Reconstructing 20+ year history of subpixel forest canopy cover, structure, and disturbances at 30-meter scale with a suite of advanced Landsat image processing systems

Background/Question/Methods:

Canopy cover fraction is one of the primary vegetation structural characteristics used in ecological studies, including habitat suitability analyses, understanding structure-function relationships, restoration planning, assessing biodiversity, etc.   Many of these questions require accurate data over large territories that quantifies multi-decadal history of canopy growth, degradation, and disturbances at the scales that are appropriate for land management.  Landsat satellite images at 30-m resolution have been frequently used to extrapolate plot or patch level measurements of canopy cover and heterogeneity to landscape scales.  However, a challenge still exists in simultaneously achieving the following three requirements: (1) accurately estimate subpixel tree canopy fractions, (2) provide these estimates over long terms, and (3) account for forest disturbances.

To address this challenge, we combine our two recently developed Landsat processing systems: eDaRT (Ecosystem Disturbance and Recovery Tracker) and MixSSMA (the Mixed Stratified Spectral Mixture Analysis).  In each 30-m cell of Landsat, MixSSMA estimates tree class fractional cover for each date, while eDaRT provides independent automated disturbance detection, thus determining the break points (regime change) in the fraction trajectories.  Combining this information, the mapping process can impose temporal stability and fraction transition feasibility constraints, thus increasing the accuracy of the canopy fraction time series product. Results/Conclusions:

We demonstrate these capabilities, in the example of mapping 20+ year history of subpixel canopy cover and disturbances in predominately mixed conifer forests across north-south and elevation gradients in the Sierra Nevada, California.  Initial validation experiments show good agreement with reference date derived from LiDAR and higher-resolution imagery, with RMSE of live tree cover fraction estimation below 0.1. Using these data, one can generate various metrics of forest heterogeneity that are relevant for habitat monitoring and restoration prioritization, and provide snapshots and assess temporal trends in forest cover and structure.