Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Etats Unis → 2022 → Using Time Series Segmentation to Generate a Novel Land Cover Classification for Xilinhot, Inner Mongolia

George Washington University (2022)

Using Time Series Segmentation to Generate a Novel Land Cover Classification for Xilinhot, Inner Mongolia

Murray, Mia

Titre : Using Time Series Segmentation to Generate a Novel Land Cover Classification for Xilinhot, Inner Mongolia

Auteur : Murray, Mia

Université de soutenance : George Washington University

Grade : Master of Science (MS) 2022

Résumé
Current modeling of land use and land cover of rangelands is limited by shortcomings in the ontological and methodological approaches to classification. For example, current classification schemes often limit rangeland cover types into one of a small number of categories, typically grassland, shrubland, or barren, and change is only measured when a pixel shifts to a new class. These coarse categories can mask the complex dynamics happening within a given land cover type, such as greening or browning that might occur prior to a state shift. A more nuanced classification of grasslands that represents the dynamics occurring within a class would contribute meaningfully to policy and management. In this study, we assess a novel method for classifying land cover change, focused on Xilinhot, a provincial city in central Inner Mongolia, located in one of the largest remaining arid grassland regions of the world. The region has faced rapid development and land use conversion in recent decades and grassland condition is declining in many areas, but the mechanisms of these changes are difficult to assess with traditional change detection approaches.In this study we apply the LandTrendr segmentation algorithm to a time series of Landsat-derived imagery of vegetation condition to capture changes in 30+ years of cover in Xilinhot. After accounting for effects of climate, we then classify the resulting segmented data within an unsupervised k-means clustering scheme. We identified 3 unique classes in the data ; these classes represent unique pixel trajectories resulting from past changes in management and land use. These data on the influence of past intervention on pixel-level responses can be used to help build predictive models to better understand the potential impact of future policy and management changes

Présentation

Version intégrale

Page publiée le 18 décembre 2022