A lake is a balance sheet you can read from the shore: water comes in, water goes out, some leaves into the air, and the sun warms the surface. In the dry summer of 2026, every line on that sheet moved the wrong way at once. Damien Bouffard, of Eawag in Kastanienbaum and the University of Lausanne’s Institute of Earth Surface Dynamics, has now done the accounting. His report, Swiss lake water levels in the dry summer of 2026 (30 September), is the clearest ledger we have of what a Swiss lake actually is in a dry year.

On 28 September, 12 of the 26 large Swiss lakes with BAFU gauges sat at their lowest late-September level since 1970. Zürichsee read 405.28 m a.s.l. That is 62 cm below the previous record for the date, set in 2009, and 17 cm below the official low-water limit of 405.45 m. On 1 July the level was still normal. Two months later it had fallen about 70 cm. The lake normally moves only a few centimetres, which is why the report puts it at “minus 12” standard deviations on its statistical scale. All 2026 values are provisional.

Inflow, outflow, sky

The inflow column explains most of it. Between July and September, Zürichsee received 23% of its 1970–2025 average, and its catchment is 1% glacier or less. Brienzersee, with about 13% glacier in its catchment, still got 89%. Ice was the buffer, and the lakes without it lost. In August and September only 15–30 m³/s flowed into Zürichsee. From mid-July the canton held the Limmat outflow at about 30 m³/s to protect trout and grayling downstream, and regulation, as the report puts it, “could no longer hold the water level”. From 1 October, Kanton Zürich cut the outflow to 20 m³/s for the rest of the month. BAFU names who pays for that: the Horgen–Meilen ferry, the ZSG fleet, the lake-rescue services.

Then there is the sky. Zürichsee lost about 49 cm to evaporation between July and late September, 17 cm more than usual. That is roughly a fifth of its incoming water, against about 4% in a normal summer. Do the physics on the back of the ledger: 490 kg of water per square metre, at about 2.45 MJ/kg latent heat, is about 1.2 GJ/m². Spread over 89 days, that is about 155 W/m² carried off the surface as vapour, around the clock. Bouffard is careful to say that absolute evaporation is uncertain by about 30%. Treat it as an order of magnitude. That is still a large share of what the summer sun puts into the water at 47° N.

The surface also warmed. Eawag’s Simstrat model, run on the Alplakes platform since 1981, gives the warmest summer surface on record at 23 of 24 lakes. They ran typically +3 °C above the 1991–2020 normal, and up to +4 °C in Vierwaldstättersee and Walensee. Zürichsee’s surface averaged 25.2 °C at 1 m depth. BAFU logged the Aare at Bern at 24.8 °C on 14 July, and about 40% of discharge stations set June–August low-water records. Behind those numbers are the hydrologists who keep the gauges running and the modellers who keep a 45-year lake simulation honest. Their work is invisible in a normal year and essential in this one.

←TODAY: Zürichsee is 17 cm below its low-water limit, with a 25.2 °C summer surface. →3012: Zurich’s lakes are read as thermal batteries, layer by layer, and no building taps the warm top. Fulcrum: The deep cold layer is the asset and the warm surface is the constraint. You only see that when you read the depth and the trend together.

Where the buildings come in

Swiss cities already draw on that cold layer. ewz has built lake-water networks around the Zürich basin since 2003. Seefeld takes water about 100 m offshore at 15 m depth and runs free cooling on 16–18 °C water. ewz’s own page states the heat-wave clause plainly: “Ist die Temperatur des Seewassers allerdings zu hoch, um die Raumwärme aufzunehmen, kommt zur aktiven Kühlung die Wärmepumpe zum Einsatz.” In other words, when the lake is too warm to take the room heat, the heat pump takes over. CoolCity, backed by CHF 303.7 million, will draw from 28 m off the Zürichhorn and return the water to the Limmat. SIG’s GeniLac in Geneva goes to 45 m in the Léman at around 7 °C. ewl in Horw draws at about 44 m, where the water stays near 5 °C. On Vierwaldstättersee, summer cooling water has to go back below the thermocline, at 20–35 m.

The trade-off is plain. The deeper the intake, the colder and steadier the water, and the longer and more expensive the pipe. CoolCity needs 2 km just to get past the shallow basin. On the return side, water-protection law (GSchV) caps river discharge at 25 °C. That is a ceiling, not a design point. A hot summer pushes the intake up and leaves less room below that ceiling.

PAZ’s GLAMOS piece made the same point about runoff, in The Alpine water tower’s 2026 statement: check how old your design data is. The same applies to lake temperatures.

Atelier: If your office connects a building to a lake-water network, its HVAC concept probably assumes the utility’s average supply temperature, not a 2026-type summer. That is a small-looking line in the brief with a large effect on coil sizing, fan energy and comfort. Monday move: ask the utility for the guaranteed summer supply-temperature band and the active-cooling fallback in the connection agreement, then size the building’s chilled-water coils for the top of that band.

Hack: Size the lake-water flow that one megawatt of cooling needs as the intake warms, with the return held under the legal ceiling. The physics is the heat balance Q = ṁ·cp·ΔT. Every kelvin the intake gains is a kelvin taken off ΔT.

cp = 4186  # J/(kg·K), water
def flow(q_kw, t_in, t_ret_max=25.0):
    dT = t_ret_max - t_in
    return q_kw * 1000 / (cp * dT) if dT > 0 else float("inf")
for t in (6, 12, 18, 22): print(t, "°C →", round(flow(1000, t), 1), "kg/s per MW")

At a 6 °C intake you need about 12.6 kg/s per MW. At 22 °C you need about 79.6, more than six times as much. That is why the depth of the intake matters so much.

The systems that last are the ones designed for the hot year, not the average one. Pull your project’s design basis today and write the intake depth and the hot-year supply temperature next to each other.