Enhancing transparency in land use change modeling: Leveraging eXplainable AI techniques for urban growth prediction with spatially distributed insights

Abstract

Recent applications of machine learning (ML) and deep learning (DL) techniques in land-use change modeling have demonstrated significant success in capturing the intricate dynamics of land development. However, their “black-box” nature restricts their utility in various contexts, such as uncovering the underlying drivers of urban expansion. To mitigate this issue, we propose to utilize eXplainable AI (XAI) techniques in ML/DL methods, which presents a promising solution to this primary constraint. To that end, we introduce DL methods to investigate and predict the non-linear dynamics of land use changes. These methods achieved notably high accuracy scores and were more computationally viable than traditional statistical approaches. Moreover, the proposed approach employed in this study surpassed the parameter interpretation capabilities of statistical methods. More specifically, the proposed XAI approach not only highlights the average effects of features on the outcome but also elucidates the factors influencing specific decisions regarding land use changes, including the number of vacant parcels, the share of single-family parcels, and certain time-lagged neighborhood features. Such analyses provide invaluable insights for researchers, practitioners, and policymakers.