Mapping the Dieback of Several Tree Species in Centre of France Using Sentinel-2 Derived Indices : a Preliminary Analysis

In recent years, there has been a massive increase in forest dieback in temperate forests. This is largely due to climate change, which is associated with an increase in droughts and biotic and abiotic disturbances. A first step for the different stakeholders involved in forest management is to map the magnitude and temporal evolution of tree decline. The use of remote sensing data, in particular multispectral imagery acquired with Sentinel-2, has been identified as valuable for a timely and accurate mapping of forest dieback. First results obtained in the same region on oak trees have demonstrated that it is possible to map forest dieback accurality using Sentinel-2 derived time series and machine learning algorithms. This paper evaluates the relevance of this approach to map the dieback of other tree species (chestnut and pine) over several years. Our first results show that our approach can be applied to various tree species (deciduous and coniferous) without the need for extensive tuning and with satisfactory accuracy (balanced accuracy ranging between 73% and 78%). This opens many perspectives to continue and improve this work, including other species and area.