Understanding recent global-scale decline in forest health requires methods that characterize forest change across broad regions, yet with sufficient temporal and spatial resolution to support management action. Operational methods that concurrently detect forest anomalies and characterize magnitude of change are in great demand. The Ecosystem Disturbance and Recovery Tracker (eDaRT) is a highly automated, broadly applicable disturbance mapping system that processes all available Landsat imagery, detecting change at 8-16 day timestep, and is operated by the US Forest Service Pacific Southwest Region to generate disturbance map products for science and land management applications. We report on a new method to estimate canopy loss using time series of spectral change associated with eDaRT disturbances.
Our study area included montane to subalpine forests in the Sierra Nevada (650,000 ha). Training data included 850 Landsat pixels, with canopy cover measured from high resolution imagery on two dates separated by a disturbance event caused by drought or insects, between 2011 and 2017. We used beta regression to model canopy cover loss as a function of eDaRT spectral change integrated over pre- and post-disturbance time periods in different variants to balance the needs for prediction accuracy with nimble delivery of products as forest management inputs.
Results/Conclusions
The resulting eDaRT Mortality Magnitude Index (eMMI) combines vegetation indices known to be related to vegetation cover, moisture, and health, including the Normalized Difference Vegetation Index, Normalized Burn Ratio, and Red-Green Angle. Canopy cover loss was best modeled by including variables representing both proportional and absolute spectral change and their temporal variability, yielding a root mean square error (RMSE) of 13%. Across our study area, the annual extent of forest mortality ranged from 20,000-80,000 hectares during 2011-2018, with spatial expansion peaking in 2015. Mean canopy cover loss was only moderately correlated to spatial expansion, reflecting the varied causes of forest mortality we mapped and emphasizing the need to detect change in terms of both extent and magnitude. We provide an overview of plans for operational implementation of this tool for the Pacific Southwest Region of the Forest Service, and its potential to improve the accuracy and efficiency of delivery of forest change products for researchers and managers.