Dasymetric Mapping of Population Using Urban Settlement-Focused LULC Classification: A Comparative Evaluation of Random Forest, OBIA, and Hybrid OBIA-RF Methods

Urbanization of cities introduces growth of population in urban settlements, formal and informal. However, these informal settlements commonly go unrecorded due to legal threats, thus the scarcity of data. This study aims to bridge the scarcity of urban settlement information by dasymetric mapping of population using census blocks and land use/land cover (LULC) map, focused on formal and informal urban settlements. To achieve this, three classification methods for LULC were compared: random forest (RF), object-based image analysis (OBIA) using a Bayes classifier, and a hybrid approach combining OBIA with RF. Land classification was applied to Sentinel-2 L1C images from 2015 and 2020, with Google Earth Engine utilized for RF and QGIS Orfeo Toolbox for OBIA and OBIA-RF. To evaluate which of LULC classification methods is most accurate, the F-scores of each generated LULC map per method were compared. Furthermore, the reliability of the LULC classification methods for population dasymetric maps were compared to built surfaces of Global Human Settlement Layer through Global Similarity Value. F-score values ranging from 86.37%–94.44% for formal settlement, 65.18%–72.19% for informal settlement, and 0.4978 Global Similarity value from hybrid OBIA-RF show that it has most capability in mapping informal settlements due to incorporation of spectral, spatial and contextual characteristics. The most effective method for LULC allowed dasymetric population mapping projected for 2025, such that LULC prediction was executed using an artificial neural network-multilayer perceptron (ANN-MLP). These outputs are expected to provide valuable insights into population distribution in informal settlements, supporting urban planning and resource management efforts.