Developing a high resolution crop classification for ecologists using Sentinel 2A/2B imagery.

Abstract:

The launch of the Sentinel satellite missions, capable of recording imagery at high spatial and temporal resolution, has widened the scope of application within the field of land change monitoring. Land cover classifications are satellite-derived end products, which can be used to monitor crops or other land cover class types through space and time. For example, the CORINE land cover dataset is a frequently used classification at a European Scale, with a spatial resolution of 100 metres. However, this spatial resolution makes the product unsuitable for small-scale applications such as identification of the land cover in a narrow buffer around a stream as well as margins beside agricultural fields that are typically only a few meters wide. While remotely sensed data, such as CORINE, can supplement in-situ data collected by ecologists, the coarse nature of the classification dataset may introduce spatial errors. We derived a 10 metre crop classification for the state of Rhineland Palatinate in Germany using Sentinel 2A/2B imagery. We aimed for an open and reproducible workflow using the open tools provided by Copernicus, including, Sentinel imagery, the SNAP toolbox and cloud computation platforms. In addition to achieving a finer spatial resolution, we also highlight the importance of using appropriate crop/land type definitions, when working at local scales, to be able to distinguish between occurring crop types more efficiently. Furthermore, we developed an interactive environment for visualising the achieved classification and thereby provide an accessible end product. Future applications also include a comparison with CORINE and other widely used classification schemes.