Fraunhofer Institute for Digital Media Technology IDMT
How much rain falls at a specific location? In the lokalRAIN research project, Fraunhofer IDMT and Pan Acoustics have developed a method to measure precipitation based on vibrations in photovoltaic modules. Structure-borne sound sensors detect the vibrations triggered by raindrops. An AI model analyzes the signals directly at the sensor node and uses them to estimate precipitation intensity. A functional demonstrator shows how additional, spatially distributed measurement points can be implemented for local precipitation measurement.
Why is local precipitation measurement important?
Heavy rain can occur on a small scale: Rainfall amounts can sometimes vary significantly between locations that are close to one another. Additional measurement points are helpful for better capturing such local differences. Conventional precipitation gauges, such as those at weather stations, involve installation and maintenance costs and are therefore often available only at select locations.
A network of additional, cost-effective measurement points can supplement existing measurement systems. Locally collected precipitation data offers the potential to map the spatial distribution of rain more accurately and to expand the data set for analyzing heavy rain events.
How does precipitation measurement using structure-borne sound sensors and AI work?
When raindrops strike a photovoltaic module, they cause its surface to vibrate. Structure-borne sound sensors detect these signals. A machine-learning model analyzes characteristic features and identifies precipitation events. Based on this, it estimates the precipitation intensity.
Initially, the researchers investigated different precipitation intensities, droplet sizes, photovoltaic modules, and sensor types and positions in the Fraunhofer IDMT rain simulator. Subsequently, three measurement systems were installed under real-world weather conditions: at Fraunhofer IDMT in Ilmenau, at Pan Acoustics in Wolfenbüttel, and in the weather garden at the Technical University of Ilmenau. The measurement data was linked to reference data and used to train and evaluate the AI models.
Signal processing and AI evaluation were optimized for an energy-efficient microcontroller platform. As a result, the calculations are performed directly at the sensor node. Compact measurement and status data are transmitted via an IoT wireless connection. Pan Acoustics developed its own hardware and software solution for the wireless transmission. A central data infrastructure at Fraunhofer IDMT consolidates the information from multiple measurement points.
What results has lokalRAIN achieved?
The project has demonstrated the technical feasibility of acoustic precipitation measurement. The investigations show that precipitation events can be detected based on the vibrations of photovoltaic modules, and precipitation intensities can be estimated directly on a microcontroller using a neural network.
For precipitation intensities between 2 and 5 millimeters per hour, the mean absolute error was 1.88 millimeters per hour. This means that, for low precipitation intensities, accuracy class C of the World Meteorological Organization (WMO) was achieved. This class allows for a maximum deviation of 2 millimeters per hour when measuring precipitation intensity. For higher precipitation intensities, less training and test data is currently available, so accuracy in this range cannot yet be conclusively assessed.
In addition to measurement accuracy, radio transmission was also examined. In tests, the measurement systems achieved a range of approximately four kilometers in urban environments and more than 35 kilometers in rural areas with a clear line of sight.
Where local precipitation data is used
Additional measurement data can supplement existing weather and environmental data. This opens up various potential applications:
Municipalities and emergancy management: Local precipitation data can help fire departments, municipal control centers, and infrastructure operators better assess the spatial extent of heavy rain events. It also offers potential for analyzing drainage systems and flood risks, as well as for planning protective measures.
Wheater services and forecast models: A denser measurement network can provide additional information about the local distribution of rainfall. This data could support the analysis of precipitation events and expand the data foundation for the further development of weather forecast models.
Agriculture and agri-photovoltaics: Location-specific precipitation data is also relevant for irrigation and water management. In the follow-up project SONICrop, 20 sensor nodes will be installed and networked for testing in the field of agri-photovoltaics.
Further development of acoustic precipitation measurement
To optimize the AI models for higher precipitation intensities as well, data collection and analysis for heavier rainfall will be further advanced. In parallel, additional pilot installations will test the practical suitability of the measurement system under various operating conditions and provide starting points for further development in specific applications. To make the transition from a demonstrator to a production-ready system, further developments are also focusing on optimizing sensor housings and electronics, expanding quality assurance, and preparing for future CE conformity.
Responsibilities of Fraunhofer IDMT
Fraunhofer IDMT was responsible for the vibroacoustic acquisition and analysis of precipitation signals. This included investigating suitable structure-borne sound sensors and sensor positions, collecting and processing training and reference data, and developing machine learning models for detecting precipitation events and estimating precipitation intensity. In addition, the institute optimized signal processing for use directly at the sensor node and developed the data infrastructure for consolidating and evaluating the measurement data.
