Research on the fault tolerance in retrieving atmospheric refractivity by microwave radiometer data

Ground-based microwave radiometer has proven to be a powerful tool to detect the atmosphere. The common algorithms are linear regression, neural network, and relevance vector machine. A fault tolerance evaluation has been considered on the above three algorithms in retrieving atmospheric refractivity. A simulation experiment reveals that the fault tolerance of relevance vector machine is very good, while the other two are worse. To improve the fault tolerance of linear regression and neural network, we can add measurement error to the brightness temperature when training.