Next generation landscape mapping: machine learning and big data methods for exploiting earth observation data

This is an exciting opportunity to explore machine learning and big data methods, in combination with earth observation data, to extract information on land cover, landscape features and habitat condition. It will use a range of remote sensing data sets, including LIDAR, aerial photography, high resolution satellite data and satellite time-series.

Land cover is a key environmental variable and is typically classified from satellite data, which has a relatively coarse spatial scale (> 20m). However, land cover change and key habitats often occur at spatial scales below this. Large data sets of aerial photography, LiDAR and high resolution satellite data are increasingly being collected, but efficiently extracting useful information from them requires development of new methods.

Over recent years the deep learning and machine learning communities have made rapid progress in developing methods for classifying aerial photography, however, this work has typically focused on the requirements of the military and disaster response.

This project will explore the advances made by the machine learning community to develop methods for classifying remote sensing data for environmental purposes, as well as for quantifying measures of habitat condition. It will explore novel large-scale applications including the combined use of LiDAR and aerial photography, and the differences in implementation required when applying the methods to different types, combinations and scales of earth observation data. The PhD will be structured around a series of case studies exploring different applications and spatial scales and will be underpinned by field data sets held by CEH. The aim of the project is to demonstrate the potential of machine learning methods and to gain an understanding of when (and how) they are most usefully applied.

Successful development of these methods will transform our ability to monitor the natural environment remotely by revolutionising the type and speed of products we produce.

This PhD is interdisciplinary in nature and as such would such would suit applicants from a wide range of numerate, scientific backgrounds, including (but not limited to) candidates with degrees in Environmental Science, Geography, Ecology, Biological Sciences, Physics or Engineering. MSc’s in a relevant subject such as Remote Sensing, Environmental Modelling or Data Science would be an advantage, but an MSc, whilst desirable, is not essential.

Enquiries should be addressed to Dr Clare Rowland, [Email Address Removed].