Thematic accuracy development of Imperviousness Density Layer for years 2006 and 2009 in Slovakia and Czechia

The Imperviousness Density Layer (IMD) of the Copernicus Programme is a raster with 100 m pixel resolution and pixel integer values from 0 to 100 % representing the imperviousness degree. So far, the IMD validations have been based on stratified systematic samples with relatively small sample sizes, in the case of the joint area of Slovakia and Czechia totalling about several hundred sample pixels, each with 25 sampling points. The objective of this paper is to evaluate the IMD thematic accuracy based on a large simple random sample of 20,000 pixels per country per year, each pixel with 100 sampling points. A large sample is capable of a more precise estimate of the omission error compared to the smaller samples used before (when dealing with a small proportion class), which is important e.g. in population disaggregation (dasymetric mapping). The focus is on the IMD for the reference years 2006 and 2009 in Slovakia and Czechia and how this accuracy developed with each new version. The analysis reveals gradual improvement in most (but not all) of the different aspects of the IMD thematic accuracy with almost each new version, especially with the latest ones (published in 2019) compared to the previous ones (published in 2009, 2010 and 2013). However, the estimates of the proportions of impervious surfaces from the total area of Slovakia and Czechia based on these latest versions seem to be the worst compared to the previous ones. Also, the area of impervious surfaces incorrectly classified as pervious in a large number of pixels with zero imperviousness map values and small nonzero imperviousness reference values (i.e. in 'major underestimation error pixels' or 'omission error pixels with a threshold value set to 1%') remains large, and its proportion of the total area of incorrectly classified surfaces reaches its maximum in the latest versions.