Content area
Abstract
Recent climatic warming has increased extent, frequency and severity of forest disturbance. These changes create considerable uncertainties on post-disturbance natural regeneration. Accurate information regarding disturbance occurrence and recovery are crucial for effective disturbance management strategies and improved understanding of disturbance dynamics in response to climatic changes. Yet, we still lack accurate and efficient remote sensing tools to allow for monitoring forest disturbances in a timely manner.
This study presents a new approach called ‘Stochastic Continuous Change Detection (S-CCD)’ using all available Landsat 5, 7 and 8 images for characterizing recent forest disturbances. S-CCD improves upon the current approach by incorporating a mathematical tool called ‘Kalman filter’, which models non-linear dynamics of trend and seasonality components as individual stochastic processes and enables a recursive model update. The dissertation surrounding this novel approach was organized into three parts. First, technical details of S-CCD were presented with a focus on its characteristic recursive form, which can facilitate near-real-time disturbance monitoring and improve computational efficiency. The generic detection accuracy of S-CCD was evaluated against a plot-based nationwide database across different disturbance agents, reporting 20% omission and 21% commission error rate.
To address high omission errors for non-stand replacement disturbances, a supervised framework named ‘PIDS’ was introduced in the third chapter for adapting S-CCD to detect subtle forest change by leveraging user-defined training samples. PIDS consists of four components: 1) Parameter optimization; 2) Index selection; 3) Dynamic stratified monitoring; and 4) Spatial consideration. PIDS was applied to map mountain pine beetle (MPB) and spruce beetle (SB) outbreaks in Colorado. The overall accuracy of PIDS is 0.84 for MPB and 0.71 for SB, making a substantial improvement (> 0.3) compared with other algorithms/products.
Lastly, built upon PIDS high-accuracy detection for beetle attack, we mapped early-stage spectral recovery following beetle and fire disturbances in Colorado using Landsat time series. The random-forest model was used to examine relationship between spectral recovery and the explanatory variables provided by S-CCD including severity, pre-disturbance density, successive disturbances and other climatic and topographic factors. The result shows that disturbance severity is the most influential variable with a strong positive correlation. This finding suggests that open space and light availability created by disturbance is critical for understory recovery for the early succession stage. Overall, this study demonstrated an accurate, near real-time, high-efficiency, and spatially explicit approach for characterizing forest disturbance, which was successfully applied to explain variability of post-disturbance regeneration in Colorado.
