In a context of global environmental modifications, the effect of climate change already generates sudden events (storms, heat waves, long drought periods for example) which, in urban environments, impose critical challenges to cities. According to the World Bank, 54% of the world's population is living today in urban areas and this number is expected to reach 80% in 2050. Given the expected environmental changes and their drastic consequences, it is increasingly important to have policies and approaches that are capable of making cities resilient and adaptable to these sustainable changes. It is now well understood that vegetation, particularly trees, provides very effective solutions for reducing these impacts: among other properties, trees can cool cities and improve air quality or biodiversity. However, to play a positive role, trees must survive and maintain a good physiological state in such a hostile context of drastic modifications of their environment (e.g., hydric and thermal stresses, resource limitations, physical constraints). As shown recently in the summer of 2022, the long period of hydric stress associated with high temperatures has caused the death of many trees. However, some others survived despite, a priori, unfavorable environmental contexts. These surprising observations disturb city managers who lack tools for monitoring tree well-being, able to alert in case of large stress, and guide future and adapted management plans for tree plantations. This lack is because we are facing a complex system with multiple interactions that are not yet understood. In order to better understand the response of trees to extreme solicitations, to prevent their senescence at an early stage, and to provide maps of the comfort of trees in cities associated with alert systems, the guideline of this project is to acquire a large number of physiological variables (in real, simulated and controlled conditions) associated with their environment (urban pattern, climatic conditions), and to exploit machine learning in order to identify links between them and to generate global maps. An important underlying question is to evaluate the ability of remote sensing data to measure physiological variables related to the internal structure of trees. We will explore learning estimation models able to derive some relations between physiological variables and their counterparts in images. An originality here is that these methods will be calibrated on the basis of physical simulations, which will allow us to i) guarantee a sufficiently large number of data to train our models and ii) ensure the physical consistency of our estimators. These models will then be refined on data acquired in controlled environments to ensure, again, a physical consistency of our models. In a second step, from the models mentioned above, the application in real conditions (i.e. using real remote sensing images and in situ measurements on trees) will be performed. To this end, the transferability of our models in real situations will be studied. In addition, to ensure a free reproduction of our processes, we will rely on freely available SENTINEL 1⁄2 images. These latter being acquired at bast at 10m resolution, a super-resolution step will be required to improve the spatial resolution at 3m, more adapted to the scale of cities. From generated maps of biophysical parameters at the city scale and internal measurements inside trees, we will develop methodologies to estimate tree comfort indexes at the scale of a city. To this end, qualitative data measured on trees will be used on a large number of trees and will be linked to stress factors (urban patterns, temperature). Specific attention will concern the analysis and interpretation of key features, issued from machine learning, connected to tree physiology in order to help the interpretability of urban patterns connected to tree physiology.