Sentinel-2 Data for Enhancing VHR-Based Mapping Applications
Smith, Geoff
Specto Natura Ltd., UNITED KINGDOM

The Sentinel-2 (S2) series of satellites are primarily designed to address the needs of 'wall-to-wall' continental scale mapping on a 'field-by-field' basis. S2 is classified as a high spatial resolution (HR) system with pixel sizes on the ground of 10, 20 and 60 m depending on the spectral band. The products and services to be supported in terms of land monitoring are similar to the current CORINE Land Cover and national land cover mapping activities with minimum mapping units (MMU) of 25 to 1 ha. As part of the Land Monitoring Core Service the S2 data are already seen as one of the key drivers for the five high spatial resolution layers (HRLs) within the Coperncius / GMES Initial Operations (GIO). Conventionally, more detailed mapping has been the sole domain of the very high spatial resolution (VHR) and aerial photography sensors with pixels sizes on the ground of less than 5 m and down to a few centimetres. However, these sensors often only record relatively limited spectral information and only at irregular intervals reducing the thematic and biophysical information that can be extracted. This also increases the requirement for manual interpretation in production flowlines with consequent increases in costs and inconsistency between operators.

Within the Geoland2 FP7 project the uses of VHR and HR data in the development of land cover products was successfully tested. The approach exploited the spatial detail of the VHR data and the spectral and multi-date capabilities of the HR systems. The SATChMo core mapping service (CMS) implemented a pre-operational flowline which produced generic land cover products for a set of sample sites within an area frame sampling scheme. A VHR image was acquired for each sample site and an object-based image analysis approach was used to produce a generic land cover map with a MMU of 0.25 ha and 10 basic classes (urban, bare, water, snow / ice, agriculture, forest, sparse woody, grassland, other vegetation and unclassified). This product specification is very close to that required by the GIO local component which is targeting biodiversity monitoring in riparian zones. In parallel to SATChMo, the Euroland CMS was developing the pre-cursors of the HRLs to provide basic land surface information such as soil sealing, forest crown cover density, forest type, etc. at a spatial resolution of 20 m with S2-like data.

By combining the 0.25 ha MMU generic land cover product and the 20 m spatial resolution HRLs it was possible to provide an enhanced land surface product and verification results for both input products. The HRL information was summarised on the generic land cover objects and a set of rules applied automatically to give additional thematic information. For instance, the generic urban class was subdivided using the average amount of soil sealing per object. Similarly, the forest crown cover density and type subdivided the generic forest class into coniferous, deciduous and mixed as well as different density classes. More subtle rules were also developed when considering such classes as scrub and processes like scrub encroachment which are important in nature conservation but were difficult to define and map successfully and consistently with categorical land cover mapping approaches. Finally, the use of products from different input data sources which were produced through different flowlines allowed verification information to be provided to identify spatial locations and situations where the products did not appear to match.

The products resulting from this approach provided additional thematic information and a rich attribute set on each object for tailored analysis at the regional and local scale. The verification information provides useful feedback to the service providers of the local and continental level products. These tests represent a valuable demonstration of the links that can be exploited within a future Land Monitoring Core Service and provide examples that can be used as prototypes for downstream services. These approaches will represent a more cost effective, efficient and consistent use of EO data in environmental monitoring. In these cases the S2 data will form an important part of European land monitoring at a broad range of scales either directly or through intermediate products such as the HRLs.