Monitoring drought impacts on street trees using remote sensing - Disentangling temporal and species-specific response patterns with Sentinel-2 imagery

Healthy street trees provide important ecosystem services to cities, but are under increasing stress from urbanisation and climate change, including drought. Traditional field observations are limited in their ability to provide city-wide and regular monitoring of the drought response of street trees, which is essential for their effective management. To overcome this, we propose a novel workflow using freely available remote sensing imagery from Sentinel-2 to identify temporal and species-specific drought response patterns of street trees. We evaluate the workflow on a sample of 2514 mature street trees of the seven most common street tree species in Leipzig, Germany. For each tree, we generate time series of eight vegetation indices from 2017 to 2022, namely enhanced vegetation index (EVI), normalised burn ratio (NBR) and normalised difference vegetation index (NDVI), each using the 10 m and 20 m resolution near-infrared band (EVI-10, EVI-20, NBR-10, NBR-20, NDVI-10, NDVI-20), as well as red edge normalised difference vegetation index (RENDVI) and red-green vegetation index (RGVI). They form the basis for a correlation analysis with the meteorological drought indicator standardised precipitation evapotranspiration index (SPEI) and for annual growing season integrals, which we subtract from those of the base year 2017. We use boxplots and statistical hypothesis testing to examine differences between and within tree species and years. The results show positive relationships between the eight vegetation indices and the SPEI, with the NBR-20 having the highest correlation coefficients. We also find significant differences in the drought response of several tree species, with Quercus robur being the most drought responsive and Platanus x acerifolia being the least. While most tree species have significantly smaller growing season integrals in the drought years 2018 and 2020 than in the non-drought years 2017 and 2021, the effects are not as pronounced in the drought years 2019 and 2022. Uncertainties arise, for example, from spectral signal variations caused by adjacent land use. Nevertheless, the proposed workflow holds promise for incorporation into holistic urban green management solutions and mixed-method approaches for further research into the causes and consequences of drought-induced damage to street trees.