Issues Associated With The Assimilation of Ground-based GNSS Observations Into Atmospheric Models

Abstract

Data assimilation is a widely used physically-based technique to combine information derived from observations with an estimate (i.e. guess or prediction) of what the environment would look like without the new information. The latter is usually referred to as “the background,” and in numerical weather prediction this is usually derived from a short-range numerical forecast valid at the time of the analysis. The purpose of using data assimilation techniques is to describe the most likely (initial) state of the environment from which an accurate prediction of a future state (forecast) can be made. In contrast to non-physical data interpolation techniques, data assimilation takes into account the estimated uncertainties (systematic and non-systematic errors) in the observations and background estimates in a way honors the underlying physical principles (laws) of the entire system. In the case of GNSS observations, zenith neutral signal delays (ZTD) estimated from GNSS pseudo range and carrier phase observations are combined with all other available observations and a 3-dimensional description of how atmospheric state variables (wind, temperature, pressure, moisture) modify the radio-refractivity of the atmosphere to achieve an optimal initial condition for all analyzed variables. Because numerical weather prediction is essentially an initial value problem, the accuracy of a weather forecast depends heavily on the accuracy of the analysis resulting from the data assimilation process. This is especially true of short range forecasts because the atmosphere is fundamentally a non-linear dynamic (chaotic) system. Chaos (manifested by temporal and spatial changes in state variables that cannot be accurately described without continuous observations at appropriate temporal and spatial scales) is the primary reason why longer range forecasts are usually less accurate than shorter-range predictions. Nonlinearity on the other hand is the primary reason why longer range forecasts depend less on the accuracy of the initial conditions than on the means and standard deviations of the state variables manifested in local to-regional climatologies. With these facts in mind, we will discuss some of the issues associated with the assimilation of ground-based GNSS observations (either ZTD or total column precipitable water vapor (TPW) retrieved from these delays) into numerical weather prediction models. The first issue involves the non-uniqueness of distributing radio-refractivity or PW vertically in the atmosphere. A co