Detecting early warning signals of tree mortality using multi-scale satellite data: a case study in boreal North Americ

Tree mortality from drought and biotic infestations represents a fundamental transition for forests at the level of an individual tree, forest stand, and even landscape. Tree mortality precipitates a cascade of ecosystem impacts and has been increasing across the continents, including in the boreal zone where climate changes and feedbacks to the Earth system are relatively large. Despite the importance for science and management communities, our ability to forecast tree mortality at landscape to continental scales is limited. However, two independent information streams have the potential to inform and significantly improve mortality forecasts. Tree-level productivity dynamics are known to precede mortality in predictable ways years to decades before death. Plot-level productivity, in turn, has been related to satellite-based indices such as the Normalized Difference Vegetation Index (NDVI). Here we link these two data sources to show that early warning signals of mortality are evident in several NDVI-based metrics up to 20 years before death. We focus on two repeat forest inventories and three NDVI products across western boreal North America where productivity and mortality dynamics are influenced by periodic drought. These data sources capture a range in forest conditions and spatial resolution to highlight the sensitivity and limitations of our approach. Overall, results indicate potential to use satellite NDVI for early warning signals of mortality. Relationships are broadly consistent across inventories, species, and spatial resolutions, although coarse-scale imagery in the heterogeneous aspen parkland was of limited utility. Longer-term NDVI data and annually re-measured sites with high mortality levels generate the strongest signals, although we still found robust relationships at sites re-measured at a typical five-year frequency. The approach and relationships developed here can be used as a basis for improving forest mortality models and monitoring systems.