Marius Comper RO

Covasna, Romania · a mofette · April–November 2022

The invisible lake the carbon dioxide pool at Covasna

In a wooden pool in the forest at Covasna, a spa town in Romania's Eastern Carpathians, carbon dioxide seeping from the ground lies like water: almost 80% at the lowest sensor, a few per cent at the highest. The invisible surface is the height at which the concentration crosses 50%. Twenty sensors recorded the concentration every second for seven months. Air pressure explains more of the surface's hour-to-hour changes than temperature does. Scroll through time and watch it.

Move the slider to choose an hour between April and November 2022. Playback advances one day a second.

A pool full of something unseen

Covasna stands on ground that breathes carbon dioxide: the gas rises through cracks and gathers in low places, where it can be dangerous. Some mofettes, as these vents are called, supply supervised gas baths. The wooden pool studied here was reserved for research; it is 1.09 m deep and measures 1.65 m by 1.53 m, and the forestry office of Comandău, in Covasna, hosted the instruments.

Between April and November 2022 Attila Gergely, Ágnes Gál and Zoltán Néda, of Babeș-Bolyai University in Cluj-Napoca, a city in Transylvania, and Alexandru Szakács, of the Romanian Academy's Institute of Geodynamics, placed a rod in one corner with 20 sensors 5 cm apart, the lowest 43 cm above the floor. On average the lowest sensor read 79% carbon dioxide and the highest, at 1.38 m, 29 cm above the rim, 5.5%; the data here reproduce their published values. The 50% surface sat on average 97 cm above the floor, just under the rim.

How it breathes

The surface does not stay put. Through the day it rises and falls by a few centimetres, in two overlapping cycles of 24 and 12 hours. Air pressure has a 12-hour cycle too, the atmospheric tide, so a pressure contribution is plausible; temperature also has a small 12-hour part, so the cycle alone does not show what drives it.

The test answers more directly. With the relationship fitted on April–June only, pressure reduces the error for the July–November hourly swings by 36% (correlation 0.67), temperature by 13% (0.45), and both together by 43%. At the same pressure, a fall faster by 0.4 hPa an hour goes with a surface about 4 cm higher. The hourly calculation leaves much of the variation unexplained; the researchers also saw bursts lasting a few minutes, whose cause remains uncertain.

A summer when the lake rose

Beside the daily swings, the surface's monthly mean rose over the summer: 88 cm in the measured part of April, 95 in June, 103 in August, and it stayed above its April level, around 99 cm, in October. The cause of the rise remains unclear; a single, incomplete year cannot settle it.

Where the gas comes from is debated too: the paper presents a deeper-source hypothesis proposed by Kovács and colleagues as an alternative to the usual "post-volcanic" explanation, which ties the gas to the volcanoes of the Harghita range in Romania's Eastern Carpathians; for this pool the question stays open. The range's youngest volcano, Ciomad, is on another page, and the region's mineral springs on another.

August 2022 · test period-15-10-505104 Aug7 Aug10 Aug13 Aug16 Aug19 Aug22 Augdeviation, cm · 3-hour means · violet: measured · gold: calculated
Three weeks of August 2022, from the test period: how far the 50% surface strayed from its three-day mean, measured (violet) and calculated from air pressure and temperature with the relationship found in April–June (gold).

Share the readings for this hour

The link opens at the selected hour.

Sources and method

Each sensor recorded one reading a second; here they are averaged by the minute, then by the hour. The surface is the height at which the profile crosses 50% carbon dioxide, interpolated between neighbouring sensors when the readings bracket it. The test uses the month split set before the build: the hourly swings of the surface, against a 72-hour centred mean, are fitted on April–June against the pressure anomaly and its three-hour tendency, and separately against the temperature anomaly, then checked on July–November, on the same hours for every version. The centred means also use later hours, so the last 36 hours of June and the first 36 of July are left out of both sets. With the pressure record shifted by 3, 3.5, 4 or 5 days, without wrapping, the score falls to at most 4%. The percentages show how much the calculation reduces squared error against a constant.

  1. A. Gergely, A. Szakács, Á. Gál, Z. Néda, “CO₂ Dynamics in a Mofette: Measurement and Modeling”, Geosciences 15(9), 368 (2025). mdpi.com The pool and sensors, average concentrations of 79% and 5.5%, cycles of 24 and 12 hours, and sudden bursts of gas.
  2. The authors' raw data and scripts. comodi.phys.ubbcluj.ro One reading a second from 20 sensors, April–November 2022.