Tracking the spatial and temporal evolution of drought-induced tree mortality using high spatial resolution satellite imagery

Vegetation mortality, due to both biotic and abiotic drivers, has recently been observed at various locations across the globe. Widespread changes in ecosystem structure and function due to mortality have the potential to significantly alter biogeochemical cycles and energy fluxes at local, regional, and even global scales. Our ability to quantify the consequences of vegetation mortality is currently limited due to a fundamental lack of information on where and when mortality events occur. Here we use multi-temporal, high spatial resolution (~ 1 m) satellite imagery acquired from a pinyon-juniper woodland in the Southwestern United States to develop and test a multi-stage tree stress and mortality detection algorithm. This region of the US has recently experienced significant amounts of vegetation mortality induced by severe drought and thus represents an ideal location for developing and testing remote sensing capabilities of mortality detection. The stress and mortality (SAM) detection algorithm combines object-oriented and spectral classification approaches to detect and delineate individual trees across the landscape. Spectrally-based mortality indices are derived from detected tree crowns and tracked through time as an indicator of the temporal evolution of relative tree stress. Initial tests of the SAM algorithm resulted in very high tree detection and stress classification accuracies (> 90%). Because our analysis is conducted at the tree level we are able to provide a demographically-based assessment of tree response to drought. Moreover, we demonstrate how statistical assessment of the spatial and temporal dynamics of tree stress and mortality can be used to reveal important information regarding the potential mechanisms by which individual trees succumb to stress factors.