12.12.23 - AI for Environmental Monitoring: The Future of Land Cover Maps

Land cover maps are essential tools for measuring and monitoring the state of natural landscapes. It enable for example to measure progress against nature recovery targets set by the UN’s Sustainability Goals. 

A new article has been published by the University of Oxford, Cranfield University and the Peak District National Park Authority : They have developed a multi-stage semantic segmentation approach that uses Convolutional Neural Networks (#CNN) to classify Land cover across the Peak District National Park (1439 km2) in the UK.
 
Land cover was predicted at high resolution (using classify RGB aerial photography obtained at 12.5 cm ground resolution) and it enables the identification of small habitat patches such as individual trees, heather patches and scrub. The detail results including the accuracy achieved for the different classes is available here : https://lnkd.in/eefsuxRi
 
Since land cover and land usage map creation is a very resource and time consuming task (for example, to produce area statistic manually on the scale of Switzerland, it is a 7 years cycle), every effort to leverage deep learning methods is driving efficiency and enable more up-to-date information. Taking advantage of all the research effort performed in that field will help improve the accuracy since it is a very challenging problem as many different vegetation classes must be identified. On top of the biodiversity monitoring, such data can enable better fire risk modelling and climate change vulnerability assessments. 

We will continue to monitor and share on our Linkedin page published results as well as presenting our own work in that field.

#remotesensing #biodiveritymonitoring #landcover