Abstract
Already proposed was a method using the Kalman filter which aimed to estimate the land-cover condition of a small area of the order of 200m×200m from remotely sensed data based on category decomposition principle. This method utilized only average values of band data for both categorical reflectance spectra and observation quantity over the small area. In the present paper we put forward a new land-cover estimation method using the extended Kalman filter which can take account of covariance features of categorical reflectance values as well as pixels brightness variation within the small area. The method is tested over the 4km×8km area drawn out from Odawara city for its validity verification. The results show that the new method outperforms the Kalman filtering method by 9.6% in terms of root mean square error in five land-cover category estimation and by 17.2% in thirteen category estimation.
