Assessing forest disturbance and recovery with spatial and temporal structural morphology

点击翻译按钮获取中文摘要

The rapid adoption and availability of LiDAR has refocused research attention on investigating the vertical structure of forests to gain critical insight about specific stand conditions for improved forest management. Additionally, a growing temporal record of satellite image data is stimulating the exploration of time-series analyses of forest change and recovery. This proposal seeks to fuse these concepts and extend the broad foundational work of landscape metrics, and specifically structural segmentation, into a third-dimension by including either vertical structure or time. The spatial-temporal results will allow us to learn the three-dimensional spatial and structural recovery trajectories of forests and to then backwards-project the dates of forest disturbances, compensating for missing or contaminated satellite scenes and more specifically to extend into the pre-1972 period of available commercial satellite imagery. Boreal forests form the largest terrestrial global biome and are complex environments comprising biotic and abiotic elements that interact across scales to interconnect chemical, biological, climatic, physical, and human processes. These forests are prone to cyclic disturbances, dominated by fire and harvesting that shape their structures. Attempts to directly estimate tree heights, volume, species, understory, disturbances, terrain, and habitats are motivating improvements to simply correlating spectral canopy signatures with stand metrics. The new structural trajectories will improve disturbance detection, mapping, classification, and correlation with landscape disturbance processes in boreal forests or the modelling of light penetration, moisture distribution, and airflow systems. While landscape metrics for characterizing forest landscape patterns have existed for, and been developed or implemented for decades, their use and interpretation are fraught with challenges. Metrics are generally constrained to 2D, and rarely extend to 2.5D characterizations, and pose difficulties when attempting to statistically compare results. Morphological spatial pattern analysis has evolved to provide a more intuitive and tangible alternative to landscape metrics in providing landscape segmentation based on relative structural relationships. We will develop the explicit logic and corresponding software to extend morphological segmentation to become a true 3D characterization of landscapes. Methods will be subject a sensitivity analysis and be used to compare the effects and recovery of landscape processes such as fire, harvesting, and forest road building activities, and to identify critical structural differences among them; improvements will benefit appraisals of database accuracy, carbon accounting, and inform forest management planning. HQP will receive extensive training in STEM as we strive to extract the vast amount of information that exists in 3D data, extending beyond what might be contained in a single image.