Remote Sensing: Statistical Testing of Thematic Map Accuracy

Abstract

In order to achieve wider acceptance among users of thematic maps derived from remote sensing data, the interpreter must be able to specify the accuracy of his product. This requires a valid sampling procedure to estimate classification accuracy. Although several alternative methods have been used in the past, none provide sufficient statistical justification for the allocation of sample points in each category of land use using remote sensing imagery, This paper describes a more detailed and more reliable method for determining the most appropriate (i.e., minimum,) sample size. The concept developed and described in the paper incorporates the probability of making incorrect interpretations at particular prescribed accuracy levels, for a certain number of errors, for a particular sample size. The remote sensing sampling strategy presented has the added advantage that it can easily be adapted for use with most forms of remote sensing imagery, including orbital data. It provides a reliable framework for testing the accuracy of any remote sensing image interpretation -based land use classification using the minimum number of sample points; thereby saving time and money, especially if it is employed in operational surveys where high specification accuracy levels need to be guaranteed.

Classifying environmentally significant urban land uses with satellite imagery

“…This equation agrees with others (Swain, 1978;Van Genderen et al, 1978;Foody et al, 1995) who found that training data size for each class should be at least 30n to form representative training samples. In this case, with 7 spectral bands and 8 classes, at least 1680 training observations are needed.…”

Integrated Landsat Image Analysis and Hydrologic Modeling to Detect Impacts of 25-Year Land-Cover Change on Surface Runoff in a Philippine Watershed

“…Another independent set of samples were likewise collected for accuracy assessment. A minimum number of 30 ground truth pixels were randomly chosen for each class, following the guidelines of Van Genderen et al [35] to obtain a reliable estimate of classification accuracy of at least 90%. The MLC was used to classify the two Landsat images.…”

Land Use Change Detection using RS Data and GIS Application A case study Aswan Governorate

“…A random distribution of such sampling points over the whole region must be sought (Van Genderen et al, 1978). The accuracy assessment was conducted through a standard method described by Congalton (1991).…”