Grid Merge

Grid Merge combines two or more grids into one raster. The output covers all the inputs at a cell size you choose, and each grid is levelled to match its neighbours before it is laid in, so surveys gridded at different times, resolutions or base levels join into a single consistent grid.

The output value at any cell comes from one grid, not an average of several. Each grid takes preference over the ones merged before it wherever it has data, so its join sits at its own data edge, and a narrow feather blends across the join. A high resolution grid dropped wholly inside a regional one keeps its full extent, with the feather running around its rim.

Opening the Dialog

Hover over the Data Processing icon in the right tool panel, then Interpolation, and click Grid Merge....

The Workflow

The dialog has two tabs.

1. Inputs - the grids to merge, in order.
2. Merge - the correction model, the feather, the output lattice and file.

OK merges and writes the file. If anything is missing or contradictory, an amber line above the buttons says what, and OK stays disabled until it is fixed. While the merge runs the tabs give way to a progress bar with a Cancel button, and the dialog closes itself when the file is written.

Inputs Tab

Add Grids... opens a file dialog; grid files can also be dragged and dropped onto the tab. Any raster format SI can read is accepted, including GeoTIFF and Geosoft GRD grids, and a grid with several bands gets a Band choice on its row.

The order of the list is the merge order, and it matters twice.

• The first grid sets the datum: every later grid is corrected to match what has been merged before it.
• Later grids take preference: where grids overlap, the later one's values are used up to its own data edge. Put the grid you trust most - usually the newest or highest resolution - last.

Drag a row to reorder, and use the bin button to remove one.

Merge Tab

Correction is how each grid is adjusted to match the grids merged before it. The parameters are fitted by least squares over the cells where the grid and the running mosaic overlap, so the grids themselves say what correction they need.

A correction that cannot be trusted falls back to a simpler one and says so when the merge finishes: a trend needs a reasonable number of overlapping cells and an overlap with some width to it, and a scale needs the grid to vary over the overlap. A grid with no overlap at all is merged uncorrected.

Feather width is how far inside each grid's data edge the join is blended, in ground units. Within that band the value ramps smoothly from the earlier grids to the later one; beyond it the later grid stands alone. Zero is a hard join. A few cells is usually enough once the correction has removed the base level difference; widen it when a residual mismatch still shows along the join.

The extent boxes and Cell size define the output lattice, as the centres of the corner cells. Input extents fills them from the union of the inputs, with the cell size of the finest input. The whole output grid is held in memory while merging, and the dialog warns when the cell count makes that expensive.

Output

Format lists every raster format SI can write; GeoTIFF is the default. The merged band carries a record of how it was made - the input files, the correction model and the feather width - as metadata items whose names start with SI_, which most GIS packages can display.

Build overviews adds reduced resolution copies once the grid is written, so a large grid draws quickly at any zoom.

Add to project as a surface loads the merged grid as a new surface layer when it is written. Undo removes the layer again; the file stays on disk.

Merging From the Assistant

The Assistant can run the same merge from a sentence: "merge the regional and detailed magnetic grids". It names the grids in order, with detail last, and can choose the correction and feather. The extent and cell size default to all the inputs together at the finest input cell. The merge runs in the background and writes the file, which the Assistant can load as a surface if you ask. The result is logged like a run from this dialog. A merge above fifty million cells is refused, and the Assistant does not use the settings remembered here.

How the Corrections Are Fitted

Each grid is levelled against everything merged before it, using only the cells where the two overlap: the offset is the mean difference, offset and scale come from a regression of one on the other, and the base slope is a planar trend in the differences. The correction is then applied to the whole of the new grid, so it carries on beyond the overlap. The log records what was fitted for each grid and how much of the mismatch it removed - worth a look when a join still shows.

The approach follows the automated grid stitching method of Cheesman, MacLeod and Hollyer (Exploration Geophysics, 1998): level the grids first, join them at the edge of the preferred grid, and keep any blending close to the join rather than distorting whole grids.