Machine Learning & Earth Observation to detect rainfall induced landslides

This project is part of the ONE Planet DTP. Find out more here: https://research.ncl.ac.uk/one-planet/

This PhD project aims to: 1) fuse Sentinel-1/2 SAR and optical data to ‘see through’ cloud; and 2) apply powerful machine learning technologies that have been highly successful at feature recognition in other contexts to automate landslide detection.

Methods: Optical imagery will be analysed to detect landslides using Object-based Image Analysis (OBIA). SAR coherence and amplitude and polarimetry, will be used to detect changes in ground surface properties due to landslides, whilst conventional and range-split-spectral InSAR and SAR/optical pixel offset tracking will be employed to measure surface movements. Supervised machine learning methods, particularly deep convolutional neural networks (CNNs) and support vector machines (SVMs) will be examined. They will be trained and tested using the existing landslide inventories. Landslide inventories from new events will be utilised to evaluate the performance of the automatic landslide detection system.

Training and Skills: The student will receive training in earth observation techniques, through taught modules (e.g. Remote Sensing, Crustal Deformation and Geohazards) and one to one tuition from the supervisory team including a landslide expert (e.g. Dr David Milledge). In particular the training will cover handling of satellite radar and optical data, feature detection, spatial and temporal functional data analysis, and machine learning. A summer internship has been arranged in DiDi AI Lab, one of the world-leading AI Labs to gain relevant experience.

The potential impact of this project is primarily on landslide disaster preparedness through improved landslide forecasting and early warning to aim risk mitigation and preparednes. However, the tools could be used for disaster response, since an automated detection system could form the basis for a functioning global platform to provide landslide estimates following every extremely weather event.

A quantitative background (e.g. mathematics, computing, engineering, geomatics/remote sensing) and an interest in Earth Sciences & Geohazards.

For more information, please contact Prof Zhenhong Li ([Email Address Removed]).