Predictability of LoRaWAN Link Quality based on Weather Data: Insights from a Long-Term Study

WSNs are often deployed outdoors and thus exposed to a variety of different weather conditions. Due to the properties of radio waves, their propagation can be negatively impacted by the prevailing weather conditions. The sender and receiver hardware, including the antennas, can also be affected, potentially leading to variations in signal quality and link degradation. In order to assess the impact of individual weather components on the quality of WSN links, we analyze a large scale LoRaWAN deployment consisting of multiple nodes and gateways over a period of several months. The WSN is used for environmental monitoring, which means that the weather data is directly measured by the nodes and available in a high spatial and temporal resolution. Our results indicate significant dependencies between different weather components like precipitation, temperature, humidity but also time of day and the link quality as measured by RSSI and SNR. We further find that every link is affected to a different degree and has its own individual characteristic. Our analysis also shows significant correlations of the link quality within clusters of nodes that are deployed geographically close to each other. Finally, we evaluate the predictability of the link quality based on the prevailing weather conditions using different supervised learning models. Our results show that the trend of the link quality is indeed predictable, especially periodic patterns and trend changes can be recognized by the models and predicted within the test sets. The best model performance as measured by RMSE ranges from 1.79dBm to 6.39dBm for RSSI and 1.15 dB to 6.06 dB for SNR.