Conclusions

RS based LC mapping and monitoring of large areas has created a new challenge with the varied spatial scale and data volume, requiring automated classification algorithms that minimise human interventions. This work has shown that use of spatial information along with complimentary ancillary and derived geographical layers is an effective way to improve LC classification performance which was demonstrated in three different terrains. In a highly urbanised area with less vegetation cover and highly contrasting features, texture played a major role in discriminating individual classes which were rather difficult to distinguish using only original high spatial resolution IKONOS MS bands. DEM played a role when the terrain is undulating, however, due to limited vegetation cover, vegetation index was not useful in classification. For the same urban area, inclusion of temperature, NDVI, EVI, elevation, slope, aspect, PAN and texture layers significantly increased the overall accuracy by 7.6% while discriminating different classes properly with Landsat ETM+ data. In a forested landscape with moderate elevation, temperature was the only factor that increased the LC classification accuracy. In a rugged terrain with temperate climate, addition of temperature, EVI, elevation, slope, aspect and PAN layers significantly improved the classification accuracy compared to the classification of only original spectral bands. Sometimes, improvements in spatial resolution of the data by integration of the spectral and spatial details from Multispectral and Panchromatic bands through image fusion also helps in object recognition, delineation and improves classification accuracy.