Using machine learning to improve resolution and bias in urban temperature projections
Global climate models use equations to represent the processes and interactions that drive the Earth’s climate, but their simulation outputs are coarse-gridded. Statistical downscaling is a popular data-driven technique that uses climate model outputs to obtain future projections at a higher spatial resolution. In this work, we demonstrate a novel and simple regression method to forecast the distribution of daily mean temperature in cities. The value of this method includes its ability to generalise better at extremes compared to traditional downscaling techniques, and its ability to quantify statistical uncertainty.
Engineered world Policy and governance