Skip to Main content Skip to Navigation
New interface
Theses

Développement de nouvelles méthodes d’homogénéisation des données atmosphériques GNSS. Application à l’étude de la variabilité climatique.

Abstract : Homogenization is an important and crucial step to improve the use of observational data for climate analysis. This work is motivated by the analysis of long GNSS Integrated Water Vapor (IWV) data which have not yet been used in this context. These series are affected by inhomogeneities linked to changes in the instrumentation, in the environment, and in the data processing procedure. Due to the natural variability of the series we actually work on the time series of differences, using ERA-Interim reanalysis as reference for the climate signal. A base assumption is that the differences contain only the signature of the abrupt changes from the GNSS series which can be detected by means of a segmentation algorithm. Careful analysis of the segmentation results allows to sort the cases when this assumption is actually not true. The main contribution of this thesis was the development a novel segmentation method dedicated to detecting changes in the mean of the GNSS-ERA-Interim IWV difference series. This segmentation model integrates a periodic bias and a heterogeneous, monthly varying, variance to properly fit the characteristics of the series. The method consists of first estimating the variance using a robust estimator and then estimating the segmentation parameters (the positions of the change-points, the means of the segments) and the periodic bias model in a sequential way. The segmentation parameters and the periodic bias model are estimated iteratively for a fixed number of change-points. The inference is achieved by the classical maximum likelihood procedure using the dynamic programming algorithm for the estimation of the segmentation parameters which provides the exact solution in a reasonable amount of time. The procedure is repeated for all the numbers of change-points tested between 0 and a maximum (about 30). Finally, the optimal number of change-points is chosen using a penalized model selection strategy. Several criteria are tested. The method is implemented in the R GNSSseg package available on CRAN. The performance of the proposed method was evaluated by numerical simulations. An application for a real dataset of 120 global GNSS stations in the global IGS network is presented for the period from January 1995 to December 2010. Inspection of the results reveals that the detected change-points contain a fraction (~ 20 %) of outliers which are characterized by double detections with two large offsets, generally of opposite signs, close together, e.g. a few tens of days apart. In order to detect and eliminate the outliers a screening method was developed. The final set of change-points is validated with respect to GNSS metadata which contain information on equipment changes that occurred at the stations. The percentage of validation remains moderate at the level of 20 % despite all the changes are statistically significant. Some of the change-points may actually be due to the reference series (ERA-Interim). Finally, the segmentation information (dates of the change-points) is included in a linear regression algorithm which is used to estimate the GNSS IWV trends. The estimated trends are tested for significance and compared to the ERA-Interim trends. Higher spatial consistency in the GNSS trends and improved consistency is found after homogenisation with ERA-Interim in regions where the reanalysis is known to perform well. [...]
Complete list of metadata

https://theses.hal.science/tel-03771164
Contributor : ABES STAR :  Contact
Submitted on : Wednesday, September 7, 2022 - 10:21:37 AM
Last modification on : Tuesday, December 6, 2022 - 11:10:07 AM

File

QUARELLO_Annarosa_2020.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-03771164, version 2

Citation

Annarosa Quarello. Développement de nouvelles méthodes d’homogénéisation des données atmosphériques GNSS. Application à l’étude de la variabilité climatique.. Sciences de la Terre. Sorbonne Université, 2020. Français. ⟨NNT : 2020SORUS457⟩. ⟨tel-03771164v2⟩

Share

Metrics

Record views

96

Files downloads

11