Modeling multi-type urban landscape dynamics along the horizontal and vertical dimensions

Abstract

Fine-scale urban simulation obtained through vector-based cellular automata (VCA) can provide more authentic parcel-level maps to aid urban planning. However, two aspects of VCA still need improvement for more detailed modeling of urban landscape dynamics. One is the lack of simulation of urban renewal prevalent in megacities. The other is the lack of simulation of vertical urban landscape dynamics at the parcel level. Hence, this study should be the first to propose a comprehensive VCA framework for modeling horizontal and vertical multi-type urban landscape dynamics (HV-MVCA) at the parcel level. The HV-MVCA consists of two modules: a deep learning-based horizontal urban landscape dynamic simulation module and a random forest-based (RF) vertical metrics prediction module. Compared with traditional VCA, the HV-MVCA can obtain the highest simulation accuracy with an enhancement of 1.45%-4.56%. The RF-based fitting model can reasonably interpret the vertical metric as the coefficient of determination and the mean absolute percentage error was 0.88 and 28.93%. By localizing shared socioeconomic pathways (SSPs), the horizontal and vertical urban landscape dynamics in the future can be simulated under different scenarios. As the multi-scenario simulation results can provide abundant horizontal and vertical information at the parcel level, the proposed HV-MVCA can help achieve a more comprehensive assessment of many urban-related issues with these indicators, such as energy consumption, climate change, and pollution emissions.