H13S-06
AN AUTOMATED ALGORITHM FOR PRODUCING LAND COVER INFORMATION FROM LANDSAT SURFACE REFLECTANCE DATA ACQUIRED BETWEEN 1984 AND PRESENT
Monday, 14 December 2015: 14:55
3020 (Moscone West)
Jennifer Rover1, Martin B Goldhaber2, Cheryl Holen3, Ray Dittmeier3, Steve Wika3, Daniel Steinwand1, Devendra Dahal3, Brian Tolk3, Robert Quenzer4, Kurtis Nelson1, Bruce K Wylie5 and Michael Coan1, (1)USGS Earth Resources Observation and Science Center, Sioux Falls, SD, United States, (2)USGS-Denver Federal Center, Denver, CO, United States, (3)Stinger Ghaffarian Technologies (SGT, Inc.), Sioux Falls, SD, United States, (4)Organization Not Listed, Washington, DC, United States, (5)USGS, EROS Data Center, Baltimore, MD, United States
Abstract:
Multi-year land cover mapping from remotely sensed data poses challenges. Producing land cover products at spatial and temporal scales required for assessing longer-term trends in land cover change are typically a resource-limited process. A recently developed approach utilizes open source software libraries to automatically generate datasets, decision tree classifications, and data products while requiring minimal user interaction. Users are only required to supply coordinates for an area of interest, land cover from an existing source such as National Land Cover Database and percent slope from a digital terrain model for the same area of interest, two target acquisition year-day windows, and the years of interest between 1984 and present. The algorithm queries the Landsat archive for Landsat data intersecting the area and dates of interest. Cloud-free pixels meeting the user’s criteria are mosaicked to create composite images for training the classifiers and applying the classifiers. Stratification of training data is determined by the user and redefined during an iterative process of reviewing classifiers and resulting predictions. The algorithm outputs include yearly land cover raster format data, graphics, and supporting databases for further analysis. Additional analytical tools are also incorporated into the automated land cover system and enable statistical analysis after data are generated. Applications tested include the impact of land cover change and water permanence. For example, land cover conversions in areas where shrubland and grassland were replaced by shale oil pads during hydrofracking of the Bakken Formation were quantified. Analytical analysis of spatial and temporal changes in surface water included identifying wetlands in the Prairie Pothole Region of North Dakota with potential connectivity to ground water, indicating subsurface permeability and geochemistry.
