A knowledge steered stratified classification approach for ERS SAR data used in an agricultural area estimation methodology

In the course of two pilot projects dedicated to the preoperational use of ERS-1 and ERS-2 SAR data in its rapid area estimation program for agricultural crops, the JRC´s Agricultural Information Systems´ MARS-STAT project team has gained valuable experience in both the processing and interpretation of SAR imagery. The use of SAR is considered in conjunction with the existing operational program that uses high-resolution optical imagery only. Furthermore, the MARS archive contains a number of important ancillary data sets for 60 sites in the European Union for use in knowledge based interpretation. This archive includes a soil data base, precision digital elevation models, agro-meteorological and agronomic data bases and a multi-year historical image data set. Single date SAR imagery has, in general, a low thematic information content, due to limited radiometric variation. The use of this information in classification is further complicated by the high spatial variability in topography and soils of agricultural scenes. The authors´ thesis is, therefore, that only multi-temporal SAR image stacks combined with optical image data used in a knowledge steered stratified classification approach can lead to a sufficiently detailed crop map product, especially if needed at an early stage in the growing season. The use of such an approach is demonstrated with a case study