National Geochemical Survey - Gridded Maps

All images that are part of this report showing the distribution of chemical elements were generated using the following method:
  1. Pooling of data. All data in the NGS database were pooled for the purpose of display in gridded maps. Data for stream sediments, soils, and other media were combined. The relative contributions of different media in different areas can be evaluated using this index map. A statistical comparison of different sample media will be done at a later time.

  2. Recalculation of data. Elemental data stored in the NGS may contain negative numbers signifying values that are upper limits. Thus "-10.0" represents a determination by the laboratory reported as "<10". These values were recalculated using the method of Grossman (1998). In this method, histograms are prepared, one showing the distribution of all data that were reported as unqualified (positive) values and one showing the distribution of all data reported as upper limits. From these histograms, a cutoff value was selected by visual inspection. Any values reported as upper limits less than or equal to this cutoff were replaced by a positive number with half the magnitude of the reported value. Continuing the example above, "-10.0" would be replaced with "5.0". Values reported as upper limits above this cutoff were deleted from the dataset.

    For most analytical methods in the NGS, detection limits are fairly constant. For a method with a constant detection limit of 1 ppm, values are either reported as positive numbers ≥1, or as an upper limit of "<1". In these cases, the selected cutoff value would be 1.0, with the effect that all of the upper limit values are converted to a value of 0.5. The INAA methods, however, have highly variable detection limits that are matrix-dependent. For such data, the cutoff was selected such that unusually high upper limits would not be converted into values greater than the 5th percentile of the distribution of unqualified determinations. A similar selection of cutoff values was done for NURE data, in which each element might contain values determined by a wide variety of analytical methods with different detection limits.

  3. Gridding. The recalculated data were used to prepare gridded maps. Grids were calculated using the Spatial Analyst extension to the the ArcView 3.2 program by ESRI. A grid-cell size of 3 km was used in the conterminous US and Alaska, and 2.5 km was used in Hawai`i. The "inverse distance weighted" (IDW) interpolation method was used, with a radius of 20 km and a power of 2.

  4. Grid display. A color-legend was created for each grid based on the distribution of the point data that was used to prepare the grid. Each of the 20 intervals in the legend represents 5 percentiles of this distribution. Note that this procedure results in somewhat different legends for the same element in each of the three regions for which maps were prepared.  The gray tones used in the first few percentile ranges are used for artistic purposes only, and are not meant to imply that the quality of data is poor.

  5. Creation of images. Each grid was displayed with its legend in an ArcView "layout," and exported as a jpeg image.

Most of the above procedure was automated using an unpublished ArcView extension written by J. Grossman. The MULTIGRID extension carried out the recalculation of data using the rules outlined above, calculated grids for all of the elements in each region, and did the statistical calculation of the intervals in the legends. In a second operation, MULTIGRID displayed each grid-layout in turn, and exported the images to jpeg files. This extension is available from the author upon request.

Please note:  These maps have been generated by a computer in an automated mode.  As a result, some problems that would be avoided in a "supervised" or manual procedure occur:  (1) there is absolutely no significance to the number of decimal places shown in the legends;  (2) for maps of elements with high detection limits, where a large percentage of the grid cells have an identical value, the automated routine to create a color legend did not work particularly well, producing many empty color slices (e.g., uranium measured by the ICP40 method).