Population Estimation Using Land-Use Change Data from Multi-Sensor Images in Maha Sarakham Province, Thailand

Abstract

Land-use changes surrounding Mahasarakham University in Thailand were investigated using multi-sensor images from 2002 and 2019. This study used aerial photographs and Landsat-7 satellite images captured in 2002, and aerial photographs from an unmanned aerial vehicle and Sentinel-2A data observed in 2019. Visual image interpretation (VII), object-based image analysis (OBIA), and random forest (RF) methods were applied to classify building areas from the multi-sensor images. Population was estimated using buildings and field-survey data, and population samples. The samples were obtained by point-, pixel-, and area-based methods. The different population estimation approaches were then compared with the actual population based on field surveys. VII yielded accuracies of 97% in 2002 and 97.5% in 2019. Built-up extraction using RF yielded accuracies of 86.55 and 90.76%, whereas OBIA was 76.47 and 82.35%, indicating a transformation in the land use from paddy fields to urban and residential areas. The area-based method were highly efficient in 2002 (r2 = 0.92) and 2019 (r2 = 0.93). The proposed area-based method provides more accurate population estimates than existing methods, with accuracies considered to be comparable to those of field data.