Characterization of Forest Disturbance in California using Landsat Spectral Trajectories

Abstract:

Natural and anthropogenic forest disturbances are a major contributor to global carbon fluxes, and can act as both a natural component of healthy ecosystem function and a threat to fragile ecosystems as well as human lives and property. The local and global impacts of a disturbance event depend in large part on the timing, intensity, and cause of the disturbance. With some disturbance types expected to increase in frequency and severity under the influence of climate change and increasing anthropogenic land use, knowledge of disturbance events and trends has become particularly crucial for a variety of scientific, political and management needs. Many studies have made use of time series analyses of multitemporal satellite imagery to study forest disturbance events at a variety of scales, but few have endeavored to attribute specific causal information to the disturbances detected, particularly at broad spatial and temporal scales. The purpose of this research is to investigate the suitability of a Landsat time series approach for detecting and describing causes of disturbance events across the heterogeneous, forested landscapes of California. Using existing GIS datasets detailing the locations of logging, fire, pest damage and land use conversion events statewide, we extracted the full Landsat time series (1984-2015) for six Landsat spectral bands at the location of each disturbance event. The characteristics of each time series vary depending on the nature of the disturbance occurring at that location. These differences can be harnessed as a way of differentiating disturbance types based on observed values before, during, and after the time of the event, which together comprise the disturbance’s temporal signature. We applied a machine learning algorithm to these temporal signatures in order to construct a classification model linking disturbance type with the associated signature. We then applied this model to all forested areas in California using all Landsat 4, 5, 7 and 8 imagery available, generating a statewide thirty-year record of disturbance timing and cause. The resulting product, which can be expanded as new data becomes available, provides for land managers, researchers, and other interested parties a consistent and comprehensive source of information on disturbance events across California.