People in pixels: Developing remote sensing-based geodemographic estimation through volunteered geographic information and crowdsourcing

Pixel-based remote sensing (RS) is widely used for social science research in particular for geodemographic interpolation data studies i.e. population distribution, poverty mapping, etc. Basically, RS research needs adequate amount of ground-truth data either for developing, validating, or improving the accuracy of the RS model. The oldfashioned ground-truth collecting from the field is time-cost consuming notably if the area is large heterogeneity. Development of crowd-sourcing methods for geoinformation science (GIS) and RS is getting attention and become more popular, known as volunteered geographic information (VGI). VGI may offer time-cost efficiency technology for ground-truth collecting data. We use mobile phone-based survey123 for ArcGIS to collect the ground-truth data. These ground-truth data used for improving the geo-demographic estimation map. Firstly, the geo-demographic map was developed by the traditional dasymetric method. Then, Support Vector Machine algorithm is used for improving the accuracy of traditional dasymetric map by using several predictor variables from remote sensing dataset. DKI Jakarta Province was chosen as a case study. We compare the geo-demographic estimation model with the existing data from Central Statistics Bureau of Republic of Indonesia. Using the combination of crowdsourcing, VGI dataset and several remote sensing-based predictor variables, a new method of geodemographic estimation was produced with higher R² of 0.66 compared with the traditional dasymetric method with R² of 0.09.