5. Remapping

Remapping is the process of mapping satellite data to a uniform grid. Mapping data to a uniform grid makes it easier to view, manipulate, and store the data. Some instrument data is provided to the user already gridded (ex. VIIRS EDR Flood, ABI L1B data) and others are not (ex. VIIRS SDR or older GOES satellites).

In Polar2Grid it is possible to perform the gridding (reprojecting) process for ungridded data or to re-project already gridded data. Mapping input data in order to create a high quality image can be a complicated process. There are different techniques that can be used to create an output image based on what grid (projection) is chosen and what algorithm is used to map input pixel to output pixel. Polar2Grid offers various options that are described below. Defaults are also configured to provide a good result without any customization necessary.

5.1. Native Resampling

Native resampling is a special type of resampling that keeps input data in its original projection, but replicates or averages data when necessary to make other processing in Polar2Grid easier. Native resampling is the default for all data that is already gridded (ABI, AHI, etc) or when a native grid is specified by the user on the command line (-g MIN). It can also be specified on the command line by using --method native. See the Command Line Arguments section below for more details and the options available.

5.2. Elliptical Weighted Averaging Resampling

Elliptical Weighted Averaging (EWA) resampling is the default resampling method for a lot of scan-based polar-orbiting instrument data. This method uses the size of each instrument scan to determine a weight for each pixel. All input pixels that map to output pixels are weighted and averaged. This helps produce an image that is typically higher quality than those produced by nearest neighbor. It fits an ellipse to the data in the two axes based upon the --weight-delta-max and the --weight-distance-max options and then filters the texture with a Gaussian filter function. It can be specified on the command line by using --method ewa.

5.3. Nearest Neighbor Resampling

Nearest neighbor resampling is the most basic form of resampling when gridding data to another grid. This type of resampling will find the nearest valid input pixel for each pixel in the output image. If a valid pixel can’t be found near a location then an invalid (transparent) pixel is put in its place. Controlling this search distance and other options are described below in the Command Line Arguments section. Nearest neighbor resampling can be specified on the command line with --method nearest and is the default when non-native grids are specified to the command line (-g my_grid) for gridded data or if polar-orbiting instrument data is not scan-based (required for EWA).

Note that nearest neighbor resampling can cache intermediate calculations to files on disk when the same grid is used. For example, the calculations required to resample ABI L1B data to the same output grid for each time step are the same. If a directory is specified with the --cache-dir command line flag, this can greatly improve performance. This has no benefit for polar-orbiting swath-based data.

5.4. Grids

Polar2Grid uses the idea of “grids” to define the output geographic location that images will be remapped to. Grids are also known as “areas” in the SatPy library. These terms may be used interchangeably through this documentation, especially in low-level parts.

Polar2Grid uses grids defined by a PROJ.4 projection specification. Other parameters that define a grid like its width and height can be determined dynamically during this step. A grid is defined by the following parameters:

  • Grid Name

  • PROJ.4 String (either lat/lon or metered projection space)

  • Width (number of pixels in the X direction)

  • Height (number of pixels in the Y direction)

  • Cell Width (pixel size in the X direction in grid units)

  • Cell Height (pixel size in the Y direction in grid units)

  • X Origin (upper-left X coordinate in grid units)

  • Y Origin (upper-left Y coordinate in grid units)

Polar2Grid supports static and dynamic grids. Grids are static if they have all of the above attributes defined. Grids are dynamic if some of the attributes are not defined. These attributes are then computed at run time based on the data being remapped. Only width/height and x/y origin can be unspecified in dynamic grids. SatPy areas are also supported by Polar2Grid, but must be specified in SatPy’s typical “areas.yaml” file.

For information on defining your own custom grids see the Custom Grid documentation.

5.5. Remapping and Grid Command Line Arguments

usage: polar2grid.sh -r <reader> -w <writer> [-h] [--method {ewa,native,nearest}] [-g [GRIDS ...]] [--weight-delta-max WEIGHT_DELTA_MAX] [--weight-distance-max WEIGHT_DISTANCE_MAX] [--maximum-weight-mode] [--rows-per-scan ROWS_PER_SCAN]
                                             [--grid-coverage GRID_COVERAGE] [--cache-dir CACHE_DIR] [--grid-configs GRID_CONFIGS [GRID_CONFIGS ...]] [--antimeridian-mode {modify_extents,modify_crs,global_extents}] [--radius-of-influence RADIUS_OF_INFLUENCE]

5.5.1. Resampling

--method

Possible choices: ewa, native, nearest

resampling algorithm to use (default: <sensor specific>)

-g, --grids

Area definition to resample to. Empty means no resampling (default: “wgs84_fit” for non-native resampling)

--weight-delta-max

Maximum distance in grid cells over which to distribute an input swath pixel (–method “ewa”). This is equivalent to the old “–fornav-D” flag. Default is 10.0.

--weight-distance-max

Distance in grid cell units at which to apply a minimum weight. (–method “ewa”). This is equivalent to the old “–fornav-d” flag. Default is 1.0.

--maximum-weight-mode

Use maximum weight mode (–method “ewa”). Default is off.

--rows-per-scan

Number of data rows making up one instrument scan. (–method “ewa”). Defaults to value extracted from reader.

--grid-coverage

Fraction of target grid that must contain data to continue processing product.

Default: 0.1

--cache-dir

Directory to store resampling intermediate results between executions. Not used with native resampling or resampling of ungridded or swath data.

--grid-configs

Specify additional grid configuration files. (.conf for legacy CSV grids, .yaml for SatPy-style areas)

Default: ()

--antimeridian-mode

Possible choices: modify_extents, modify_crs, global_extents

Behavior when dynamic grids are converted to ‘frozen’ grids and data crosses the anti-meridian. Defaults to ‘modify_crs’ where the prime meridian is shifted 180 degrees to make the result one contiguous coordinate space. ‘modify_extents’ will attempt to surround the data but will often cause artifacts over the antimeridian. ‘global_extents’ will force the X extents to -180 and 180 to create one large grid. This currently only affects lon/lat projections.

Default: “modify_crs”

--radius-of-influence

Specify radius to search for valid input pixels for nearest neighbor resampling (–method “nearest”). Value is in geocentric meters regardless of input or output projection. By default this will be estimated based on input and output projection and pixel size.