Detection of moving part malfunctions in paired ground radiometers using deep learning

The USDA UV-B Monitoring and Research Program utilizes a network of ground-based solar radiometers to track UV and VIS radiation. These sensors are prone to malfunctions impacting data accuracy. This work presents a novel two-step automated error detection process. First, a deep learning model, trained on extensive error-free UV and VIS data, identifies deviations from expected spectral and temporal relationships. Second, a specialized model, trained on manually verified malfunction data, analyzes these deviations to diagnose specific sensor issues, like moving part failures. This automated approach improves data accuracy, enabling more reliable UV radiation monitoring and research.