Map Skepticism: Recognizing Map Limitations
By Wayne E. Thogmartin
LA CROSSE, WI – Maps are easily-manipulated models of
reality. As such, they are never entirely true. Digital land cover maps are fast becoming one of the most
readily-used data sources in ecological research and natural resource
management. However, processes for accurate land cover
classification over large regions are still evolving.
At the Upper Midwest Environmental Sciences Center, we identified
inconsistencies in the patterning of land covers in the National Land Cover
Data 1990, the most current and widely available digital representation of
land cover in the conterminous United States.
In a geographic information system, we placed a regular
lattice structure over the digital land use map of the upper midwestern
United States, and then summarized for each cell of the lattice the
proportion of individual land cover types contained within. These proportions were then placed onto the lattice
and the resultant lattice was compared to Landsat Thematic Mapper (TM)
paths, state boundaries, and regional map classification units.
The inconsistencies we identified in the mapping of land cover
were particularly related to grassland/herbaceous and emergent herbaceous
wetland land covers (Figures 1 and 2),
both rare land covers (each <3% of land use) in the upper Midwest. We attributed these inconsistencies
to differences between TM image dates, especially as they reflect differences
in plant phenology, processing of mapping classification regions, and image
interpreters. For our purposes of modeling
regional populations of birds, we found these internal inconsistencies
across the region were more than we could allow in our statistical applications. Thus, we combined grassland/herbaceous and pasture/hay
land uses to minimize problems associated with the mapping discrepancies
we identified. To minimize the influence of emergent
herbaceous land cover, we identified a surrogate of ephemeral wetlands
on the basis of a static wetness index.
Because current accuracy assessments do not consider patterns in land cover, these sorts of internal inconsistencies can be missed and thus influence analyses involving rare land covers. We recommend elucidating patterns in digital land cover data in relation to TM path, state boundaries, and map classification region for other areas before their use in regional habitat evaluation, especially if rare land cover is of particular interest. We intend for further details of this research to be published shortly.
Figure 1. Association
between Thematic Mapper (TM) path and summarizations of land use for 800-ha
cells in the upper midwestern United States as determined from the National
Land Cover Data 1990. Note the abrupt decrease
in mapping frequency moving from TM path 22 to TM path 21.
Figure 2. Seven-fold difference in mapping frequency of emergent herbaceous wetland between Minnesota and Wisconsin as determined from the National Land Cover Data 1990.