A methodology for fixed observational network design: Theory and application to a simulated global prediction system

A methodology called the retrospective design algorithm (RDA) for fixed observational network design is presented. The RDA makes use of archived ensemble forecasts to estimate the value of various spatial configurations of fixed observational networks. For large dimensional geophysical applications of interest, the RDA can test a variety of fixed networks without having to repeatedly integrate the prediction model equations, unlike observing system simulation experiments (OSSEs). The RDA does not require the adjoint of the prediction model or observation operator. As a proof of concept, the utility of the RDA is tested by making direct comparisons to OSSEs for a simple surface pressure network design problem in a simulated global atmospheric prediction system. A high degree of correspondence between information obtained from OSSEs and the RDA was found for a variety of verification regions and forecast lead times.