Crop classification in rain-fed and irrigated agricultural areas using Landsat TM and ALOS/PALSAR data

Abstract

Multispectral optical remote-sensing systems have been the base for crop identification and monitoring for many years. However, cloud cover limits and even prevents the use of optical data for this activity. Synthetic aperture radar (SAR) sensors provide an interesting alternative to conventional multispectral classification schemes. Radar sensors can acquire data regardless of cloud cover, and their observations provide complementary information to that captured by optical sensors. The main objective of this paper is to evaluate whether the incorporation of polarimetric SAR observations to a multispectral classification scheme enhances classification results using the proposed method. With this aim, one Landsat thematic mapper scene and two ALOS/PALSAR quadrature polarization scenes acquired in 2007 were processed. The results demonstrated the ability of SAR data to improve the classification based on optical images. However, improvements were slight (an increase of around 2% in the classification's overall accuracy). The results improved significantly when barley, wheat, and oats were considered a single class, called cereals. The best results were achieved using Landsat and the two ALOS/PALSAR scenes together, obtaining an overall Kappa coefficient and accuracy of 0.67 and 90%, respectively. Probably, scenes acquired on other dates (June and July) would have yielded clearer classification enhancements.