South-east Asia experiences some of the world’s most severe convective storms, causing flooding and landslides which endanger human life, agriculture and infrastructure. There is, therefore, a strong socio-economic need to improve our understanding of the occurrence of such storms and their underlying physical mechanisms, to aid forecasters. A recent effort to track mesoscale convective systems (MCSs) in geostationary satellite data has provided a 5-year data set of MCSs over the entire south-east Asia region (Fig. 1, left). The PhD candidate will update this data set with recent satellite observations and use machine learning techniques to discover different types of MCSs based on properties such as their geometry and lifetime. A statistical survey of these MCS types will reveal which MCSs are most associated with high-impact weather. The student will further investigate machine learning techniques to reveal which satellite-observed features of an MCS are of greatest importance in determining its type and its contribution to high-impact weather such as heavy precipitation. Convection-permitting simulations will be used alongside observations to investigate the underlying dynamics of each type of MCS, to determine whether they behave according to existing MCS theories.
The project addresses the following research questions:
- What morphology of MCSs can be found over south-east Asia?
- What are the typical storm-scale and large-scale conditions of the different morphologies of MCSs?
- Which morphologies are most common and where and when do they occur?
- Which morphologies of storms are associated with high-impact weather?
- How do the large-scale flow conditions modulate the storm characteristics?
- How are the different morphologies of storms represented in state-of-the-art convection permitting MetUM simulations?
- How do the different morphologies of storms match up with existing theories on the dynamics of MCSs?
- How can the information gained in this project aid local forecasters? Can the new insights improve nowcasting?
The main observational data sets used for the project are brightness temperature from the Himawari satellite and precipitation from the Global Precipitation Measurement (GPM) mission. Convection-permitting and global MetUM simulations at various grid spacing will also be used to study the storm dynamics.
Specifically, the project will:
- Expand the existing data set of tracked storms using more recent data.
- Develop storm-type classification by application of unsupervised machine learning techniques (clustering) on measured morphological properties of the tracked storm data set.
- Train the convolutional neural network to predict storm-type classification and identify key morphologies using ‘layerwise relevance propagation’ on the trained model.
- Characterise different groups of MCSs according to their morphologies in a statistical sense.
- Analyse the evolution of the storms in the different morphology-categories through their lifetime.
- Redo the same analysis for storms in 4.4-km convection permitting MetUM simulations.
- Investigate if the morphologies of the MCSs from the observations and the model are the same or how and why they might differ.
- Investigate selected high-impact weather events in detail.
- Investigate the underlying dynamics of the different morphologies of MCSs in both datasets, with reference to existing theories (e.g., Rotunno et al., 1988, and Moncrieff and Liu, 1999).
- Investigate ways in which the results can be utilised by forecasters in Southeast Asia.
The Met Office CASE award will allow the student access to a suite of state-of-the-art model products and computing facilities as well as staff expertise in remote sensing and in-situ observations, modelling and understanding of atmospheric processes such as convection. The CASE award will also provide additional funding. The student will be based and registered at the University of Leeds.
All SENSE students will receive extensive training on satellite data and AI/Machine Learning, as well as attending a field course on drones, and residential courses hosted by the Satellite Applications Catapult (Harwell), and ESA (Rome). All students will experience extensive training on professional skills, including spending 3 months on an industry placement. See http://www.eo-cdt.org