When reprojecting data from the original reference grid to a new projection, cells are relocated, the grid cells values need to be re-assigned values using resampling. The three industry standard methods found in MapInfo Pro Advance are Nearest Neighbour, Bilinear and Bicubic or Cubic Convolution. Each has it's own merit's and are best suited to certain data types and end applications. Understanding the end result and the application are key to determining which to use.
Nearest Neighbour will assign each values from the original cells to the new cells, which in many cases can be beneficial, especially for applications where it is required to keep the original cell values unaltered such as categorical grids. However, positional errors can be noticeable as the nearest point on the reference (original)grid's value is assigned to the cell.
For example, the blue points below show the original reference grid point locations, the green points represent the new cell centre points that will make up the new output image. As the values are assigned from the nearest located data point, the type of positional changes that could occur can be seen clearly in the illustration below. As the value is assigned from the blue reference grid to the new green output grid the location changes for this value. The value is however is exactly the same as it was previously, in linear data it may sometimes appear that the new grid is blocky or jagged (as illustrated with the line below) as the values may change along linear features.
Nearest NeighbourSee Attachment
With Bilinear, the reference grid points (blue) have their values assigned to them from the nearest four surrounding points using a distance weighted average. The closer the data point the higher the influence on the newly assigned value. In this method the new value will have altered from the value in the original reference grid. The range of values in the surrounding nearest points will influence how great the difference between the original reference value
Using cubic convolution, the new cell value is assigned from a distance weighted average of the data values found within the area of the nearest 16 cells. The images or grids created using this method are generally more visually appealing or ‘smooth’ in appearance, however the values will likely have been altered more drastically than the other methods. This is however, dependant on the range of data point values within the surrounding 16 cells.
Understanding how the methods work and understanding shifts, rotation and scaling changes between projections will give clues to as which is the best method to use for each application.