CEEDS Seminar: Machine Learning for the Natural Environment

Latest blog articles


On 14th October, we held a CEEDS seminar on the topic of "Machine Learning for the Natural Environment", hosted by Pete Henrys (UKCEH). This seminar was attended by over 170 people – a fantastic and amazing attendance across Lancaster University and UKCEH. It helps build the strong links between LU and UKCEH, but also the links across UKCEH science areas and different sites.

The seminar started with a quick update on CEEDS and housekeeping. Chris Nemeth then kicked off the seminar with a whistle-stop tour of Machine Learning, which laid the foundation for the subsequent talks. Chris introduced the three main categories of Machine Learning; namely Supervised, Unsupervised and Reinforcement Learning, with examples of how each of these approaches can be applied to the field of Environmental Science. Next up was Tom August, who spoke on AI Naturalists in the digital undergrowth. Tom collected high-quality georeferenced images from social media platforms such as flicker, and fed them into a deep Plantnet to differentiate plant diversity. Such a machine learning approach faced several challenges including the spatially biased data and the suboptimal images used for training purposes, where data exploration and spatial uncertainty need to be considered properly. Thereafter, Ce Zhang spoke about deep learning in environmental data science. Ce discussed the state-of-the-art deep learning methods for image classification in agriculture, land use and ecology, amongst many other environmental applications. The latest research in deep learning demonstrated great potential and transformative impact on environmental science and data science. Finally, Clare Rowland talked about using random forest and Earth observation (EO) data for a range of applications. Clare summarised the new product CEH Land Cover Map and EO data sets derived from Random Forest classification and regression. The success of the Random Forest algorithm depends on the quality of EO data and the spatial distribution of training sample sets. These talks were all superb and mutually supportive, and brilliantly chaired by Pete Henrys. The seminar marked the start of a truly interesting and vibrant discussion within CEEDS around the application of machine learning for the natural environment; a topic that will be further developed in the coming months, through a dedicated CEEDS workshop that will aim to build and sustain collaborative research in this space. Opportunities to be actively involved in this process will be announced soon, so please make sure that you are registered as a CEEDS member to be notified of future events. 

The next CEEDS seminar will be on 4th November 15:00-16:30 on “Using remote observations in biodiversity research.”

Ce Zhang

Member (Staff)
LU - Lancaster Environment Centre