A machine learning methodology to quantify the potential of urban densification in the Oxford-Cambridge Arc, United Kingdom

Regional-scale urban residential densification provides an opportunity to tackle multiple challenges of sustain-ability in cities. But framework for detailed large-scale analysis of densification potentials and their integration with natural capital to assess the housing capacity is lacking. Using a combination of Machine Learning Random Forests algorithm and exploratory data analysis (EDA), we propose density scenarios and housing-capacity es-timates for the potential residential lands in the Oxford-Cambridge Arc region (whose current population of 3.7 million is expected to increase up to 4.7 million in 2035) in the UK. A detailed analysis was done for Oxfordshire, assuming different densities in urban and rural areas and protecting lands with high-value natural capital from development. For a 30,000 dwellings-per-year scenario, the land allocated in Local Plans could cover housing growth in the four districts but not in Oxford City itself (which accounts for 48% of the demand); only 19% of the need would be covered in low but 59% in high housing density scenarios. Our study suggests a decision-support method for quantifying how the impact of housing growth on natural capital can be significantly reduced using more compact development patterns, protection of land with high-value natural capital, and use of low-biodiversity brownfield sites where available.