DensRas: Raster densifying and dedensifying (resampling)

Access this help text as a web page: DensRas

Presentation and options	Dialog box of the application
Syntax

Presentation and options

This application changes the cell side of a raster (densifies or dedensifies it, as appropriate), without exceeding the bounding box of the original raster, and allowing, if necessary, a change of the origin grid. The procedure operates using one of the following four resampling methods: nearest neighbor, 4-neighbor mode, bilinear interpolation, or bicubic interpolation. In these last two options (bilinear interpolation and bicubic interpolation), when the value resulting from the resampling is a value outside the range of values of the original raster (lower than the minimum value or higher than the maximum value of the original raster), the user is allowed to select between to saturate the resampled value to the minimum value of the original raster (or maximum as appropriate) or assign NoData. On the other hand, the program makes a careful treatment of the NoData value (it does not consider it in the resampling: for example, if one of the 4 neighbors at the position of a cell in the new raster is NoData and the criterion is the mode, this mode will be calculated considering only the 3 neighbors with data).

In order to change the cell size of a raster without changing the grid origin, use the CanviRes program. For more sophisticated operations, and particularly to expand the geographic extent by filling it with NoData, use the program AdapRas. At the end of this text there is a summary explaining when CanviRes should be used and when DensRas should be used, including the options for each one.

It must be considered that if the cell side change is a factor greater than 2, the mode created may be unsatisfactory because it is not using all the original cells that fall geometrically on the final cells, but only the 4 central ones. In these cases, it is preferable to approach the desired cell size using CanviRes before using DensRas. This "defect" is also present, by definition, when bilinear (or bicubic) interpolations are performed. In this case, you can proceed by progressive increments of cell size in factors of 2. However, if the user prefers each cell of the new raster to have a value calculated from the original values closest to its center, it is correct to apply the DensRas directly. This would be the case if the user has a Digital Elevation Model (DEM) of 2 m cell side and wants to obtain a new DEM with cells of 20 m side but in which the value in each cell is not calculated (operation that would be carried out with CanviRes) with the median or mean of the 100 2x2 m² cells corresponding to each new 20x20 m² cell, but rather it is calculated with bilinear interpolation of the 4 2x2 m² cells closest to the center of the new cell, or with the bicubic interpolation of the 16 2x2 m² cells closest to the center of the new cell. In this case, the data from the cells of the new raster will report the elevations in the center of the 20x20 m² cell, while if it had been done with CanviRes they would report a more representative data (by the median or mean) of all elevations of the 20x20 m² cell. It is important to realize that:

One method is not better than the other: Depending on the objective of changing the raster cell size, CanviRes (values are more representative of the cell surface) or DensRas (values more representative of the center of the cell) may be more convenient, and the values of cells in the output raster may be quite different from those located at approximately the same cell centers in the original raster.
If the difference between input and output cell sizes is small or non-existent (such as when the user only wants to change the origin grid), using one method or the other will not imply large differences between the values of the output raster cells and those located at approximately the same cell centers in the original raster.

In the process of densification (or dedensification) of a raster, with the same bounding box as the original raster, the treatment that is carried out on the edges during resampling is, in reality, a simplification, due to the non existence of cell information beyond the bounding box. Therefore, it is not convenient to directly connect adjacent tiles resulting from a densification, but it is necessary to have a certain extra area around the desired one in order to be fully reliable. In other words, when the user densifies exactly between the original raster coordinates on a tessellation and does the same procedure on a neighboring tessellation that does not have data in common geographic areas (i.e., no overlap between tiles), the application of algorithms related to hydrological flows may cause inaccuracies or calculation artifacts in the border areas between tiles once the tiles are mosaicked; in these cases the ideal is to have, for each tile, an extra "concentric" strip (all the way around), do the resampling and finally, before making the mosaic of the tiles to obtain a single work area, cut each tile to exclude the extra area and that they are again "exactly touching, without overlapping" (but having been calculated with an extra area for more accuracy of resampling):

If bilinear interpolation is to be used, it is recommended to extend a 1-cell strip to the entire margin of the original raster, run DensRas and finally remove this 1-cell strip.

If bicubic interpolation has been used, it is recommended to proceed in a similar way but in this case extending (and removing at the end) a strip of 2 cells to the entire margin of the raster.

RECOMMENDATIONS SUMMARY (where DensRas program is indicated also applies to AdapRas)

EXPANSION:  + Categorical map:   CanviRes [followed by the method in X. Pons' Ph D thesis]
            + Sensor image:      CanviRes for classification (not recommended)
                                 DensRas L, for visual analysis
            + DEM:               DensRas L

CONTRACTION:+ Categorical map:   CanviRes, M option
            + Sensor image:      CanviRes, C option for classification
                                 CanviRes, d or m option for visual analysis
            + DEM:               CanviRes, d or m option for median or mean heights
                                 DensRas L, for heights that prioritize central values in the resulting cell

Dialog box of the application

DensRas dialog box

Syntax

Syntax:

DensRas OriginalRaster DensifiedRaster Xmin Xmax Ymin Ymax Spix Resampling

Parameters:

OriginalRaster (Original raster - Input parameter): Name of the input raster file to densifier.
DensifiedRaster (Densified raster - Output parameter): Name of the densified raster file. If not indicated, it is assumed that both files have img extension.
Xmin (Minimum X - Input parameter): Define the minimum X of the densified raster bounding box.
Xmax (Maximum X - Input parameter): Define the maximum X of the densified raster bounding box.
Ymin (Minimum Y - Input parameter): Define the minimum Y of the densified raster bounding box.
Spix (Cell side - Input parameter): Cell side or pixel of the densified raster. If Xmax does not correspond to an integer value n such that Xmax=Xmin+n*Cpix, Xmax will be adapted accordingly. Similarly for Ymax.
Resampling (Resampling mode - Input parameter): Points the resampling mode: 'V' for the nearest neighbor, 'M' for the 4 neighbors mode, 'L' for bilineal interpolation and 'C' for bicubic interpolation interpolation.