How Local Perspectives Matter. Local and Outsider Perceptions of Urban Deprivation: A Deep Learning Approach Using Remotely Sensed Imagery of the Accra-Tema City Region

Abstract

The rapid growth of cities in low and middle-income countries (LMICs) has put increasing pressure on livability, resulting in deprived urban areas (DUAs). Previous research has shown that local livability perspectives can be accurately modeled using deep learning techniques. The downside of this modeling approach is that it requires substantial amounts of local training data, which limits the scalability and transferability of the method. This study investigated if an outsider perspective could serve as a proxy variable for a local DUA livability assessment using a deep learning approach. Data on the outsider perspective of the Accra Tema city region (ATCR) was collected with participants from The Netherlands (n = 56). The participants performed pairwise image comparisons of remotely sensed imagery, to evaluate which area is relatively better or worse in terms of livability. The outcome of the evaluations was used as labeled training data for an AI livability prediction model. The predicted livability assessment of the trained AI model (outsider-trained model) was systematically compared to the outcomes of preliminary studies in the ATCR (locally trained model) using an urban morphometric framework. The study found significant differences between the two perspectives, which manifested unevenly across space. Correlations between the predicted livability scores showed considerable variation based on urban morphology. Within clusters of compact low rise, a strong positive relationship was found between the local and outsider perspectives, whilst open low rise showed a weak correlation. Furthermore, the differences between perspectives varied in their dependence on urban morphometrics. Local participants perceived areas with lower building densities, less greenery, and paved roads as more livable, while the outside group preferred green spaces and surface water. In sum, the findings reinforce the necessity of including the local perspective in DUA livability assessments while showing that the differences from the outsider perspective depend on urban characteristics. Future research could focus on finding avenues for leveraging the outsider perspective in a complementary role in data-scare environments.