SNAGGING DROUGHT-RELATED TREE MORATLIY OVER TIME WITH MACHINE LEARNING

Abstract:

Drought-related forest mortality events are forecasted to increase globally with climate change, which challenges managers tasked with improving forest resilience to multiple and compounding stressors. In the fire-frequent forests of the Intermountain West of the United States, low to moderate intensity fires produce forest structure patterns of individual trees, tree clumps, and openings that confer resistance to future wildfire. We examined how drought-related tree mortality varied by forest structure patterns and biophysical factors immediately before, during, and one-year post drought, 2018 in the southern Sierra Nevada, California. We developed a neural network using multi-temporal, paired airborne lidar and hyperspectral data (NEON); smallsat 4-band imagery (NASA’s Commercial Smallsat Data Acquisition Program; and field data to: 1) model snag detection with lidar intensity over time; and 2) identify changes in mortality patterns with respect to time, forest structure patterns and biophysical factors. We present our methodology for multi-temporal and sensor investigations of drought-related mortality patterns as well as our findings regarding drought-related mortality patterns to inform future forest interventions.