July 2027

Mobile phone masts could be the key to saving lives in floods – if operators will share their data

The world’s mobile phone networks already function as a vast, unintentional network of rain gauges. The technology works. The science is proven. Now comes the hard part …

Every time it rains, the signal weakens slightly on the hundreds of thousands of microwave links that form the invisible backbone of mobile phone networks around the world. Engineers installed them to carry voice calls and data traffic between base stations – but scientists have discovered that the way rainfall disrupts those signals can be read as a highly accurate measurement of precipitation, in near real-time, at ground level, across entire cities and regions.

The principle is straightforward. The more intense the rainfall, the greater the attenuation – the reduction in signal strength – across a given microwave link. By developing algorithms that compare transmitted and received signal levels, researchers can estimate average rainfall along the path between two antennae with remarkable accuracy. A dense urban mobile network may contain thousands of these links, providing spatial rainfall coverage at a level of detail that would be prohibitively expensive to replicate with conventional rain gauge networks.

The implications are considerable. In countries where weather radar is sparse or absent – much of sub-Saharan Africa, large parts of South and Southeast Asia, Central America, and across the Pacific – these ‘commercial microwave links’, or CMLs, could provide the kind of ground-level rainfall intelligence currently available only to wealthier nations. For flood forecasters, disaster managers, farmers, water utilities, and public health agencies, that data could be the difference between timely warning and catastrophe.

The concept is especially compelling in lower- and middle-income countries, where modern mobile networks are increasingly ubiquitous but conventional meteorological infrastructure remains chronically underfunded. Sub-Saharan Africa has reportedly seen a dramatic decline in functional weather station density since the 1970s. Weather radar coverage across much of South and Southeast Asia, Central America, and the Pacific is fragmented or absent. Establishing and maintaining the kind of observation networks that support meaningful hydrological modelling requires sustained capital investment and skilled technical personnel – resources that are often just not available.

CML infrastructure, by contrast, is often denser and better maintained in rapidly urbanising lower-income countries than in many rural parts of the developed world, penetrating communities that have likely never seen a professional weather station. Every microwave link erected for commercial purposes is, in principle, a potential rainfall sensor. The challenge is unlocking that potential.

The science is settled

The scientific case for CML-derived rainfall estimation has been building for two decades, with research groups in Germany and the Netherlands leading the way.

At the Karlsruhe Institute of Technology (KIT), Professor Christian Chwala and colleagues have been central to developing the theoretical and algorithmic foundations. KIT’s work has produced key advances in correcting for the so-called “wet antenna effect” – where water films forming on antenna housings can distort measurements – as well as baseline signal determination and the statistical characterisation of CML-derived rainfall fields. The team also developed OpenMRG, an open-source Python library that has opened access to CML analysis methodologies for researchers worldwide. Studies in Germany have demonstrated that CML networks can resolve rainfall at spatial resolutions comparable to or exceeding operational weather radar, particularly in urban areas.

At Delft University of Technology in the Netherlands, Professor Remko Uijlenhoet and colleagues have pioneered the move towards operational use. Most notably, in a landmark collaboration with Dutch mobile operator T-Mobile, they produced the world’s first near real-time, country-wide rainfall maps derived entirely from CML data – maps that proved broadly comparable to composite products derived from rain gauges and radar, but with far better spatial coverage in areas with sparse gauge networks. TU Delft research has also done much to quantify measurement uncertainties and explore how CML data can be merged with other data sources to improve estimates further.

Elsewhere, significant contributions have come from researchers in Italy, Slovakia, France, and the Czech Republic. European collaborations under programmes such as COST Action CA16209 (OPENSENSE) have produced multi-country comparative evaluations of data quality and processing methodologies. The evidence base is now extensive.

Germany’s operational experiment

Germany’s national weather service, Deutscher Wetterdienst (DWD), has gone further than most in testing how CML data performs under real operational conditions.

Working with a mobile network operator and the University of Augsburg, DWD developed an automated system combining weather radar, ground rain gauges, and CML data to produce faster and more accurate rainfall estimates, in the BMBF-project HoWa-PRO, with particular focus on early warning for flash floods. The project ran for 28 months from September 2022 to December 2024, generating rainfall maps every few minutes, continuously, for more than a year.

Maximilian Graf, an opportunistic sensing expert at DWD, is enthusiastic about the technology’s potential – while clear-eyed about its limits. The key advantage of CML data, he says, is timeliness, though not every user needs that. “Some end users – people doing city planning, people who want to know how large a drainage pipe needs to be to handle once-in-a-hundred-years rainfall – they don’t need to know what’s happening right now. The people who need real-time data are mainly in weather warnings and flood forecasting.”

During the project, CML data performed well in summer months but proved less reliable in winter, when melting snow and ice introduced errors into rainfall measurements. The project concluded with a proof of concept demonstrating that CML data have significant potential to improve heavy precipitation analyses and, consequently, flood forecasting. However, for the operational use of CML data, data availability remains the primary challenge.

That outcome illustrates a pattern that has repeated itself across CML research worldwide: technically impressive results that nonetheless fall short of full operational adoption.

The limitations

Notwithstanding its considerable promise, CML-derived rainfall data has real limitations that its advocates acknowledge.

The measurements are indirect. Rainfall intensity is inferred from microwave signal attenuation rather than measured directly, introducing uncertainties that correction algorithms can reduce but not eliminate. The wet antenna effect – water accumulating on antenna housings during and after heavy rain – remains a source of error despite significant algorithmic progress. Cold weather complicates matters further, as snow and ice behave differently from rain in their effect on microwave signals.

Geographic coverage can be uneven. Rural and remote locations typically have fewer microwave links than urban areas, limiting the technique’s usefulness precisely where conventional observation networks are also sparse. And because CML systems were designed for telecommunications rather than meteorology, mobile operators may alter frequencies, transmission power, or network configurations for commercial reasons with no consideration of the meteorological implications. Equipment varies considerably between operators, between vendors, and between countries, complicating efforts to develop standardised data processing methodologies.

Raw CML data is also vulnerable to interference from sources other than rain. Atmospheric refraction, seasonal vegetation changes, and hardware-induced signal drift can all mask or mimic rainfall signatures, requiring careful correction before the data becomes useful.

These are solvable problems – and researchers have made substantial progress on most of them. But they are not yet fully solved.

The bigger obstacle: getting operators to share

Technical challenges, though real, are not the primary reason CML rainfall monitoring remains largely confined to pilot projects and academic collaborations. The bigger problem is a financial one.

Mobile network operators have had little incentive to share their data with third parties. The information is proprietary, collected mainly for network management and quality-of-service monitoring rather than meteorological purposes. Accessing it requires either voluntary cooperation from operators – through bespoke commercial agreements or research partnerships – or regulatory mandates. Neither route is straightforward.

Commercial negotiations can be protracted and sensitive, particularly given operators’ concerns about revealing information on network structure and performance that they consider commercially sensitive or strategically significant. The result has been a patchwork of pilot projects and academic collaborations that, while producing excellent scientific results, have fallen well short of the scale, continuity, and service levels required for operational meteorological services.

“The biggest bottleneck is to get mobile network operators to share their data, because it would mean putting in an effort that wouldn’t gain them anything – initially, at least,” says Professor Uijlenhoet of TU Delft.

Moving beyond pilot projects

The most concerted effort to demonstrate the technology’s potential in developing-country contexts has come through the GSMA, the global trade body representing mobile network operators, via its Mobile for Humanitarian Innovation and Climate Resilience initiatives.

GSMA-facilitated pilots in Sri Lanka, Nigeria, and Papua New Guinea – funded respectively by the UK Foreign, Commonwealth and Development Office and Australia’s Department of Foreign Affairs and Trade – have generated valuable proof-of-concept data showing that CML networks in tropical and subtropical environments can produce rainfall estimates accurate enough to be useful in early warning applications.

In Sri Lanka, researchers from TU Delft, Wageningen University, the Dutch meteorological institute KNMI, and GSMA collaborated with mobile operator Dialog Axiata to use CML data from networks around Colombo and other urban centres, exploring potential improvements to flash flood early warning in a country where monsoon-driven flooding causes significant annual mortality and economic loss.

In Nigeria, GSMA worked with MTN, one of Africa’s largest mobile operators, using CML data from networks in Lagos and its surrounding region to illustrate what a country with millions of mobile subscribers but sparse weather observation infrastructure might gain from unlocking its telecommunications network for hydrological purposes.

However, it was Papua New Guinea – partnering with operator Digicel PNG in terrain among the world’s most challenging, and a virtually non-existent conventional observation network – that provided perhaps the most dramatic demonstration of what the technology could offer in extreme data-sparse environments.

The key lesson from these three pilots, says Panos Loukos, Director of Financial Inclusion at GSMA, is that future efforts to support a commercially sustainable project should begin by establishing demand before approaching operators. He says: “Before I go to the mobile operators, I’d have a call of action to actually say, do you need rainfall estimations in Ghana, Kenya, India, wherever there is a lack of radar and meteorological weather stations? Do you need rainfall estimations that we don’t have access to now? Make a commitment that you will buy these rainfall estimations off a mobile operator or whoever else. And once I have this commitment, I’d go to the mobile operator and tell them there is a business case here. We have a long list of organizations in your market that are interested in exploring this.”

Developing a business case is a key issue. Martin Fencl, of the Department of Hydraulics and Hydrology at the Czech Technical University in Prague, and a collaborator on several of the GSMA and other projects, agrees that the business case is the central challenge. It would be technically feasible, he notes, to provide operators with information that improves their network operations in exchange for CML data. But most operators want to see potential revenue streams before committing. He says: “This is difficult because this data has the largest value in low- and middle-income countries where there are no observations – and in these countries it is very difficult to create a business model which would pay off.”

The path to operational services

Moving beyond the pilot stage means shifting from project-based data sharing to institutionalised, contract-governed, long-term data supply arrangements. Two broad approaches are on the table, and the answer may involve elements of both.

The first is commercial: developing a compelling business case that gives operators a genuine financial reason to participate.

According to Professor Remko Uijlenhoet of Delft University of Technology (TU Delft): “The biggest bottleneck is to get mobile network operators to share their data because it would mean putting in an effort that wouldn’t gain them anything, initially at least.”

He adds: “Providing CML data on a continuous, operational basis would assist MNO’s to run their networks more effectively because it would give them information about the functioning of their microwave link network in real-time, and possibly short-term forecasts about expected disruptions due to extreme rainfall. Also, it could give MNOs the opportunity to act as data provider for business operations downstream, notably in the area of weather forecasting and early warning for floods and landslides. Providing data for ‘public good’ could also help them build a positive public image. Finally, space agencies launching earth observation satellites, notably NASA, have also shown an interest in CML data because it could serve as ground validation for their satellite-based precipitation products. We are currently, together with other European partners, involved in a project (EU COST Action) to promote CML data sharing: SetGMDI (https://www.cost.eu/actions/IG20136/).”

The second approach involves government mandates. Under this model, operators would be required to supply rainfall-relevant CML data to national meteorological services as a condition of their telecoms licences – in the same way that operators in some jurisdictions are already required to provide emergency call routing or network access to public safety agencies. Governments could establish the regulatory framework while also offering incentives, technical support, and partnership opportunities to make compliance less burdensome. In practice, hybrid approaches may prove most viable.

What will an operational service cost?

Looking first at the costs of the pilot projects, Mr Loukos says it’s difficult to give an accurate estimate due to the multitude of factors involved in costing these projects. He adds: “Also, our activities were bound by contractual terms & conditions which makes it nearly impossible to disclose amounts or ranges. For example, the cost of the project depends on the cost of labour in the market, the characteristics of the network, the availability of ground-level data sources for validation, and many other factors.”

What about the costs of setting up an actual real-time operational CML rainfall estimation service?

According to Alan Seed, an internationally recognised expert in developing an operational radar rainfall estimation and nowcasting service, if we take the annual salary (or full time equivalent, FTE) of a tech/science developer in Australia as AUD200,000, then in dollar terms, for a credible operational pilot-to-service transition we’re probably looking at AUD 250k–1.0M for the establishment of a CML rainfall estimation system and AUD 100k–400k for its annual operation.

Mr Seed says: “For a small national or regional service with a cooperative operator and modest service level agreement or SLA, the lower half of that range is plausible. For a warning-critical national service with strong availability requirements, audited security, redundancy, and 24/7 escalation, the upper half is more realistic.”

Looking at these costs in more detail, he adds: “Based on my experience with developing an operational radar rainfall estimation and nowcasting service, I would think in terms of 1 – 4 years of full-time equivalent (FTE) research scientist / IT effort to transition a pilot real-time CML rainfall estimation system to an operational service and 0.5 – 2 FTE annually to support operations and incremental improvements to the underlying algorithms.

“Once established, the mobile-network-operator effort to maintain the real-time CML signal-level data feed is expected to be modest, of order 0.1 FTE per year for routine maintenance, provided the feed is integrated into existing network monitoring and support processes. For an operational warning service, this should be supplemented by an SLA allowance covering availability monitoring, incident response, change notification, security management and periodic reporting. A reasonable annual allowance for an operator side data service that includes an SLA to support a warning service would be of the order of 0.25 FTE, not including any licensing fees for the data.”

However, it should be stressed that these costs, while based on the view of someone highly experienced in this field, do not constitute a systematic review of operational CML deployments and actual costs will inevitably depend on the specific context.

Global initiative

An important recent development in efforts to build an operational CML system is the SetGMDI project, or Global Microwave link Data collection Initiative. This brings together European research institutions as well as meteorological services, MNOs and international UN organisations WMO and ITU. The aim is to develop a consortium-based framework to provide secure, scalable access to CML data from operators worldwide for hydrometeorological applications.

Beyond the technical issues, the project is working on possible business cases for operators and developing template legal documents to address data governance questions around licensing, confidentiality, and liability. Such legal templates are essential, says Mr Graf at DWD, to facilitate data exchange and alleviating potential legal or organizational concerns.. He says: “It’s not a personal data issue. It’s not a GDPR issue. We just want performance data for the network – signal levels.” If this initiative succeeds, it could be instrumental in transforming CML rainfall monitoring from a collection of isolated pilots into a key component of the global meteorological infrastructure.

From sensing to warning: the dual role of mobile operators

There is a further dimension to the CML opportunity that makes mobile operators potentially transformative actors in disaster risk reduction – one that goes beyond their role as data sources.

Mobile operators are already the dominant channel for communicating weather-related alerts to at-risk populations. Cell phone broadcast technology allows geographically targeted emergency alerts to be pushed simultaneously to every active handset within a defined area, without prior registration, without internet connectivity, and without any action required from the recipient.

This capability is precisely what is needed to reach, for example, smallholder farmers in rural areas who may have limited literacy, no radio or television access, but who very likely do have a basic mobile handset. An operator that detects rainfall attenuation on its CML network consistent with a rapidly developing storm capable of triggering flash flooding could, in principle, simultaneously generate a rainfall warning product and push it to mobile subscribers in the affected catchment – all within minutes, and all within a single organisation.

This integration of sensing and communication functions within a single actor could dramatically simplify the multi-agency coordination challenges that can plague conventional early warning systems. It does not eliminate the need for authoritative meteorological judgement in the warning chain – the operator’s observation still needs to be assessed within a broader hydrological and meteorological framework – but it could substantially reduce the time from observation to alert in ways that save lives.

The bigger picture: opportunistic sensing

CML rainfall monitoring is part of a broader shift in meteorology towards what researchers call ‘opportunistic sensing’ – the use of data generated incidentally by technologies designed for other purposes.

Traditional weather observation relies on dedicated instruments: rain gauges, weather radars, radiosonde balloons. Opportunistic sensing draws on signals that exist anyway – the attenuation of a mobile phone link, the wiper speed of a connected car, the barometric pressure reading from a smartphone. None of these was designed to measure the weather. All of them, properly interpreted, can.

Looking ahead, CML rainfall data could become one strand in a much richer fabric of incidental environmental observation. Connected vehicles could provide real-time information on rainfall intensity, road surface conditions, and visibility. IoT devices, smart city infrastructure, and connected industrial systems could contribute additional meteorological observations. Wearable devices may eventually add to the picture. The challenge for the meteorological community will be to develop the frameworks, algorithms, and governance structures needed to integrate these diverse streams into reliable, operationally useful products.

But for now, the more immediate challenge remains simpler and more stubborn – persuading the companies that own the data to share it. The science has done its work. The question is whether the economics and the politics can catch up.

Primary sources:

Interviews with

Professor Christian Chwala, Karlsruhe Institute of Technology

Martin Fencl, Department of Hydraulics and Hydrology, Czech Technical University in Prague,

Professor Remko Uijlenhoet, Delft University of Technology

Max Graf, opportunistic sensing expert at Deutscher Wetterdienst (DWD)

Panos Loukos, Director of Financial Inclusion at GSMA

Alan Seed, consultant and expert in developing an operational radar rainfall estimation and nowcasting service

ENDS