Skip to end of metadata
Go to start of metadata

Community for Data Integration - Technology Stack Working Group

Use Cases

Land Cover Trends Research Project

Use Case – Perform Change Analysis

Note

  • This use case was written with the assumption that GDP was going to be used to support the workflow defined. This does not need to be the case--other technologies could be used to replace GDP if desired.
  • GDP should probably be thought of as an "working prototype." It should not be considered an operational tool. If CDI or others want to move GDP to this next level, then planning (time, people, funding) for this should occur.
  • Most of the following assumes little or no change to currently existing GDP functionality or even configuration. To support the workflow, Jeanne needs little or no interaction with the GDP development team. I (rviger) see this as a very positive characteristics of this use-case.

Summary

The Land Cover Trends (LCT) research project examines the rates, causes, and consequences of contemporary U.S. land-use and land-cover change. The strategy involves classification of sample blocks of Landsat imagery into eleven land-cover classes. The blocks are chosen from a random sampling of each of the eighty-four Level III ecoregions in the conterminous U.S. The classification is performed on the same spatial blocks for five time periods: 1973, 1980, 1986, 1992, and 2000. The temporal series of land-cover classifications is then analyzed for each block to identify and quantify any land-cover changes occurring over time. Change statistics that result from this analysis form the basis of work documenting current change and projecting future land-cover changes.

Data

            Data for Trends analysis come from change grids previously generated from the land cover classification products for the study years. This derived product consists of categorical data where each cell contains a value between 1 and 121. These categories represent the “from-to” land cover conversions that occurred for each cell. The study is based on a total of 2688 sample blocks analyzed for the four change periods: 1973 to 1980, 1980 to 1986, 1986 to 1992, and 1992 to 2000. This results in approximately 11,000 change grids, and each grid is approximately 15k bytes in size. Trends also uses about 200 polygon shapefiles that define the boundaries of blocks and ecoregions. The Trends analysis will utilize the GeoData Portal (GDP) to retrieve the change grids and shapefiles needed for the statistics. In addition, access through the GeoData Portal to a selection of shapefiles of regions, states, watersheds, etc. would provide a range of choice for the researcher interested in defining a study area.

Workflow

            A researcher who is interested in obtaining land-cover change statistics for a given study area accesses the Trends analysis through a web-based portal that provides simple GIS capabilities. The LCT portal will use GDP functionality according to the following use-case.

  •   The researcher selects or supplies a “study area” boundary, a watershed boundary for example. This could be from a user-specified shapefile or by querying a WFS-accessible store of polygons.
    • This content (i.e. the shapes) is provided by or is specified by the user.
  • Determine which Level III ecoregions the study area covers.
    • This content is provided by the LCT portal.
    • Although the Land Cover Trends (LCT) project currently stores the ecoregions in shapefiles, they could convert these to NetCDF-CF format rasters and serve via THREDDS in order to let GDP do this work.
    • Returned: List of the identification number of each ecoregion within the study area.
    • Returned: simple feature geometry delineating each study area-ecoregion intersection.
  • Within each study area-ecoregion intersection, determine total number of all blocks. A block is “inside” if 51% of its area is within the study area-ecoregion footprint.
    • The shapefiles of blocks could be converted to raster and made available via THREDDS to help using GDP (as was done with the ecoregion map).
    • GDP could do this. GDP can already figure out the percentage area of other data within a study area. A little extra work might be needed to apply the 51% threshold.
    • Returned: A per-ecoregion list of the number of blocks within the study area. 
  • Within each study area-ecoregion intersection, determine total number and identity of sample blocks. A block is “inside” if 51% of its area is within the study area-ecoregion footprint.
    • See first two sub-bullets on previous bullet.
    • Returned: A per-ecoregion list of the sample block identifiers and corresponding area.
  • LCT processing service runs an SQL query that asks for the per-sample block table of results.
    • Not GDP.
    • Could LCT project build/convert into such a database? Sounds like it.
  • Statistics are generated for each ecoregion by LCT processing service.
  • These statistics are then displayed as tables and graphs in report form. This would also be provided by the LCT processing service.

One possible view of user getting per-study area (in shapefile) summary of land cover changes using methods and data provided by the Land Cover Trends project.

Results

            The tables and graphs returned to the researcher at the end of the analysis by the LCT portal. The portal would provide change statistics on land cover within the study area.  For each of the “from-to” land cover conversions such as “from agriculture to developed” or “from forest to wetland”, the researcher can view estimates of the total change and variance and the change and variance as percent of study area.  Relative error and confidence intervals provide additional information on the estimates.  

Future Enhancements

            Current Trends analysis uses a stratified sampling approach, utilizing full sample blocks in the statistics even if some percentage of the block falls outside the study area.  A future enhancement would clip or mask the change grids that fall on study area boundaries and perform a recount of conversions based on the masked grid.  The analysis function would request the full grid for the sample block from the GeoData Portal and perform the masking and recalculation unless the Portal provided this functionality.             Also, eventually a Trends web-based GIS would provide access to the complete Trends data set, comprised of roughly 150,000 satellite images, land cover and conversion grids, aerial photographs, videos, and reports.  The Trends database, acting as a metadata catalog for the archive, would be available for searches and would provide the names of the files to the GeoData Portal for retrieval.  These files could either be displayed at the Trends GIS website or made available for download.   Jeanne Jones & Roland Viger
August 2010