Application of machine learning to satellite data for classification of the precipitation spatial characteristics

Mariia Kavalerova

Master's thesis

Abstract:

Precipitation analysis of the trends and changes are taking part in climate change understanding. In addition, it plays a vital role in predicting and investigating extreme weather events, such as floods and droughts. Extreme events threaten climate-sensitive economic sectors, including agriculture, forestry, energy, insurance and others. In precipitation analysis, temporal changes and spatial variability are taken into account. Therefore, it is crucial to identify areas with the same and different precipitation patterns to improve precipitation analysis. Only ground-based observations were available and used in the studies for a long time. However, spatial data with high resolution became available, and many researchers attempted to use satellite data in their analysis. Spatial data includes an enormous amount of data compared to ground-based observation; thus, analysis involving the application of the unsupervised machine learning approaches is becoming more popular and is considered more efficient. In this research, the self-organising algorithm merged with hierarchical clustering was used on GPM data to classify precipitation of the Czech Republic spatially. GPM data is stored in the structured compressed "nc4" format; thus, it was pre-processed and converted to the tidy table format. As a result, homogeneous clusters with different precipitation patterns were received. Different grid sizes and numbers of clusters were applied; however, most of the clusters remain the same; this confirms the homogeneity of the regions. Interestingly, the mountains regions and the South Moravia region formed separate and homogeneous clusters. This approach showed a convenient, low computational and time cost method to classify precipitation patterns spatially. One of the main benefits is clear visualisation of the clusters on the map, which can benefit the analysis of the precipitation patterns and improve spatial precipitation variance analysis. Furthermore, spatial classification of the precipitation patterns can improve flood and drought modelling and predictions. However, this study did not include a comprehensive analysis of the frequency and intensity of the precipitations and their relation to the atmospheric circulation, which can be considered future development and an extension of the analysis. In addition, the GPM data quality should be considered while interpreting the results of this study.

Keywords

Self-Organizing Map, clustering, precipitation in Czech Republic, GPM data

Language used: English

Date on which the thesis was submitted / produced: 31. 3. 2022

Thesis defence

• Supervisor: doc. Mgr. Ing. Ioannis Markonis, Ph.D.

Czech University of Life Sciences Prague

Faculty of Environmental Sciences

Master programme / field:
Landscape Engineering / Environmental Modelling