Assimilation of global positioning system radio occultation measurements into numerical weather forecast systems

2004 marks the centennial of Bjerknes' conceptual solution to the problem of weather forecasting, based on the application of physical and mechanical laws to a set of initial conditions. Today many satellites observe our atmosphere to help provide these initial conditions to numerical weather forecast systems. However, there is still considerable room for improvement when it comes to weather forecasts, namely by better constraining the atmospheric initial conditions through more frequent and more accurate atmospheric observations. The radio occultation (RO) technique which observes the atmosphere from the side (instead of from above, like today's numerous weather satellites) was used for decades to study planetary atmospheres before it was finally used observe our atmosphere. The level of accuracy required in order to make RO competitive with or superior to other techniques already observing the atmosphere was achieved with the advent of the Global Positioning System (CPS) and later with advanced CPS receiver technology. We demonstrate here that useful one-dimensional (1D) vertical atmospheric temperature information can be retrieved from GPS RO refractive index measurements. We develop a first-generation system capable of retrieving that information from real CPS RO refractive index measurements, and assimilating it into a numerical weather forecast system. Resulting five day weather forecasts issued with that system show improved accuracy in the Northern hemisphere and degraded accuracy in the Southern hemisphere. Analyzing data from late CPS RO missions onboard CHAMP and SAC-C satellites, we find that improved receiver technology and processing techniques yield more CPS RO data and of better quality. We develop a two-dimensional (2D) ray-tracing iterative procedure to simulate CPS RO measurements from the three-dimensional (3D) atmospheric representation provided by a numerical weather forecast system. We demonstrate that the 2D ray-tracing results agree better with the CHAMP and SAC-C measurements than the 1D approach used in our first-generation system. In order to render the use of this type of modelling possible under the stringent time constraints of operational numerical weather prediction centers, we propose a new operator called the Fast Atmospheric Refractivity Gradient Operator (FARGO). FARGO is shown to be 250 times faster and nearly as efficient as 2D ray-tracing in fitting real observations. The work presented here will serve to implement a second-generation assimilation system of CPS RO data into numerical weather forecast systems. Also, the approach used in FARGO can be extended to simulate other refraction measurements which suffer from horizontal inhomogeneities in the Earth's lower troposphere.