The Impact of Respondents’ Background Towards Slum Conceptualisations and Transferability Measurement of Remote Sensing–Based Slum Detections. Case Study: Jakarta, Indonesia

Updated and reliable data on slums’ location, extent, morphology, and living conditions is critical to implementing and evaluating the effectiveness of slum improvement programs. Object-based image analysis (OBIA) allows extracting such information from imagery using shape, texture, density, and context to resemble human recognition of image objects. Although slums share similar characteristics, such as density, locations, and building orientations, they may differ locally, which requires adaptations of the OBIA ruleset; this reduces its transferability. Also, the most common approach in measuring transferability is using accuracy assessment by comparing the OBIA result with reference data created by domain experts using visual indicators. However, image interpreters have various local and professional experiences. Consequently, their choice of indicators to conceptualise slums and how they are delineating objects may lead to ambiguous results regarding mapped slum existence and extent. Our research aims to understand how respondents’ backgrounds impact slum conceptualisations and transferability measurement. We used three subsets in Jakarta, Indonesia, with different morphological characteristics and asked respondents with varied backgrounds to indicate slums and their characteristics. Our research concludes that different sources of uncertainties come from different understandings of slums, and the inability of rule-based OBIA to handle all uncertainties limits the ability to create a transferable OBIA ruleset for slum detections. We also conclude that the usage of accuracy assessment in measuring the performance of the image classification algorithm might be misleading when we are unaware of uncertainties in reference data.