Spatial-temporal graph neural networks for weather prediction in South Africa
| dc.contributor.advisor | Moodley, Deshendran | |
| dc.contributor.author | Davidson, Mohamed | |
| dc.date.accessioned | 2026-05-26T07:30:44Z | |
| dc.date.available | 2026-05-26T07:30:44Z | |
| dc.date.issued | 2023 | |
| dc.date.updated | 2026-05-26T07:26:21Z | |
| dc.description.abstract | Spatial-temporal graph neural networks (ST-GNN) have been shown to be highly effective for flow prediction in dynamic systems but are under-explored for weather prediction applications. Additionally, current approaches for evalu- ating ST-GNN models do not take into account the robustness and stability of the trained models. This research compared and evaluated two ST-GNN mod- els, i.e. Graph WaveNet (GWN) and the Low-Rank Weighted Graph Neural Network (WGN), for weather prediction in South Africa. The results of these two ST-GNN models are compared to two basic temporal deep neural network architectures, i.e. the LSTM and the TCN, for temperature prediction across 21 weather stations in South Africa. A novel framework is presented in which to reliably evaluate model robustness and stability for weather prediction. This framework was used to perform rigorous experiments to evaluate the stability and robustness of the ST-GNN models for temperature prediction. The results show that the GWN model outperforms the other models across different predic- tion horizons with an average SMAPE score of 8.30%. Despite the GWN model outperforming the other models on average, the TCN model outperformed both ST-GNN models at particular weather stations. The results indicate that an ensemble approach consisting of ST-GNN models and basic temporal deep neu- ral network architectures would be the most effective approach for temperature prediction. Finally, the learnt adjacency matrices of the two ST-GNNs were analysed and compared to gain insights into the prominent spatial-temporal dependencies between weather stations.. | |
| dc.identifier.apacitation | Davidson, M. (2023). <i>Spatial-temporal graph neural networks for weather prediction in South Africa</i>. (). University of Cape Town ,Faculty of Science ,Department of Computer Science. Retrieved from http://hdl.handle.net/11427/43286 | en_ZA |
| dc.identifier.chicagocitation | Davidson, Mohamed. <i>"Spatial-temporal graph neural networks for weather prediction in South Africa."</i> ., University of Cape Town ,Faculty of Science ,Department of Computer Science, 2023. http://hdl.handle.net/11427/43286 | en_ZA |
| dc.identifier.citation | Davidson, M. 2023. Spatial-temporal graph neural networks for weather prediction in South Africa. . University of Cape Town ,Faculty of Science ,Department of Computer Science. http://hdl.handle.net/11427/43286 | en_ZA |
| dc.identifier.ris | TY - Thesis / Dissertation AU - Davidson, Mohamed AB - Spatial-temporal graph neural networks (ST-GNN) have been shown to be highly effective for flow prediction in dynamic systems but are under-explored for weather prediction applications. Additionally, current approaches for evalu- ating ST-GNN models do not take into account the robustness and stability of the trained models. This research compared and evaluated two ST-GNN mod- els, i.e. Graph WaveNet (GWN) and the Low-Rank Weighted Graph Neural Network (WGN), for weather prediction in South Africa. The results of these two ST-GNN models are compared to two basic temporal deep neural network architectures, i.e. the LSTM and the TCN, for temperature prediction across 21 weather stations in South Africa. A novel framework is presented in which to reliably evaluate model robustness and stability for weather prediction. This framework was used to perform rigorous experiments to evaluate the stability and robustness of the ST-GNN models for temperature prediction. The results show that the GWN model outperforms the other models across different predic- tion horizons with an average SMAPE score of 8.30%. Despite the GWN model outperforming the other models on average, the TCN model outperformed both ST-GNN models at particular weather stations. The results indicate that an ensemble approach consisting of ST-GNN models and basic temporal deep neu- ral network architectures would be the most effective approach for temperature prediction. Finally, the learnt adjacency matrices of the two ST-GNNs were analysed and compared to gain insights into the prominent spatial-temporal dependencies between weather stations.. DA - 2023 DB - OpenUCT DP - University of Cape Town KW - networks KW - South Africa LK - https://open.uct.ac.za PB - University of Cape Town PY - 2023 T1 - Spatial-temporal graph neural networks for weather prediction in South Africa TI - Spatial-temporal graph neural networks for weather prediction in South Africa UR - http://hdl.handle.net/11427/43286 ER - | en_ZA |
| dc.identifier.uri | http://hdl.handle.net/11427/43286 | |
| dc.identifier.vancouvercitation | Davidson M. Spatial-temporal graph neural networks for weather prediction in South Africa. []. University of Cape Town ,Faculty of Science ,Department of Computer Science, 2023 [cited yyyy month dd]. Available from: http://hdl.handle.net/11427/43286 | en_ZA |
| dc.language.iso | en | |
| dc.language.rfc3066 | eng | |
| dc.publisher.department | Department of Computer Science | |
| dc.publisher.faculty | Faculty of Science | |
| dc.publisher.institution | University of Cape Town | |
| dc.subject | networks | |
| dc.subject | South Africa | |
| dc.title | Spatial-temporal graph neural networks for weather prediction in South Africa | |
| dc.type | Thesis / Dissertation | |
| dc.type.qualificationlevel | Masters | |
| dc.type.qualificationlevel | Masters |