Supervised classification of multisource satellite image spectral and texture data for agricultural crop mapping in Buenos Aires Province, Argentina

We developed and tested an operational methodology to map agricultural crops from the classification of RADARSAT and Landsat TM spectral and texture data in southeast Buenos Aires Province, Argentina. The main crops of interest were wheat (in winter) and corn, sunflower and potato (in summer). The crops rotate with pasture for livestock every 4-5 years, creating a high spatial heterogeneity in the land use and land cover characteristics. Me used RADARSAT data (SGF W2 Ascending, 10 July, and 7 November 1997) and textural derivatives obtained from co-occurrence matrices in combination with optical satellite data (Landsat TM - bands 3, 4 and 5, 27 November 1997). Standard methods for georectification, backscatter calculation and texture calculation were employed. Training and test areas were selected based on field surveys, and a supervised classification was performed using the maximum likelihood decision rule. The mapping results indicate that RADARSAT tone and texture data provide a superior classification result (approximately 10% better) when compared to the use of Landsat TM data alone in the decision rule. The overall classification accuracy was approximately 62% correct (95% Confidence Interval = 57.8% - 66.5%). Of principal interest was the wheat class, which was determined to be approximately 81% correct.