Complex mathematical models are often computationally expensive to run and involve many different input parameters. Therefore, the propagation of errors through these models is often difficult to evaluate. In this paper we will have a closer look at the error propagation through one such model, namely, Neon, an electro-optical Tactical Decision Aid (TDA) for predicting the apparent brightness temperature contrast between a target and its background, which is run operationally by the UK Met Office. Neon consists of four different parts, a land surface model (LSM) for predicting background temperatures, a land/maritime target model (TM), a radiative transfer model (RTM) and a detect and recognize model (D and R). Although, the accuracy of the individual Neon components has been studied before, no overall sensitivity analysis has ever been done for Neon. In this paper we utilize Morris’ Method to study the sensitivity of the Neon prediction system to uncertainties in its input parameters. The key message from this analysis is that Neon is particularly sensitive to uncertainties in the optical properties, in particular the albedo, of the target and background. Thus, this study allows us to focus further development of the system on the elements that contribute the greatest uncertainty.
|