Predictions of land cover & land surface temperature based on machine learning algorithms to support climate-resilient Urban planning in UK cities

Abstract

Cities are becoming increasingly warm as a result of climate change and increasing population density. UK cities are projected to experience temperature increases of 3-7°C by 2100 under current climate scenarios, with urban areas warming faster than rural counterparts due to extensive impervious surfaces, reduced vegetation, and anthropogenic heat emissions. This research develops and validates an integrated machine learning framework for predicting land cover and land surface temperature changes in Glasgow, Cardiff, and Cambridge, providing evidence-based tools for climate-resilient urban planning. The study combines remote sensing data from 1990-2023 with advanced analytical techniques to project urban thermal conditions to 2031.

The MOLUSCE artificial neural network-cellular automata framework achieves robust land cover prediction performance with Kappa coefficients ranging from 0.79-0.87. Random Forest emerges as the optimal algorithm for temperature prediction, demonstrating R² values of 0.87-0.90. ERA5-Land calibration aligns satellite-derived LST with ground-level temperature conditions through weighted least squares regression, achieving correlation coefficients of 0.76-0.86. Analysis reveals divergent urban development trajectories: Glasgow demonstrates successful green transformation with minimal projected warming (0.11°C), Cardiff balances urban expansion with grassland growth yet faces substantial warming (2.07°C), whilst Cambridge exhibits concerning environmental degradation with temperatures approaching critical stress thresholds (21.90°C in urban centres).

High-resolution analysis of 23 land use types reveals that commercial-residential centres generate the highest temperatures, whilst water infrastructure provides cooling of 1.40-3.90°C. The research establishes evidence-based temperature thresholds for planning interventions and demonstrates that urban thermal destinies can be actively shaped through strategic planning. This framework provides transferable methodologies enabling any UK city to develop targeted adaptation strategies based on quantified relationships between land cover, land use, and temperature.