Living England project, led by Natural England, is a multi-year programme delivering a satellite-derived national habitat layer in support of the Environmental Land Management (ELM) System and the Natural Capital and Ecosystem Assessment (NCEA) Pilot. The project uses a machine learning approach to image classification, developed under the Defra Living Maps project (SD1705 – Kilcoyne et al., 2017). The method first clusters homogeneous areas of habitat into segments, then assigns each segment to a defined list of habitat classes using Random Forest (a machine learning algorithm). The habitat probability map displays modelled likely broad habitat classifications, trained on field surveys and earth observation data from 2021 as well as historic data layers. This map is an output from Phase IV of the Living England project, with future work in Phase V (2022-23) intending to standardise the methodology and Phase VI (2023-24) intending to produce a baseline version using the agreed standardised methods.
The supplementary material “NERR108 Edition 1 Supplementary data: Confusion matrices” (Living_England_PhaseIV_Confusion_Matrices_May2022.xlsx) provides confusion matrices produced from the random forest classification models used to predict the habitat classes across England and to create the Living England Phase IV Habitat Probability Map. These statistics are broken down for each habitat class and biogeographic zone.
The Living England Habitat Map (Phase 4) is available on the Natural England Open Data Geoportal.
Update 07/2022 – The habitat probability map has some known under mapping of urban areas, with major roads, airports, car parks and dockland areas being classified under a number of other habitat types. This mainly affects habitat predictions around urban areas for the following broad habitat types: Broadleaved, Mixed and Yew Woodland; Coastal Sand Dunes; Bare Sand; Dwarf Shrub Heath; Acid, Calcareous and Neutral Grasslands). This is an issue the Living England team are developing solutions for and are looking to improve in the next iteration of the map due for release in 2024.