Understanding the Influence of Different Sample Sizes and Sample Techniques on Accuracy Assessment of Land Cover Mapping: Case Study of Salatiga city, Indonesia

Remote sensing data has been proven capable and efficient as a powerful resource for large-scale land cover mapping. However, a map is considered acceptable with the required accuracy value. The problem related to sampling is how the sample amount and sample technique affect the accuracy of the land cover mapping. Furthermore, the accuracy assessment for mapping usually only utilizes accuracy measurement standards, which are commonly used. This research was conducted to measure the effect of the different sampling sizes and sampling methods on the accuracy value of large-scale land cover mapping using area based assessment approach. A visual interpretation was used as a reference while multispectral classification was carried out independently as an object to be tested for accuracy assessment. The number of classes interpreted was 25 and 9. We demonstrated the sampling methods applied were random sampling, stratified random sampling, and systematic grid sampling. A confusion matrix method was used to gain the overall accuracy. The result of this study showed that the number of 200 samples for land cover with 25 classes and 36 sample for nine classes could start the regularity against the actual accuracy. While the sample number below 200 and 36 for both land cover classes showed irregular fluctuations in the accuracy value. Using stratified random sampling was satisfactory for modeling the accuracy compared to random and systematic grid sampling. Thus, those results could be used to indicate accuracy value against different scenarios and gain a recommendation for assessing the accuracy ofland cover on a large scale. © 2024 SPIE.