Global Bare Ground Gain in the First Decade of 21st Century from Landsat Data: the Preliminary Results on Estimation and Distribution

Abstract:

Bare ground gain (BGG) is one of the most intensive land surface transformation because of the complete alteration of ecosystem functioning, the complex nature of temporal nonlinearity and spatial heterogeneity, the fast growing trend along with the global population boom and urbanization. However, it is not yet clear that what are the global dynamics of BGG and how it is spatially distributed. It is therefore important to monitor BGG as an essential component of land cover change on a locally relevant and globally consistent base. In this study, we try to answer these questions using over-a-decade Landsat satellite observations.

Recent developments in optical remote sensing hold tremendous promise for global BGG detection. One data source is the legacy of annual Landsat mosaics from the research of Hansen et al. on global forest dynamics. Following previous research by Hansen et al., BGG observed by Landsat data is defined as a process of land cover change featuring permanent or semi-permanent clearing of vegetation cover by human land use or natural disturbances at the 30-m Landsat pixel scale. A sophisticated method has been developed to capture the change signal from high dimension metrics derived from time series of Landsat spectral bands and continuous bare ground field. By examining the contribution of each metric to the effectiveness of BGG detection, 140 metrics were selected and put into a supervised machine learning algorithm, the bagged classification tree. A recursive strategy was adopted to complete training data and improve result. A global BGG training data set counting to around 27.5 million pixels was produced. Additional training was obtained from regional sources like the bare ground gain layer of Web-enabled Landsat data (WELD) and the impervious surface layer of National Land Cover Database (NLCD). Independent validation was performed by interpreting stratified samples on Google Earth high resolution images. The preliminary results of the first global BGG map will be presented at the meeting with accuracy estimates.Bare ground gain (BGG) is one of the most intensive land surface transformation because of the complete alteration of ecosystem functioning, the complex nature of temporal nonlinearity and spatial heterogeneity, the fast growing trend along with the global population boom and urbanization. However, it is not yet clear that what are the global dynamics of BGG and how it is spatially distributed. It is therefore important to monitor BGG as an essential component of land cover change on a locally relevant and globally consistent base. In this study, we try to answer these questions using over-a-decade Landsat satellite observations.

Recent developments in optical remote sensing hold tremendous promise for global BGG detection. One data source is the legacy of annual Landsat mosaics from the research of Hansen et al. on global forest dynamics. Following previous research by Hansen et al., BGG observed by Landsat data is defined as a process of land cover change featuring permanent or semi-permanent clearing of vegetation cover by human land use or natural disturbances at the 30-m Landsat pixel scale. A sophisticated method has been developed to capture the change signal from high dimension metrics derived from time series of Landsat spectral bands and continuous bare ground field. By examining the contribution of each metric to the effectiveness of BGG detection, 140 metrics were selected and put into a supervised machine learning algorithm, the bagged classification tree. A recursive strategy was adopted to complete training data and improve result. A global BGG training data set counting to around 27.5 million pixels was produced. Additional training was obtained from regional sources like the bare ground gain layer of Web-enabled Landsat data (WELD) and the impervious surface layer of National Land Cover Database (NLCD). Independent validation was performed by interpreting stratified samples on Google Earth high resolution images. The preliminary results of the first global BGG map will be presented at the meeting with accuracy estimates.