Automatic Classification of Farmer’s Weather Station Siting Based on Geodata

More and more farmers are equipped with private weather stations which transmit local weather data in real-time. However, those weather stations are sometimes installed near obstacles such as farm buildings or trees, which influence the measurements. Those biased data have consequences when included in a wider weather monitoring network or in decision support systems, but the experts’ managing networks of hundreds of stations cannot afford to visit all the stations to check the siting. The goal of this study was to develop an algorithm to automatically assess the siting quality of weather stations based on their geographic coordinates. The algorithm relied on existing public high-resolution geodata. A land cover map was used to identify heat sources such as buildings or artificial surfaces, which can bias temperature measurements. Digital terrain and surface models were used to identify obstacles located too close to the station compared to their height. Ultimately, one of the following siting classes was assigned to each station: excellent, good, poor or bad. The siting class was computed for 30 stations from a public network. In 73.3% of cases, the siting class determined by the algorithm aligned with the field experts’ assessment. The algorithm never diverged from the experts when they announced a class “excellent” or “bad”. The method was then used to assess the siting of 108 farmers’ stations. More than 65% of these stations had poor or a bad siting. This proves the importance of assessing the station siting.