Project rationale and aims
The identification of plant individuals, their health and their dynamics is essential for environmental monitoring and management, whether this is in the context of plant population, habitat condition or ecosystem services assessment, invasive species or weed control action, or climate or disturbance impact assessment. While, currently, the most common approach is intensive field survey, the combined developments in unmanned aerial vehicles (UAV) and machine learning has opened up opportunities for automation.
This project is about developing an UAV-machine learning based approach that will help map and monitor the health of frailejones, a plant of the Páramos of Colombia, Ecuador and Venezuela. Páramos are highly diverse mountain ecosystems and are found above the timberline in the Northern Andes. Frailejones are iconic plants due to their unique shape and beauty, and many of which are endangered species. They typically have a stemmed rosette of large, coriaceous leaves, that gives them the appearance of palms. Their morphology is very diverse, with plant sizes ranging from fist-sized to more than 15 m tall.
The Páramos and their vegetation are under pressure from climate and environmental changes. Recently, Frailejones have been affected by moth caterpillars, beetle larvae and fungi, and have started to die in large numbers. It is yet unclear why this is happening although an increase in average temperatures is believed to be the main cause. Finding an effective approach to mapping and monitoring frailejones’ health will be key to help evaluate current hypotheses.
Traditional remote sensing approaches are unsuitable for identifying and mapping frailejones because individual plants can be very small laterally relative to their height and be both clustered and highly dispersed. However, recent research using machine learning by the supervisors has shown great promise for this type of problem. This PhD will develop a novel deep learning framework for UAV-derived imagery and 3D point clouds to map individual frailejones and their condition (alive, diseased, dead) within a large Páramo complex in Colombia. This deep learning framework will then be extended to include temperature, precipitation and land use as covariates to explore and identify the possible drivers of the frailejones’ decline, thus, resolving the currently unanswered but important and urgent question of why the frailejones are declining rapidly.
Programme of training
The student will be part of the large and vibrant Lancaster Environment Centre by becoming a member of the Geospatial Data Science research group and will be affiliated with the Centre of Excellence in Environmental Data Science (CEEDS; https://ceeds.ac.uk/) and the Data Science Institute. The student will benefit from the research training programmes offered by the Faculty of Science and Technology at Lancaster University, the skills training and seminars offered by CEEDS and DSI and the interdisciplinary supervision across LEC, DSI and UKCEH.
Eligibility
We are seeking applications from graduates with a good (i.e., 1st class or 2.1) Undergraduate degree, and preferably also a Master’s degree, in a Machine Learning-related subject. You should have a strong background in computer science, mathematics, geography/environmental science with strong quantitative (e.g., programming) skills. You must have demonstrable potential for creative, high-quality PhD research.
Continue with Facebook
