|
Habitat mapping is key to meeting land management and conservation objectives, from supporting estimation of natural capital in a landscape to monitoring habitat change over time. Current land management challenges, including conserving rare species and habitats, illustrate a growing need for spatially extensive and rapidly updatable biodiversity information; a need that can best be met through remotely sensed imagery combined with the remarkable data processing potential of machine learning. We assess the potential for optical satellite data to classify complex upland habitat types in the Scottish Highlands, following two UK national habitat classification frameworks. We explore how the differences in spatial and spectral resolution of satellite sensors affects the accuracy of derived habitat maps. Specifically, we contrast the performance of open-source Sentinel-2 data (20 m spatial resolution) against higher spatial-resolution data from the commercial Worldview-2 satellite (0.5 m resolution). We then compare the results produced with these satellite datasets against equivalent results obtained with high-resolution (25 cm) colour airborne photographs, to better inform users on the utility of available optical data before subsequent analysis, e.g. natural capital assessments, in comparable settings. We demonstrate that high-fidelity habitat maps (93% overall accuracy) can be produced using high resolution optical satellite data (Worldview-2). This level of accuracy exceeded that of maps derived from airborne surveys (~75%) and is deemed sufficient to be useful to ecologists in-situ. In contrast, the capacity of Sentinel-2 data was considerably more limited (~50% overall accuracy). This highlights the importance of spatial resolution for characterising habitat mosaics at scale, especially in settings such as upland Scotland where shifts in habitat and species composition of importance to land managers may occur at relatively fine length scales (<10m). Provided high spatial resolution optical data is available, the framework developed is scalable to a national scale, therefore, facilitating effective land management strategies
|
