Mapping agricultural land use using retrospective ground referenced data, satellite sensor imagery and GIS

A postal survey, using questionnaires, has been used to collect retrospective land cover information for comparison with Landsat TM imagery. The questionnaires targeted selected farms in Warwickshire, UK based on spectral data from an image produced by an unsupervised classification of TM bands 2, 4 and 5. The information from the questionnaires was used as 'training data' in a supervised classification of the imagery and as 'testing data' for the assessment of classification accuracy. The analysis was performed using IDRISI, a raster based Geographical Information System (GIS). The overall accuracy of the classified image was 87%. Individual class accuracy ranged from 80% for oilseed rape to 94% for water. The Kappa coefficient for the classified image was 86.5%. The total area and percentage occupied by each class on the classified image was calculated. Comparisons with independent ground survey data indicated that difference in terms of area percentage coverage ranged from 0.45% for wheat to 1.49% for grassland. The methodology is workable for obtaining and using compatible ground referenced data with imagery taken in the recent past.