How to think global : Exploring different alternatives for global cropland classification in the framework of the project Sentinel 2 Agriculture

When talking about crop identification, trying to think global is a big challenge as we have to deal not only with the huge diversity of cropping systems around the world, but also with the different agro meteorological conditions and the local heterogeneity of the agricultural practices. Not to mention the difficulty of counting with reliable field data to feed the algorithms. The future Sentinel 2 mission will offer the opportunity to develop agriculture applications with unprecedented properties in terms of resolution (10-20 meter), revisit frequency (5 days) and coverage (global). In this context, ESA has launched the Sentinel2 for Agriculture project which aims at developing open source algorithms and software to process Sentinel-2 data in an operational manner for major worldwide representative agriculture systems. One of the products that will be developed in the framework of the Sentinel -2 Agriculture is a dynamic cropland mask. This mask will consist in a binary “annual cropland - no annual cropland” map produced several times along the season, to serve as a mask for monitoring crop growing conditions along the agricultural season In a first phase of the project, 5 different algorithms for each of the products will be proposed and benchmarked in 12 different sites from Europe, Africa, Asia and South America. Imagery from SPOT Take 5, Rapid Eye and Landsat 8 will be used as substitute of the Sentinel 2. This work will present the methodologies developed ( and the first results) to handle the lack of sufficient and accurate field data that generally is needed to feed the algorithms. These methods include; i) an unsupervised classification based in reflectance and temporal features and ii) study of the change using an iterative trimming algorithm.