AREA ESTIMATION OF MULTI-TEMPORAL GLOBAL IMPERVIOUS LAND COVER BASED ON STRATIFIED RANDOM SAMPLING
Keywords: Area estimation, Impervious Land Cover, Stratified sampling, Design-based Inference
Abstract. Estimating area of impervious land cover is the most useful and one of the ecological assessment indexes of urban and regional environment. Global land cover maps are inevitably misclassified, which affects the quality and application of the data. Statistical approach for assessing the accuracy is critical to understand the global change information and area estimation is usually based on sample data with a probability-based estimator. However, research on evaluation of multi-temporal global impervious land cover maps has not been implemented. In this study, spatial characteristics of the data are considered to assess the thematic map accuracy with a two-stage stratified random sampling plan. The first stage of stratification is determined by the global urban ecoregion and the second one is determined by land cover classes. Additionally, sample size of both map stage and pixel stage are calculated using a probability sampling model. A response design is constructed for a per-pixel accuracy assessment and blind interpretation is implemented using sample pixels and its surrounding area. Our method is applied to the multi-temporal global impervious land cover maps between 2000 and 2010 with a time interval of 5 years and the estimated area in different epoch are listed in detail. The main contribution of our research is illustrating the details for calculating the proportion area of impervious land cover and corresponding confidence intervals based on the reference classification. The experimental results show that the increasing area of the impervious surface according to the sample unit shows good agreement with the urbanization and descriptive accuracy assessments by user’s, producer’s and overall accuracy are shown respectively.