The urban heat island effect is one of the most studied of all environmental phenomena and intersects with three pressing challenges of the 21st century: urban population growth, global climate change and increasing energy demands. More people are likely to crowd into denser, noisier and more polluted cities, experiencing higher temperatures and more extreme events such as heat waves. Wealthier citizens will demand energy-consuming air conditioning systems to avoid impacts on health and well-being that others cannot escape. In seeking solutions to this scenario, the starting point must be understanding how and where temperatures differ within cities at a spatial resolution relevant to people. While process-based modelling has provided good understanding of what should affect urban temperatures, detailed empirical studies of what actually drives local temperatures is still needed, particularly in structurally complex, smaller cities where classic heat island effects may not be dominant. In this paper, the complex interactions between multiple land–use drivers and land surface temperature (LST) are explored using machine-learning tools. Specifically, decision trees with stochastic gradient boosting are used to build an empirical non-linear, non-parametric model of Landsat-derived LST at 30 m resolution for the city of Southampton, UK. With a cross-validated correlation between predicted and measured LST of c. 0.95, this highly successful model is then used to simulate how spatial pattern in land use affects LST and heat island metrics on neutral landscapes. In particular, the authors address the “land-sharing v. land-sparing” debate in the context of thermal refugia within an urban setting. Analysis shows that Southampton has high heterogeneity in LST, driven by both immediate and adjacent land use. Grouping key land use types can regulate LST and the optimum design is neither at the land sharing or land sparing extremes, but a balance between the two. Spatial planning guidelines should consider the cooling effects of green space both on people to minimise the risks of heat exhaustion and on buildings to limit energy demand. Seeking such opportunities for urban cooling should be required in the planning process, enabling cities to build capacity for climate change and a rising population.