Isolating and comparing the cooling effects of trees and short vegetation in large cities across the globe

Increasing vegetation cover is regarded as a nature-based solution to mitigate urban heat. Satellite-derived land surface temperature (LST) data at high spatial resolution can indicate comparatively warm and cold places within cities. This offers the opportunity to analyze the cooling effect of vegetation cover, and to separate it from other drivers. Further, it is possible to compare the cooling effects of different urban vegetation such as meadows and trees. Here, we use daily high-resolution LST data jointly with land cover information from >100 cities worldwide during their warmest three months in 2013. We train random forest models to predict LST patterns from land cover information for each day and city. As a first result we find that random forest models generally outperform linear regression models in predicting LST, and are therefore better suited to study the relative roles of individual drivers. Then, we estimate the influence of tree cover and short vegetation cover on LST by calculating SHapley Additive exPlanations (SHAP) values. We find that trees contribute to decreasing urban LST in most cities and days while only half samples indicate decrease of LST caused by short vegetation. Thereby trees have a much larger cooling effect than short vegetation. This is probably related to sustained transpiration during warm and dry conditions thanks to deep rooting systems, which is typically not the case for short vegetation. Also for trees, the cooling effect varies across climate regimes, with the largest effects in cities with temperate climate. Moreover, we find that the cooling effect of trees is particularly large during the hottest days while it is limited by high relative humidity. This probably reflects the impacts of radiation and vapor pressure deficit on tree transpiration. Overall, our analysis demonstrates how remote sensing data and machine learning methods can inform urban vegetation cooling to deal with more frequent hot extremes.