The fast proliferation of new satellite, aerial and terrestrial remote sensing techniques has provided new technological and scientific opportunities to investigate geohazards during the last decade. Driven by this impressive technological progress together with the development of innovative data treatment and data-driven models, we are now in a good position to upscale and enhance current modelling strategies to forecast landslide initiation. There is a plethora of on real-time data -including in-situ, close-range and satellite remote sensing techniques- that are awaiting to be fully exploited. A new generation of Early Warning Systems able to utilise these data and to couple them with efficient landslide modelling algorithms are actually needed in order to forecast the complex interactions between rainfall and landslide acceleration/failure, with sufficient response time to implement risk mitigation strategies. This interdisciplinary project aims to develop and implement a data driven approach to forecast accelerating landslide movement using real-time big data from satellites and in situ monitoring. In other words, we aim to make a disruption on the current strategies for risk management of landslides affecting infrastructures by laying the foundations of a new generation of Early Warning Systems (EWS). To do so, we will apply advanced data analytic methods to extract knowledge and insights from a rich stream of heterogeneous and structured time-series of both observational and forecasted datasets –e.g. weather conditions and slope deformation measures using satellite remote sensing and in-situ sensors-. The development of a modelling strategy able to predict, with sufficient response time, both the initiation and the acceleration of a sliding mass using forecasted meteorological data has a great potential to revolutionize current risk management strategies through the implementation of a new generation of early warning systems. We will first train and test the ability of a series of algorithms to explain past behaviour using both well-established physically based models and also semi-supervised or unsupervised deep learning (including –but not limited to- Hidden Markov models, recurrent neural networks, autoregressive integrated moving average modelling, Gaussian Processes, etc). Back and forward analysis of tipping point behaviours will be carried out in order to model better extreme and often unexpected events. The emphasis of this study will be on developing real-time algorithms for predicting the future behaviour of hazardous slopes through a nowcasting strategy and on a multi-scale basis.
This 3.5 years EPSRC DTP award will provide tuition fees (£4,500 for 2019/20), tax-free stipend at the UK research council rate (£15,009 for 2019/20), and a research training and support grant of around £5,000