| Reprojection |
|
Some helpful additional concepts for Reprojecting imagery
There is not a one-to-one correspondance between the pixels of the input file and those of the output reprojected file. To handle this, the Reproject tool masks out invalid pixels and ultimately reprojects only the valid pixels of the source file. This hidden automated pixel validation step is acheived using the following two criteria: the valid pixel expression and the designation of a no-data value. You can view/modify this criteria by right-clicking on any band, and selecting Properties. Alternatively you can view this with the Band Info tool (Analysis > Band Info) when selecting the band.
In general this is the desired behavior, unless the valid pixel expression is masking on an ideal range, outside of which there may actually be valid data. If you wish to include this out-of-range data in the reprojection then you will need to edit the valid pixel expression of the band(s) of the source file before reprojecting. One example of this would be when creating true color imagery using surface reflectances. Because the surface reflectance product is designed based on water leaving radiances for the ocean, pixels which image cloud tops can exceed a reflectance value of 1. In this case, for the case of including bright clouds in your true color imagery, the valid pixel expression band(s) of the source file would need to be modified so as to not throw away values above 1.
In general, it may be best to consider mapping a file one way for statistics and another way for displaying data. For statistics you are best served using an equal area map projection (such as Albers Conic Equal Area, Lambert Azimuthal Equal Area, Sinusoidal, ...). With an equal area projection, the Earth surface area is the same for each resultant pixel and can be treated as such with an equal per pixel weighting. However, for large scenes, an equal area map projection may not look natural so using a conformal projection (such as Mercator and Stereographic) is more ideal when displaying this type of imagery.
Note: selection of the "best" map projection is dependent on scene size (is it global, regional, local), scene shape (does it run narrow north-south, narrow west-east, or is it more square in shape), and scene location (is it at the equator, at one of the Poles, or some span of latitudes soemwhere between the Pole and equator.
Upsampling is the process of resampling (remapping) your data to a higher resolution. Upsampling has no inherent problems, other than the fact that the resultant file will be increasingly larger with higher resolutions. A remapped upsampled image will tend to be clearer than one remapped at a resolution comparable with the original input file. The higher the resampling resolution, the closer you approach to seeing the pixel shapes of the original unprojected file.
Note: Ideally a remapped image should not be displayed at a zoom level such that it displays the pixellation of the output file. Such a visual might cause a viewer to incorrectly infer a measurement of noise (or lack of noise) pixel to pixel, when in fact these pixels do not bear a one-to-one correspondance with the native satellite detector measurements.
Downsampling is the process of resampling (remapping) your data to a lower resolution. Downsampling by more than a factor of 4 does have some inherent problems, at least in the SeaDAS Reproject and its related GUI tools (see note 1). This is because SeaDAS currently only offers 3 interpolation methods (nearest neighbor, bilinear, and bicubic) which use a matrix size on the input file of 1x1, 2x2, and 4x4 respectively to generate each remapped pixel of the output file. (See method details.) Downsampling by greater factors than these matrix sizes, will result in many of the pixels in the source file being ignored. Basically, in order to ensure that all source data pixels are being use, if you are downsampling by a factor of 4 then use bicubic, a factor of 2 then use bilinear or bicubic, and a factor of 1 then use any of the three.
Note 1: The resampling(remapping) concept and process is the same in all of the SeaDAS visualization resampling tools: Reproject, Collocate, Mosaic, and Time Series.
Note 2: The SeaDAS OCSSW tools for binning and mapping behave differently and do not ignore any source data pixels for large factor downsampling