaScript is required" shell to both curl and WebFetch. It was verified with a known-positive — a resolution successfully fetched earlier the same day returns the identical shell. Three separate research strands hit this independently. Direct PDFs on files.diariodarepublica.pt still work if the exact path is known. Related: ffms.pt's own search returns HTTP 200 with results that ignore the query entirely, which reads as "no results" and is not.

What this page could not settle

• Whether five of the eight basin WEI+ values are usable. They come from a 2021 consultancy study for APA. Whether it computed WEI+ on gross or on consumptive abstraction decides the question, and the study itself was not obtained.
• Whether "Água que Une" is one document or two. RCM 133/2026 and a 9 March 2025 prime-ministerial presentation under Despacho 7821/2024 may be the same instrument or two. Blocked on DRE. That single document would also refresh sixteen of nineteen stale records carried forward from earlier work.
• The 2026 Alqueva tariff in €/m³. Despacho 1645/2026 exists and its over-quota penalty structure was obtained. The base rates were not.
• The 4,300–4,400 hm³ aquifer reserve. Widely quoted. No primary source located.
• Whether Ponte de Muge is a live station. The SNIRH register says EXTINTA; CADC reports through December 2025 treat it as the Convention control station.
• Spain's Douro non-compliance projection. The 11% → 21% by 2039 figure reached the record through a delegated agent and was not personally re-extracted.

One error this page caught in itself Building the compliance grid meant reading the CADC annual report against the extraction it came from, and the two disagreed. Three 2024/25 cells were wrong — Crestuma's 327% had been attributed to Cedillo, and Crestuma and Badajoz carried nothing. They are corrected in section 02, marked in the grid, and the verbatim source text is quoted beside them. The correction is shown rather than absorbed, because a page that promises its arithmetic can be overturned has to demonstrate what that looks like. It also means the same class of error may survive elsewhere in cells that were never independently re-checked.

On methodEvery figure on this page carries a source, a year and a type. Where a source could not be found, the probes attempted are recorded rather than the gap being filled. Nothing here is drawn from general knowledge, and no number appears without a document behind it. Where a figure is derived, the arithmetic is shown so that it can be checked and overturned. The sector balance is 2018 because the PGRH third cycle is the most recent full reconstruction that exists; the compliance record ends at 2024/25 because that is the last hydrological year CADC has reported.

What prompted thisThis study was prompted by the FFMS/RTP documentary Onde está a água?, June 2026. Several of its figures are tested on this page. Two hold, several could not be sourced, and one — the Viana do Castelo precipitation normal — is wrong by 19%. That is not a criticism of the film so much as a demonstration of how quickly a plausible water statistic detaches from its document.