An Adaptable and Open-source Approach to Merging Satellite and Gauge Data over Australia

The estimation of rainfall through gridded spatial analyses is important for water resource management and water-related disaster risk mitigation, as well as for being an input into scientific models. The current rainfall analysis used by the Bureau of Meteorology (BOM), Australian Gridded Climate Dataset (AGCD) rainfall, relies purely on gauge data, leading to high uncertainty over gauge-sparse regions such as interior parts of the country. Utilising additional sources of rainfall information to form blended datasets would be highly valuable but the advancement of operational rainfall analyses has been hindered by a lack of research in the development and comparison of blended datasets over Australia, and in the verification of such datasets over gauge-sparse areas. This thesis addresses this research gap by developing multiple correction and merging techniques that are tailored to the Australian context, and subsequently comparing them against each other to determine which is the optimal method, and whether they are still effective over gauge-sparse areas. This thesis aims to produce the optimal blended rainfall analysis for operational use in Australia, providing users globally with an adaptable technique for assimilating in-situ and gridded data. All methods investigated demonstrated the capability to generate a blended analysis that maintains similar performance to AGCD over gauge-dense regions but with a notably more realistic representation of rainfall and generally improved performance over gauge-sparse regions. In particular, adapting the current algorithm used for creating AGCD, Optimal Interpolation (OI) also known as Statistical Interpolation (SI), to incorporate a satellite estimate as the background field was appealing for operational usage. This was because of its ability to be adapted for different datasets and variables, its relatively low computational requirements, and its removal of the need for a preliminary correction. An open-source Python implementation of the algorithm in two dimensions, which offers adaptability to other domains and datasets, was developed. The research completed, along with the provision of an open-source and adaptable algorithm for creating a gridded analysis, is an important contribution to the knowledge of the value of blended datasets, offering a strong motivation to the use of a blended dataset over Australia for operational purposes. The adaptable and open-source nature also ensures applicability over other regions and for other geospatial variables.