Methodological developments in data fusion for lake water reflectance from satellite sensors

Zou, Zhaoyuan (2025) Methodological developments in data fusion for lake water reflectance from satellite sensors. PhD thesis, University of Glasgow.

Abstract

Fusing satellite-sensed reflectance data from different sources is of interest to monitor lake water quality, and the satellite sensors have possibly different spatial, temporal and spectral supports. The nonparametric statistical downscaling (NSD) model is an existing state-of-the-art fusion model which can account for a change of spatial and temporal support between two remote sensors [Wilkie et al., 2019]. However, the NSD model is computationally demanding for large datasets and does not allow multivariate responses with an additional spectral dimension. Thus, the aim of this thesis is to improve the computational efficiency of the NSD model and then extend this model to provide an approach that is suitable for a multivariate response to enable the fuse of reflectance data with different spectral and temporal supports from two sensors. The NSD model assumes that the discrete data at each location within a lake from each data source are observations of smooth functions over time and that the coefficients of these smooth functions are modelled as spatially correlated via a covariance matrix. In this thesis, a novel approach proposes using a Gaussian predictive process to approximate the spatial varying coefficients in the NSD model, which requires the inversion of a matrix with smaller dimensions in the Gibbs sampling process and hence reduces the computational time for the parameter estimation. The predictive performance and computational efficiency of the proposed nonparametric statistical downscaling model with Gaussian predictive process (NSD-GPP) are compared to the NSD model through simulation and using satellite reflectance data from Lake Garda. It was found that the NSD-GPP model achieves a similar predictive performance as the NSD model using less computational time. To enable data fusion from the two sensors with a multivariate wavelength dimension, a novel method using the two-dimensional B-spline basis functions was developed where the basis functions were used to represent the reflectance over both time and wavelength at each location, and a different precision parameter was used for each wavelength. Lake Garda is used as an example of interest here, and methods are general for any lake of interest in principle. Overall, it is found that the proposed multivariate NSD-GPP model could be used to make predictions for the unobserved wavelengths and time points within the observed range. It may be beneficial to provide reflectance data at higher temporal and wavelength frequencies, and this model could in principle be extended to consider similar challenges in space.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: H Social Sciences > HA Statistics; Q Science > QA Mathematics
• Colleges/Schools: College of Science and Engineering > School of Mathematics and Statistics
• Supervisor's Name: Haggarty, Dr. Ruth, Miller, Professor Claire and Lee, Professor Duncan
• Date of Award: 2025
• Depositing User: Theses Team
• Unique ID: glathesis:2025-84879
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 10 Feb 2025 10:25
• Last Modified: 10 Feb 2025 10:25
• Thesis DOI: 10.5525/gla.thesis.84879
• URI: https://theses.gla.ac.uk/id/eprint/84879