A new dynamic approach for statistical optimisation of GNSS radio occultation bending angles

Li, Y 2013, A new dynamic approach for statistical optimisation of GNSS radio occultation bending angles, Doctor of Philosophy (PhD), Mathematical and Geospatial Sciences, RMIT University.


Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Li.pdf Thesis Click to show the corresponding preview/stream application/pdf;... 5.05MB
Title A new dynamic approach for statistical optimisation of GNSS radio occultation bending angles
Author(s) Li, Y
Year 2013
Abstract Climate change has become a serious issue for our society. It is of great importance to accurately monitor climate change and provide reliable information to the society so that proper actions can be taken to alleviate the significant change of climate. Global Navigation Satellite Systems (GNSS) based radio occultation (RO) is a new satellite remote sensing technique that can provide high vertical resolution, long-term stable and global coverage atmospheric profiles of the Earth’s atmosphere. However, the quality of the retrieved atmospheric profiles decreases above about 30 km due to a low signal-to-noise ratio of GNSS signals at these high altitudes, since errors in bending angle profiles are propagated to refractivity profiles through an Abel integral and subsequently propagated to other atmospheric profiles through the hydrostatic integral. It is therefore important to carefully initialise the bending angles at high altitudes to minimise these error propagation effects and thereby optimise the climate monitoring utility of the retrieved profiles.

Statistical optimisation is a commonly used method for this purpose. This method combines the observed bending angle profile and background bending angle profile based on their error covariance matrices to determine “optimised” bending angle profile. The focus of this thesis is to investigate an advanced statistical optimisation algorithm, which dynamically estimates both background and observation error covariance matrices, for the best determination of RO optimised bending angle profile. In this new algorithm, background bending angle profiles and their associated error covariance matrices are estimated using bending angles from multiple days of the European Centre for Medium-range Weather Forecasts (ECMWF) short-term (24h) forecast and analysis fields as well as the averaged observed bending angle. The background error matrices are constructed with geographically varying background error estimates on a daily-updated basis. The observation error covariance matrices are estimated using multiple days of RO data with geographically varying observation errors for an occultation event. The most distinctive advantage of the new algorithm is that both background and observation error covariance matrices are realistically estimated using large ensemble of climatological and observed data, while existing algorithms use crude formulations to estimate both error matrices.

The new algorithm developed is evaluated against the algorithm used by the Wegener Center Occultation Processing System version 5.4 (OPSv5.4) by calculating statistical errors of retrieved atmospheric profiles relative to their reference profiles. Since the background errors at different heights are highly correlated and their covariance matrix is critical for the resulting optimised bending angles, the dynamically estimated background error covariance matrix is first used in statistical optimisation to retrieve atmospheric profiles from simulated MetOp as well as observed CHAMP and COSMIC RO events on single days. The dynamically estimated observation error covariance matrix is then used in the statistical optimisation together with the estimated background error covariance matrix to retrieve atmospheric profiles using the same test data.

It can be concluded from the evaluation that if the estimated background error covariance matrix is solely used for the statistical optimisation, it can significantly reduce random errors and generate less or similar residual systematic errors (biases) in the optimised bending angles. The subsequent refractivity profiles and atmospheric (dry temperature) profiles retrieved are benefitted from the improved error characteristics of bending angles. If both observation and background error covariance matrices estimated from the new approach are used, the standard deviations of the optimised bending angles are only further reduced for simulated MetOp data, while for the observed CHAMP and COSMIC data, large random errors of bending angles are found at higher altitudes (e.g. > 50 km). This is likely to be that the observation errors are underestimated at high altitudes, where bending angles are largely affected by ionospheric effects and observation errors, and more weights are given to the noisy observed bending angles in the estimation of the optimised bending angles. Errors in CHAMP and COSMIC observed bending angles are further transferred downwards to their subsequently retrieved refractivity and dry temperature profiles, the quality of which is also degraded.

The effects of the estimated background and observation error correlations on the atmospheric retrievals are investigated using simulated MetOp data. It is found that these realistically estimated correlations alone can reduce the random errors of the optimised bending angles significantly and improve the quality of the subsequent refractivities and temperatures. The performance of the new approach that uses only the new background matrix in the statistical optimisation on monthly occultation data is evaluated. The results show that the monthly errors are similar to those from single days, but in a smoother manner.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Mathematical and Geospatial Sciences
Keyword(s) Climate monitoring
GNSS radio occultation
bending angles
dynamic
statistical optimisation
Versions
Version Filter Type
Access Statistics: 323 Abstract Views, 430 File Downloads  -  Detailed Statistics
Created: Mon, 16 Dec 2013, 09:41:22 EST by Brett Fenton
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us