Raindrops in the Signal: Open Dataset Turns TV Satellites into a Rain Gauge Network

Every raindrop that falls between a geostationary television satellite and a rooftop dish leaves a fingerprint in the received signal. Rain attenuates microwaves in the Ku band so effectively that engineers have spent decades fighting it as a nuisance to satellite broadcasting. Now a team of researchers has flipped that nuisance into a measurement, releasing the first large-scale openly accessible dataset of satellite microwave link observations designed for precipitation monitoring. The OpenSat4Weather dataset, described in the journal Earth System Science Data, packages five months of received-signal measurements from 215 satellite links across Southern France, together with the conventional weather data needed to turn those signals into rainfall estimates.

The scientific idea behind the dataset is known as opportunistic sensing. Rather than building dedicated instruments, researchers exploit signals that already exist for other purposes. The concept was pioneered nearly twenty years ago with commercial microwave links, the terrestrial backhaul connections of cellular networks, which were shown to double as a vast, unintended rain gauge network. Satellite microwave links extend the same principle to Earth–space paths. A ground terminal receiving television broadcasts from a geostationary satellite records a time series of received signal level, typically in dBm, and any dip in that series can indicate rain along the slanting path between the dish and the spacecraft. Because the signal traverses the entire atmosphere, the measurement geometry differs fundamentally from terrestrial links, and so does the processing challenge.

That geometry is precisely why the new dataset includes more than just satellite data. Only the lower portion of the Earth–space path, extending from the ground up to the height where ice particles have melted into raindrops, actually carries liquid precipitation. Converting an observed attenuation into a rainfall intensity therefore requires knowing the rain height, which is tied to the altitude of the 0 °C isotherm. OpenSat4Weather bundles hourly ERA5 reanalysis values of the 0 °C isotherm height for every link, allowing users to compute how much of the propagation path is rainy at any moment. The dataset also contains 6-minute rainfall accumulations from 113 operational Météo-France rain gauges located within 10 kilometres of at least one satellite station, and radar-derived rainfall intensity along each link path, derived from the gauge-adjusted PANTHERE product of the French national radar network.

The satellite measurements themselves come from HD Rain, a French company that operates a network of low-cost weather stations built around standard satellite television hardware. Each station consists of a dish roughly 75 centimetres in diameter, a conventional low-noise block receiver, a solar panel for autonomous power, and a device called the Rainbox, which logs the received signal power, averages it over 15-second intervals, and transmits the data over the mobile network to a central server. Of the 215 stations, 147 point at the Astra 19.2° E satellite, 67 at Hotbird 13° E, and one at Astra 23.5° E. The published signal levels integrate the upper Ku band from 11.70 to 12.75 GHz at horizontal polarization, sampled at one-minute resolution with a precision of 0.01 dBm.

The observation window runs from 1 August to 31 December 2022, a period chosen to capture the Mediterranean autumn, when the region is most exposed to intense rainfall. The study area along the southeastern French coast is prone to so-called épisodes cévenols, torrential Mediterranean rain events that can trigger devastating flash floods. Thunderstorms contribute roughly 40 percent of the annual precipitation there, a far higher share than in northern France. The region’s complex topography produces highly variable rainfall patterns that sparse conventional gauges struggle to resolve, which is exactly the gap that cheap, decentralized satellite sensors are meant to fill. In 2022 the summer was exceptionally dry, but several severe convective events struck in August and September, including episodes with hail.

The descriptive statistics published with the dataset reveal both the promise and the quirks of the technique. Across all stations and the full period, the median received signal level was −16.4 dBm, with values ranging from −60.5 to −8.8 dBm. The absolute level matters less than its variability, since it depends heavily on installation details such as the length of the coaxial cable between dish and logger. Data availability proved excellent: the proportion of missing values ranged from 0.04 to 20.6 percent across stations, with a median of just 0.09 percent. During a severe rain event on 15 December 2022, a coastal station recorded a signal drop of about 10 dB coinciding with the rainfall peak measured by nearby gauges and radar, a textbook demonstration of the rain-attenuation relationship.

The dataset also documents the pitfalls that anyone hoping to extract rainfall from these signals must confront. Temperature swings on sunny days can modulate the receiver gain, producing signal variations that mimic rain. Water films on the antenna after a storm cause a lingering extra attenuation, the well-known wet-antenna effect, so the signal recovers gradually hours after the rain stops. During downpours the received signal can sink below the minimum detectable level of the receiver, around 14 dB of attenuation for typical hardware, because raindrops themselves radiate microwave energy that raises the background noise. Cable disconnections and wind-induced dish misalignment add further artefacts. The authors illustrate each of these failure modes with real examples, effectively providing a field guide to distinguishing genuine rain signatures from instrumental noise.

Seasonal physics adds another layer of complexity. In August, when the 0 °C isotherm sat near 5000 metres, the rainy portion of a typical satellite path stretched about 6 kilometres, and heavy convective storms produced attenuations up to 12 dB. By December, with the freezing level down to 1500 to 2000 metres and precipitation mostly weak and stratiform, the rainy path shortened below 3 kilometres and attenuations shrank to about 2 dB. The authors note the resulting dilemma: in summer, intense rain can saturate or even extinguish the signal, while in winter the faint attenuation makes establishing a reliable dry-weather baseline difficult. The freezing level itself can swing by one to two kilometres within a few days, which is why the dataset provides dynamic, hourly isotherm heights rather than climatological averages.

OpenSat4Weather fills a conspicuous gap in the opportunistic sensing literature. Several open datasets of terrestrial commercial microwave links already exist, including countrywide data from the Netherlands, the OpenMRG collection from Gothenburg, and recent releases from Italy and New York, but no comparable satellite link dataset had been publicly available. Data ownership has long been the bottleneck: for bidirectional satellite services the operator owns the measurements, and without open data the community cannot develop, compare, and validate processing algorithms. The new release follows the data format conventions proposed by the European COST Action OpenSense, stores everything in NetCDF files with standardized variable names, and is published on Zenodo under a Creative Commons Attribution 4.0 licence, satisfying the FAIR principles of findability, accessibility, interoperability, and reusability.

The authors also offer practical guidance for would-be users: interpolate short data gaps, flag suspicious links using both signal level and variability, cross-validate against neighbouring stations, classify wet and dry periods carefully before establishing baselines, and account for the melting layer when converting attenuation to rain rate. They point to tools such as the pycomlink Python package, developed for terrestrial links, as adaptable starting points, and suggest that machine learning methods are well suited to recognizing rain-induced signal patterns. The ultimate ambition is to accelerate the development of satellite-based rainfall monitoring in regions where conventional networks are sparse or absent. If the approach matures, the constellation of television satellites already broadcasting over most of the planet could become a global, low-cost complement to rain gauges and radars, one dish at a time.

Subject of Research: Opportunistic rainfall monitoring using Ku-band satellite microwave link signal attenuation

Article Title: The OpenSat4Weather dataset: Ku-band satellite link data for precipitation monitoring

Article References: Nebuloni, R., Graf, M., Cazzaniga, G., Mercier, F., & Turko, M. (2026). The OpenSat4Weather dataset: Ku-band satellite link data for precipitation monitoring. Earth System Science Data, 18(10), 7455-7473. https://doi.org/10.5194/essd-18-7455-2026

Image Credits: AI Generated

DOI: 10.5194/essd-18-7455-2026

Keywords: OpenSat4Weather, satellite microwave links, Ku band, opportunistic sensing, precipitation monitoring, rain attenuation, rain gauges, weather radar, ERA5 reanalysis, 0°C isotherm, open data, Southern France