AGROMET II : local and quality weather data

In a context of climate change

Weather conditions are central to many decisions in the agricultural world. In the context of climate change, applying the same old methods no longer works. It is therefore essential to provide reliable meteorological information to management structures, the research community, and farmers.

Strengthen the reference network

The CRA-W's Pameseb network is an expert network of 35 weather stations whose data is integrated into the Agromet.be platform. For each station, data is recorded every minute for eight weather variables. This represents a considerable volume of data to check, given that sensors are not infallible and errors will always occur at source. During the Agromet II project, several actions were taken to improve the quality of the data reported. Preventive maintenance rounds were carried out at the stations. Tipping-bucket rain gauges, which are prone to clogging, are being replaced one by one with state-of-the-art weighing rain gauges (Figure 1). An automatic quality control system has been put in place to identify suspicious data, supporting quality control and measurement correction by a human operator. The algorithm successfully identifies clogged rain gauges, humidity sensors that saturate too low, and frozen anemometers.

Figure 1. Weighing rain gauge (left in the photo) installed at the Libramont station.

Two birds with one stone

In Wallonia, the Pameseb reference network coexists with private networks consisting of weather stations purchased and installed in fields by farmers or other agricultural stakeholders (Figure 2). As part of Agromet II project, the CRA-W collaborated with the WalDigiFarm weather network, which brings together nearly 200 connected agricultural stations.

The primary objective was to apply the methods developed to improve the quality of the Pameseb network to private stations. Automatic quality control has been adapted and will soon be available to private station owners. An algorithm has also been developed to evaluate the siting of a weather station based on its coordinates and remote sensing data: land use map, digital surface model with a resolution of 0.5 m. This makes it possible to determine remotely whether a private station is located too close to obstacles or heat sources. A study of 108 private stations revealed that 65% of them were poorly positioned.

This work opens the door to combining private networks with the Pameseb network. These networks are complementary: Pameseb stations have high-quality sensors but are few in number. Areas not covered can be filled in by private stations. This approach would be particularly useful for measuring precipitation, which is highly variable in time and space.

The provision of data from private stations for the project was a pioneering case study in the context of agricultural data sharing. The data was shared in compliance with the GDPR and with the agreement of the owner of each station.

Figure 2. Comparison of the number of stations in the Sencrop network (left) and the Pameseb network (right) in part of Wallonia.

Comparing commercial weather stations

At the Humain weather station, near Rochefort, seven commercial station models were studied over a three-year period. Their measurements were compared with two reference stations at the site: a Pameseb station and an RMI station. The portable connected stations proved to be reliable tools, although they required a certain amount of maintenance.

• The temperature probes were accurate and required little maintenance, with errors limited to a few tenths of a degree. The quality of the measurement depended on the quality of the radiation shield.
• The connected rain gauges accurately recorded rainfall and its intensity. However, they are susceptible to clogging and may underestimate or overestimate long-term rainfall totals.
•  Measuring relative humidity, on the other hand, is quite tricky, and several major problems have been observed: overestimations of more than 10% and probes that saturate too low (at 94%, for example, instead of 100%). These probes need to be replaced or recalibrated after a few years. Particular care must be taken not to feed disease models with probes that overestimate humidity or for which humidity saturates too low.
• Cup anemometers slightly underestimated wind speed but remained reliable sensors for tracking fluctuations. The station must be located well away from obstacles that could block the wind. The other type of sensor tested, the 2D sonic anemometer, did not perform satisfactorily.Solar radiation sensors may be slightly biased and cause under- or overestimations, but they record variations in solar radiation accurately and reliably. Recalibration after several years of operation is recommended. It should also be noted that estimating radiation from a solar panel has not been proven for the model tested, but could be a relevant and inexpensive option in the future.

And what about weather forecast ?

Weather forecasts are essential in agriculture, particularly for planning field work. In this project, we studied the performance of 10 forecasting models and determined the best-performing model for each weather variable and at each location in the Pameseb network. An important initial finding was that the ranking of the models did not change depending on the region considered. General conclusions can be drawn for Wallonia. The German ICON D2 model was the most effective at predicting air temperature, relative humidity, and wind classes for a forecast horizon of up to two days. For forecasts beyond three days, the European ensemble model performed very well. It was the best model for predicting air temperature between 3-10 days and one of the best for predicting wind speed beyond 3 days. The Dutch HARMONIE model was the best for predicting the occurrence of precipitation.

The impact of weather forecast model outputs within the SprayVision decision support tool was studied. This tool uses weather forecasts for the next 3 days to calculate the optimal time windows for spraying. Three weather forecast models stood out with similar results: ICON D2 (German model), AROME (French model), and HARMONIE (Dutch model).

Main results

• Installation of 22 weighing rain gauges.
• 10 preventive maintenance rounds carried out over 3 years.
• 1 automatic quality control system for the Pameseb network.
• 1 scientific article describing the automatic quality control system.
• 1 algorithm for automatic control of weather station locations.
• 1 scientific article describing this algorithm.
• 1 map of station siting quality posted on Agromet.be.
• 1 report comparing commercial stations in Humain from 2022 to 2024.
• 1 report on the use of spatialized weather data to feed agro-meteorological decision support tools.
• 1 report comparing weather forecast models.

Development of two tools for farmers:

→  My weather report: Weekly weather maps sent to farmers (e.g., cumulative precipitation over 7 days).

→  My quality control: Alerts sent to farmers in the event of a suspected problem with their weather station (e.g., clogged rain gauge).

• 1 event to present the results: The Agrometeorology Morning Meeting

Partners

• WalDigiFarm : Arnaud Verlinden, Thomas Servais, Clémence Privé, Sébastien Weykmans
• Royal Meteorological Institute : Michel Journée
• UCLouvain : Patrick Bogaert
• Agence du Numérique : Sandrine Quoibion

Funding

This project was funded by the Walloon Recovery Plan. #WallonieRelance (axe 3, sous-Axe 3.1, projet 142 : « Déployer le Smart farming : le digital au service de la transition »).

Team