Performance evaluation of supervised classification of remotely sensed data for crop acreage estimation

Remotely sensed satellite data have shown their ability to map and monitor different crops. In this research work, in addition to crop acreage estimation of a province in Iran, different supervised classification techniques have been used to study their performances for crop area estimation in comparison with statistics provided by the Agricultural Statistics and Information Department (ASID) of the Ministry of Agriculture of Iran. A total number of nine IRS LISSIII scenes acquired in 1998 pertaining to Markazi province of Iran were registered and used to make a mosaic of the whole province using EASI/PACE image processing software. Since images have been acquired in different date of the year 98, different training sites were selected for a single crop. Different supervised classification algorithms were applied to estimate crop acreage using similar training sites. The results of comparison between areas derived from classification techniques and area reported by ASID shows that a maximum likelihood classification (MLC) algorithm in conjunction with a parallelepiped algorithm are suitable for total crop area estimation. And based on the KAPPA coefficients of different classification algorithms, it is concluded that the best algorithm is MLC