Abstract
The challenge set by the Peak District team involves making a robust and scalable algorithm to identify changes in the land cover, including changes in land use and habitats, using aerial photography images from two time periods roughly 10 years apart. This model should ideally be able to incorporate new data every 5 years and potentially be portable to application by other national parks in England.
The data consists of 12.5 cm ground resolution, ortho-rectified, geo-referenced true colour (RGB) files and 50 cm ground resolution,
ortho-rectified, geo-referenced infrared (IR) files. The files represent 891 square-km areas chosen as a training set, representing a total area of
1,439 square-km of the park. In addition, there is secondary data, including files for a 2 m ground resolution Digital Surface Model and a 5
m ground resolution Digital Terrain Model. The primary objectives for this challenge are:
• Develop unsupervised AI/ML algorithms for detecting change across The Peak District National Park that can be used to monitor changes
in land use in future surveys and for other National parks.
• Produce change maps that detail regions where land use has been altered between 2010 and 2020.
• Identify the degree of changes in areas of conservation interest (i.e. peat restoration, deforestation, growth of bracken habitats and disappearance of dry stone walls).
An unsupervised approach that classifies the changes in land cover features between the two-time points is superior to applying the same supervised model on the two-time points and comparing the results, as erroneous predictions in the two models can compound into misleading change predictions. Re-labelling a new dataset is extremely time-consuming and represents a huge challenge. Differences in land use will be impacted by the seasonal variation between
the two acquisition dates. In addition, the impact of different acquisition times (which will elongate the shadows of various features) has not been considered.
Citation information
DOI: 10.5281/zenodo.10090623
Additional information
Contributers:
- Antonia Marcu is a PhD student at University of Southampton
- Mayank Sharma is an Associate National Project Officer in Data Science at UNESCO MGIEP.
- Alex Milne is a 3rd year PhD student studying at Swansea University.
- Jennifer Stiens (’Jen’) is a 3rd year LIDo/Bloomsbury fellowship PhD student at Birkbeck University of London
- Alicja Polanska is a PhD student in astroinformatics at University College London’s Data Intensive Science CDT.
- Stephen Town is a senior research associate in the Faculty of Brain Sciences at University College London.
- Nurul Abedin is a research student in the Aeronautics and Aerospace research centre at City, University of London.
- Ben O’Driscoll is a Data Visualisation Developer based at Plymouth Marine Laboratory (PML)
- Wenlan Zhang is a PhD student in advanced spatial analysis at the Centre for Advanced Spatial Analysis, UCL.
- Sokipriala Jonah is a PhD researcher at the Centre for computational science and mathematical modelling Coventry.
- Miguel Espinosa is a PhD researcher at the University of Edinburgh SENSE CDT.