Land disturbance characterization based on Landsat time series

Abstract:

Land disturbances, such as harvest, mechanical, stress, wind, hydrology, debris, fire, and climatic variability, occur on the Earth’s surface with high temporal and spatial variability. It is a key component of ecological systems and influences terrestrial ecosystem at a wide range of scales. However, most of the efforts in remote sensing community are only focused on mapping land cover and land use change - the target of change, with very limited studies on land disturbance - the agent of change. Here, we will introduce a land disturbance agent classification approach for all land areas across the Conterminous US (CONUS) based on Landsat time series. First, the COntinuous monitoring of Land Disturbance (COLD) algorithm will be used to detect disturbances (with unknown agent). Second, high-quality disturbance agent training sample data will be established based on existing open-source datasets. Third, a Random Forest (RF) model will be created to classify various kinds of disturbance agents based on an object-based approach. The purpose of this approach is to produce annual land disturbance agent products over the CONUS at 30-meter resolution.