Prediction of climate-induced forest mortality using mechanistic vegetation models still remains challenging, mostly due to complex interactions of physiological mechanisms that lead to mortality. An approach to improve predictions of climate-induced forest mortality is through detection of resilience-based early warning signal (EWS). When approaching a ‘tipping point’ i.e., close to mortality or a similar transition event that leads the system to a new equilibrium state, natural systems tend to show abnormally low resilience. This low resilience, which is characterized by low recovery rate from perturbations, may serve as a potential EWS. Using forests in California as a test bed, this study evaluates the time varying resilience using a Bayesian dynamic linear model and time series of Landsat 7 images and climate forcings. By comparing the location and timing of EWS occurrences with observed mortality from forest aerial surveys, this study seeks to address the following questions: (1) Can the resilience based EWS be identified prior to observed forest mortality? (2) How early does the EWS appear? (3) Can the EWS be used to predict forest mortality?
Results show that most areas with mortality exhibited an EWS 6 to 18 months prior to mortality. The lead time of EWS varies with climate regions and plant genus. Temporally, mortality probability of a plant genus is found to be strongly related to the fraction of area with EWS in that plant genus, explaining 73% of the inter-annual variation of mortality on average. However, the signal vanishes when collated for each climate region, highlighting that the resilience signals are highly genus dependent. Spatially, incorporating characteristics of EWS, in addition to using topography and competition that have been considered in previous studies, further improves the estimation and prediction of forest mortality.
Liu, Y., M. Kumar, G.G. Katul, and A. Porporato (2018): Reduced resilience as a potential early warning signal of forest mortality. Ecological Society of America Annual Meeting, August 5-10, 2018, New Orleans, Louisiana.
This Paper/Book acknowledges NSF CZO grant support.