Mapping dominant genus/species types in natural and seminatural landscapes across Wales through application of Sentinel-2 time-series data

The unprecedented loss of habitats and biodiversity in the UK has been particularly evident during the past few decades and strongly emphasises the need for monitoring habitat states and dynamics on a more regular and consistent basis. Timely availability of free, high resolution Sentinel optical and radar data offers an opportunity to develop capacity for monitoring and informing natural resource management and interventions. This study demonstrates a workflow to utilise the full potential of Sentinel-2 datasets through a systematic, scalable approach for routinely mapping dominant vegetation genus/species in the natural and semi-natural habitats across the landscape. The present study is part of the Living Wales project which is developing a landcover monitoring framework for Wales with operational capabilities. The project utilizes the Food and Agriculture Organisation Land Cover Classification System (LCCS) as the base taxonomy. The LCCS provides capability to integrate Essential Environmental Descriptors (EEDs) of the environment retrieved from Earth observation (EO) data to generate the land cover maps. A precursor classification step first involved identification and separation of natural/semi-natural vegetated and cultivated areas across Wales into woody (trees and shrubs) and herbaceous lifeforms. Within the natural/semi-natural lifeform category, spectrally distinct genus or species were separated first, followed by further segregation of groups with similar spectral response. A machine learning algorithm was optimized by identifying appropriate spectral variables and time windows for the stepwise separations. This study generates most recent dominant species/genus map for Wales with validation and pixelwise probability statistics. The map represents a composite period of four years (2015-19), with capacity for routinely update (e.g., for 2016-20), thereby ensuring temporal consistency. The method demonstrates a systematic EO-based approach for genus/species mapping by efficiently utilising dense time-series information from ~10 m resolution Sentinel-2 data.