Accuracy Assessment of Land Cover Fraction Map Derived from Landsat 8 Imagery Using Linear Spectral Mixture Analysis Method

Abstract

Accuracy assessment is conducted to assess how well a classification method is. One of the land cover classification method for mixed pixel problem is Linear Spectral Mixture Analysis (LSMA). Error matrix is commonly used accuracy assessment of categorical data. But accuracy assessment for fraction map as result of LSMA processing is not really examined yet. LSMA method produces map fraction which contains land cover proportion. This experiment aims to examine the effect of sampling unit on land cover fraction map accuracy assessment result. 70 samples were taken using stratified random sampling. Geoeye 1 image is used as reference data corresponding to ground truth data. Accuracy assessment was performed in two scenario sampling unit which is single pixel area and a cluster of pixels area. The accuracy of LSMA model was influenced by position error between classified map and the reference data. Selection the right sampling unit is a one way to minimize the deviation caused by positional error corresponding reference data.