Page History

...

• Application to Geographically Gridded data, to regridding of data without reprojection, and to general reprojection of projection-gridded data
• The impact of the rows-per-swath parameter and options of rows-per-swath = 1 and rows-per-swath = 0 => all rows … .
• New modules to replace ll2rc – lldim2rc (Geographic Grids), rc2rc (Regridding, no projection math), xy2rc (Projected Grids, Double projection, Source-to-Target)
• Application to Geographically Gridded data

  • Expand
    title More details...
    There is a relatively direct relevancy of the EWA algorithm to the reprojection of geographically gridded data.   The application of EWA in this case appears valid from a reprojection perspective. For geographically gridded data we have an array of data, with correlated latitude and longitude values.  It is easy to see how the algorithm would be applied.   When you look at a geographically gridded data array versus a swath data array, you can see that the geographically gridded data is generally wider than the swath, but there is no elliptical stretching of the grid cells in their geographic spacing.  The ellipse of influence for the source data is circular.   There remains, however, the non-circular ellipses of influence due to the reprojection to a non-geographic, projected grid.  This is properly handled by the EWA implementation. EWA should be a relatively efficient implementation relative to reverse projection techniques. The source-cell to target-cells mapping does not have the elliptical spread of the sample-spots to consider - only the elliptical spread of the projection mapping.  There should be no more, and likely fewer cells per source cell, and thus greater efficiency per source cell count than for a swath dataset. The algorithm has a gaussian filter built-in to mitigate down-sampling aliasing effects, where relevant in the output projected grid. In this case, the preprocessing step of EWA - the Latitude-Longitude-to-Row-Column calculation (ll2rc) can be handled in one of two ways The latitude/longitude arrays of values can be precomputed (if not directly available) and then fed into the ll2rc method Alternatively, the ll2rc method can be modified into a method that directly uses the 1D dimension-scales (dimension-variables) providing latitude and longitude values.  I would call such a method llDim2rc.

• Application to Regridding - without reprojection

  • Expand
    title More details...
    This is a curious and perhaps less clear "off-label" application of the EWA algorithm In this case the mapping of source cells to target cells is exactly circular, not elliptical.  Note however, that the EWA approach of looking at location-mapping deltas between adjacent row and column data still results in an appropriately defined "circle of influence" between the source and target grids. Note also that this particular application of EWA does not care if the source/target arrays are geographically gridded or projection gridded - only that the source and target grids have the same projection. Again the ll2rc method has some special handling considerations There is nothing preventing the use of the first approach outlined above for application to reprojecting Geographically Gridded data - excepting for the inefficiency and processing time required. I refer here to computing a pair of lat/lon datasets that can be fed into ll2rc to compute the mapping between source cells and target cells. This requires projecting the source cell locations to lat/lon, and then projecting that back to the target cell locations. As noted, ll2rc will work in this way but the double projection math involved is significant and ultimately unnecessary Alternatively, the ll2rc method can be replaced with a method to compute the relative locations of the source and target cells using the two grid definitions and without consideration of the grid-projection involved.  This can be done producing the same data results as the ll2rc method - a source to target mapping of row-column values, floating point results, i.e., to the target row-column space. I refer in this case to the grid definition as the combination of x and y projected locations of the corner points, in projected meters (or lat/lon for geographic grids); the number of rows and columns; and implicitly, the cell resolution in projected meters (degrees for geographic grids). Reprojection math can be replaced with affine-matrix math, as one approach I would call such a method rc2rc.

• Application to General Reprojection

  • Expand
    title More details...