Seasonal Variation in Land-Cover Classification Accuracy in a Diverse Region

This study determined the effect of seasonal spectral variability on land-cover classification of Landsat TM by comparing accuracies in different seasons in a mid-latitudinal (29 degrees 30' to 31 degrees 0'S) region with summer and winter rainfall, a broad altitudinal range, a temperate to subtropical climate and diverse land uses (e.g., summer and winter crops and nature conservation). The TC, PCA, and NDVI images were used alone and in combination with original bands for classification. Results using PC1-3, Bands 1 through 5, Bands 1 through 4, and 7 combinations were better than others, but the Bands 1 through 4 combination performed best across all dates. The highest accuracy was obtained with a mid summer (January) image (96.5 percent, Kappa 0.96) and the lowest with an early spring (September) image (86.7 percent, Kappa 0.84). The superiority of mid-summer images in generating highly accurate land-cover classifications was demonstrated by analysis of additional images under varying rainfall conditions in different years.