thumbnail

Methods for converting continuous shrubland ecosystem component values to thematic National Land Cover Database classes

Open-File Report 2017-1119
By: , , , and 
https://doi.org/10.3133/ofr20171119

Links

Abstract

The National Land Cover Database (NLCD) provides thematic land cover and land cover change data at 30-meter spatial resolution for the United States. Although the NLCD is considered to be the leading thematic land cover/land use product and overall classification accuracy across the NLCD is high, performance and consistency in the vast shrub and grasslands of the Western United States is lower than desired. To address these issues and fulfill the needs of stakeholders requiring more accurate rangeland data, the USGS has developed a method to quantify these areas in terms of the continuous cover of several cover components. These components include the cover of shrub, sagebrush (Artemisia spp), big sagebrush (Artemisia tridentata spp.), herbaceous, annual herbaceous, litter, and bare ground, and shrub and sagebrush height. To produce maps of component cover, we collected field data that were then associated with spectral values in WorldView-2 and Landsat imagery using regression tree models. The current report outlines the procedures and results of converting these continuous cover components to three thematic NLCD classes: barren, shrubland, and grassland. To accomplish this, we developed a series of indices and conditional models using continuous cover of shrub, bare ground, herbaceous, and litter as inputs. The continuous cover data are currently available for two large regions in the Western United States. Accuracy of the “cross-walked” product was assessed relative to that of NLCD 2011 at independent validation points (n=787) across these two regions. Overall thematic accuracy of the “cross-walked” product was 0.70, compared to 0.63 for NLCD 2011. The kappa value was considerably higher for the “cross-walked” product at 0.41 compared to 0.28 for NLCD 2011. Accuracy was also evaluated relative to the values of training points (n=75,000) used in the development of the continuous cover components. Again, the “cross-walked” product outperformed NLCD 2011, with an overall accuracy of 0.81, compared to 0.66 for NLCD 2011. These results demonstrated that our continuous cover predictions and models were successful in increasing thematic classification accuracy in Western United States shrublands. We plan to directly use the “cross-walked” product, where available, in the NLCD 2016 product.

Suggested Citation

Rigge, M.B., Gass, Leila, Homer, C.G., and Xian, G.Z., 2017, Methods for converting continuous shrubland ecosystem component values to thematic National Land Cover Database classes: U.S. Geological Survey Open-File Report 2017–1119, 10 p., https://doi.org/10.3133/ofr20171119.

ISSN: 2331-1258 (online)

Table of Contents

  • Abstract
  • Introduction
  • Section I—Modelling Thematic Land Cover from Continuous Cover Classes
  • Section II—Preparation of Validation Data
  • Section III—Accuracy Assessment
  • Conclusion
  • Summary
  • Acknowledgments
  • References Cited
Publication type Report
Publication Subtype USGS Numbered Series
Title Methods for converting continuous shrubland ecosystem component values to thematic National Land Cover Database classes
Series title Open-File Report
Series number 2017-1119
DOI 10.3133/ofr20171119
Year Published 2017
Language English
Publisher U.S. Geological Survey
Publisher location Reston, VA
Contributing office(s) Earth Resources Observation and Science (EROS) Center
Description iv, 10 p.
Online Only (Y/N) Y
Google Analytic Metrics Metrics page
Additional publication details