Future climate change is projected to raise temperatures and induce extreme droughts, resulting in more severe fires and forest mortality. In California, considerable management efforts are ongoing to preserve carbon and water stocks and prevent catastrophic wildfires. However, current records of fire, management, and forest die-off are represented as coarse polygons, which ignore spatial heterogeneity within perimeters of disturbances, resulting in large uncertainties in the trends, vulnerability, and resilience of California’s forests. To address this, we combined databases of perimeters of fire, management, and insect-driven forest die-off (n = 324,826) with multi-decadal (1984 – 2019) time series of 30 m resolution Landsat remote sensing data to estimate the drivers, area, and timing of forest loss across California. We used the Continuous Change Detection and Classification (CCDC) algorithm to estimate the pixel-level timing and magnitude of disturbance from an ensemble of spectral trends and combined them with the disturbance polygons to identify areas of non-disturbance within perimeters (e.g. “unburned islands” within fires), and disturbed areas that extend beyond the perimeters. We then combine the spectral and refined area data with individual-level estimates of tree mortality in the Sierras to train a classifier and identify harvest, fires, and mortality across the state.
Results/Conclusions
Preliminary results show large differences in the estimated areas of fires mortality between using our combined remote sensing and polygon-based database and using just the polygon-based database. Over 4,752 fires, Landsat-derived burned area estimates were, on average, just 57.7 ± 13.3% of the polygon-based estimates, though average this varied widely by ecoregion (up to 72.7 ± 25.3% in the Sierra Nevada Mountains and as low as 23.5 ± 13.0% in the Southern California Mountains). A large number of the fire polygons contained considerable areas “unburned islands”, which represent areas where trees survived the fire. Using the refined areas of fire, we estimate a trend over 1984-2019 in burned area per year that is approximately half of that when estimated by the polygons. We present analogous results for drought and harvest-driven forest losses and compare and contrast the extent, rate, and trends of forest mortality driven by these causes. By harnessing the full Landsat time series, we generated more refined estimates of the timing and locations of forest disturbances and estimate the rates of forest mortality resulting from fires and the recent extreme drought, which will provide considerable assistance to forest managers seeking to efficiently protect California’s water and carbon resources.