Join Operation

On this page, you will find general help for the ProRaster product family including links to documentation, instructional videos, and training videos.
Previous Topic: Statistics Analysis Operation
Next Topic: Classify Operation
Back to: ProRaster Scientific Help
Back to: ProRaster Help
Back to: ProRaster
Go to: ProRaster Essential help page, ProRaster Premium help page, ProRaster Scientific help page.
The ProRaster User Guide is available for download as a PDF.

Watch this video on YouTube to learn how to use this processing operation. Use the chapter indexing to skip to the detail you need.

The Simple Join operation is accessed from the Raster Processing “Merge” menu and implements a simple and efficient raster merging operation, powered by virtual raster technology. 

Merging rasters is complicated by the plethora of different terminology used to describe the process, and by the fact that the definitions of these terms are not strictly defined or uniformly adhered to. Terms like Merge, Join, Stitch, Blend, Feather, Mosaic, Composite, and Collate are commonly used in different ways. For the record, this is how I define these terms in ProRaster. 

Join                      The simplest kind of merge, which combines two or more rasters into a single raster. The data values in all source rasters are copied without modification. If source rasters overlap, then a “painters” algorithm is employed, and the rasters that are “on top” replace the content of rasters below. The order of the source rasters is critical.

Merge                  A true merge will apply some modifications to data values in source rasters to attempt to ensure that values from different rasters are compatible. These modifications could involve both scaling and offset. The offsets may have a spatial dimension (a 1D offset, 2D planar offset, or 2D non-linear offset). The scaling could be linear or non-linear (e.g. histogram matching). These modifications could be manually controlled or automated.

Stitch                   An advanced merge operation that attempts to completely blur the boundary between overlapping source rasters by feathering the boundary between rasters. Feathering techniques like cosine weighted summing or minimum curvature smoothing could be employed.

Blend                   A method of combining data from overlapping source rasters by using some statistical technique (averaging, median value, mode etc.)

Feather                A method employed to eliminate the boundary between overlapping source rasters by manipulating the data using smoothing techniques like minimum curvature.

Mosaic                 In the context of satellite multispectral imagery, a mosaic is a “Join” of adjacent satellite scenes.

Composite           In the context of satellite multispectral imagery, compositing satellite scenes combines multiple scenes using masking and blending to improve the quality of the product. In other disciplines this can be referred to as ‘stacking”.

Raster Source      When you employ a Raster Source that contains multiple source rasters, ProRaster effectively “Joins” those rasters when you consume the raster in the rendering engine or anywhere else. 

The Coordinate System, Cell Size, Cell Alignment, and Structure of the source rasters in any Merge operation is also of critical importance. In ProRaster, the following rules apply. 

1. In a “Raster Source”, all source rasters must have the same coordinate system. They must have the same structure (meaning the same Fields and Bands, and the same Events). Cell size is expected to be similar in all source rasters, but cell size and alignment do not have to match.
2. In a “Simple Join” operation, the source rasters will have the same coordinate systems, but cell size and alignment can differ. All source rasters should have the same structure.
3. In a “Complex Join” operation, the source rasters can have different coordinate systems, cell size, and cell alignment. All source rasters should have the same structure. 

The Simple Join operation takes a single Raster Source as input and outputs a virtual raster (MVR) that executes the merge operation in real-time. Create the raster source in advance using the Raster Source Editor. The raster source ought to contain two or more source rasters that share the same coordinate system and raster structure. Cell size and cell alignment can differ. Simple Join does not support batch processing. 

The primary application of this is to take a collection of rasters that are in a tiled arrangement (or postage stamp arrangement if you prefer) and bring them together into a single raster. In this scenario, it is expected that all the source rasters will have the same coordinate system. It is likely they will share the same cell size and that the cells will align and not overlap, but this is not a requirement and will not prevent a valid merge. 

Where cells in rasters overlap, the order of the rasters in the list determines what value will be retained in a cell where there are multiple possible inputs. Rasters in a Raster Source are processed from First to Last where rasters that are later will overprint rasters that are earlier. In the Raster Source Editor, rasters are listed in first to last order (from top to bottom of the list). These rules only apply to valid cell values – invalid (null) cells do not overprint valid cell values. 

Complex Join Operation

The Complex Join operation outputs an MVR virtual raster that merges a collection of rasters. You can browse for one or more rasters in any supported format and/or select one or more raster sources (which may contain many rasters). Complex Join does not support batch processing. 

Unlike the Simple Join operation, in a Complex Join operation you can combine rasters that are spatially different. There is no requirement for the coordinate systems or any registration information to match. There is still an expectation that the rasters will have the same field, band, and event structure. 

You can choose to join all fields and bands, or a selected field and all its bands. 

If the input rasters have different coordinate systems, it is likely you will need to define a suitable coordinate system for the output virtual raster. For example, if you join data over multiple UTM zones, you might choose a geographic output coordinate system. 

You can define the cell size and origin coordinate for the output virtual raster. In fact, this will make little or no difference to the appearance of the raster when it is rendered, but it will have an impact if you choose to export the MVR to a crystallised raster format, and it can impact performance. 

If you define a coordinate system, make sure you define a cell size that is suitable for that coordinate system. For example, if it is projected, you might define a cell size in meters. If it is geographic, you would define a cell size in fractions of a degree. The most important function of the origin coordinate is to define the cell alignment. The actual location of the origin coordinate with respect to the raster data is not important. 

Where input rasters overlap, the merge operation must determine a final cell value from two or more candidate values. In a Simple Join it uses the last valid value obtained. In a Complex Join you can use the “Combine Values“ rule to control how cell values are computed. 

First                      The first valid value obtained for the cell is retained and never overwritten. In effect, this will reverse the order of the source rasters being merged.

Last                      The last valid value obtained for the cell is retained. This is the default setting and is equivalent to the Simple Join operation.

Minimum             The minimum valid value for the cell is retained. The order of the source rasters will make no difference to the outcome.

Maximum            The maximum valid value for the cell is retained. The order of the source rasters will make no difference to the outcome.

Running average A pseudo-average value is computed. The system computes the average of the existing cell value and the new cell candidate value and retains this value. In effect, the average is weighted towards the last value received.

Average               The average of all contributions is computed. The order of the source rasters will make no difference to the outcome. This is a blending process.

Sum                      The sum of all contributions is computed. The order of the source rasters will make no difference to the outcome.

Count                   The number of valid contributions to each cell is counted and recorded. The cell values themselves will be discarded. The order of the source rasters will make no difference to the outcome. 

In general, the output virtual raster will have bands that match the data type of the input source rasters. Usually, this is appropriate, but there are scenarios where this causes issues. 

The system is engineered so that virtual rasters return data in the band data type at every resolution level. This differs from normal rasters that return data in the band data type for the base and overview levels, but in 4-byte decimal (REAL4) for underview resolution levels. Underview resolution levels are often smoothly interpolated and so, even if the data type is integer, they use a decimal data type. If the band data type is integer, then a virtual raster will never properly implement smooth interpolation. To work around this problem, by default all integer band data types are promoted to decimal (either 4 or 8-byte as needed). This only applies to “Continuous” field types – Image, ImagePalette, and Classified fields are not affected. 

When the merge operation is executed, it must obtain input cell value data from the source rasters. By default, it will acquire that data as either 4 or 8-byte decimal numbers. It can be more efficient to acquire the data in the data type of the source raster and the “Inherit data type for all bands” option gives you the ability to make this happen. However, note that this can trigger the same kind of interpolation issues. If the source raster is being tapped for an underview, this may then be forced into an integer data type and smooth interpolation will be lost.