A robust approach and analytical tool for identifying early warning signals of forest mortality

Abstract

Forests are facing unprecedented stressors, evidenced by increases in the rate of forest mortality. Characterizing the state of forest ecosystems and their responses to disturbances remains a complex and crucial task. Existing methodologies have rarely been evaluated in real-world ecosystems due to challenges such as limitations in data availability and analytical techniques. To address these gaps, this study employs remotely sensed spatio-temporal data to identify early warning signals of forest mortality using satellite imagery. Utilizing local spatial autocorrelation methods, specifically local Geary's c and local Moran's I, a robust approach that yielded consistent results across multiple study sites is developed. This approach successfully generated early warning signals based on time-series analysis of local spatial autocorrelation metrics, providing up to a two-year advance notice of impending forest mortality events. The results demonstrated that the proposed approach could outperform previous techniques in reliably generating early warning signals of forest mortality, as shown by significant trend analysis. Additionally, a new R software package, “stew”, is introduced that is designed to facilitate user-friendly spatio-temporal analysis of ecosystem state changes. In summary, this study corroborates the potential of spatio-temporal indicators as valuable tools for predicting climate-induced forest mortality up to two years in advance.