Development of spatio-temporal data fusion frameworks for point and gridded soil moisture data

Zheng, Weiyue (2026) Development of spatio-temporal data fusion frameworks for point and gridded soil moisture data. PhD thesis, University of Glasgow.

Abstract

Monitoring soil moisture can play an important role in helping to inform researchers, regulators, and landowners about the available water content of the soil for agriculture and vegetation. However, the capacity to observe soil moisture is constrained by practical and financial limitations, making it challenging to observe continuously across space and time. We can only monitor soil moisture at a finite number of spatial locations and time points. One of the most accurate methods for measuring soil moisture is to use in-situ sensors. However, the high cost of deploying these sensors extensively means that soil moisture data tends to be collected from a sparse network of monitoring points.

Given the limited in-situ sensor data, it becomes essential to explore the benefits of utilising other data sources, such as satellite data, by developing and using data fusion techniques. Data fusion allows for the integration of different data sources, enhancing the ability to make informed decisions and understand environmental phenomena with more precision, despite the limited direct monitoring of soil moisture.

The research question is motivated by the in-situ soil-moisture data provided by SEPA in Elliot Water and the satellite images provided by Copernicus. It is necessary to develop a data-fusion method for point data and gridded data, so that the accuracy of the in-situ data can be combined with spatial and temporal information from satellite data to generate a fine-resolution map with uncertainty quantification.

This thesis introduces three INLA-based data-fusion methods in Chapter 3, 4, and 5, which include a spatio-temporal regression with misaligned covariates, a spatial data fusion method, and a spatio-temporal data fusion method. A comprehensive simulation study varies sensor density, grid resolution, percentages of missing grid data, and temporal window length k. Across scenarios, joint fusion consistently outperforms point-only and grid-only baselines in RMSE. This thesis also introduces an XGBoost-based constrained ensemble method with conformal prediction in Chapter 6, developed to merge in-situ point and satellite gridded data under different spatio-temporal supports.

This thesis presents the background, motivation, model development, and application of the novel data fusion methods, addressing the gap in the literature by accounting for spatio-temporal change-of-support problems. Results are presented throughout to demonstrate the use of the data fusion model in soil moisture data.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: H Social Sciences > HA Statistics; Q Science > QA Mathematics; S Agriculture > S Agriculture (General)
• Colleges/Schools: College of Science and Engineering > School of Mathematics and Statistics
• Supervisor's Name: Scott, Professor Marian, Miller, Professor Claire and Elliott, Dr. Andrew
• Date of Award: 2026
• Depositing User: Theses Team
• Unique ID: glathesis:2026-85693
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 16 Jan 2026 11:20
• Last Modified: 16 Jan 2026 11:24
• Thesis DOI: 10.5525/gla.thesis.85693
• URI: https://theses.gla.ac.uk/id/eprint/85693