Grid Merging Fundamentals

A common problem with airborne geophysical coverage is that surveys are flown in a patchwork fashion over many years, and are not all registered to the same datum. This means that surveys seldom match exactly along their common borders, making it difficult to merge surveys into larger compilations. However, both gamma-ray spectrometric and magnetic surveys can be levelled and merged together by using the differences in the areas where the surveys overlap to estimate correction factors. In general, magnetic surveys require a level shift adjustment, whereas radiometric surveys require both a shift and a scaling adjustment. In many cases, both magnetic and radiometric data may also require a tilt adjustment.

The approach taken by GridMerge is to treat the estimation of the shift and scale adjustments required to level the surveys as a single inverse problem. However, in order for a solution to exist, the following criteria must be met:

• The grids must be interconnected in some way. If a grid does not overlap any of the grids in the collection, then it cannot be levelled.
• In order to level the grids to a common baseline, this baseline must be specified. The baseline could comprise (a) one or more "base grids"; (b) a set of control traverses that intersect one or more of the grids; or (c) a combination of both. The control traverses and base grids are not adjusted –  they are assumed to be at the correct level. All the other grids in the collection are adjusted to fit this baseline.
• Where grids intersect, there must be sufficient overlap points, and (in the case of radiometric data) sufficient dynamic range in the data  to accurately estimate both shift and scale adjustments.

After levelling the grids using shift and scale adjustments, the fit between surveys can often be further improved using tilt adjustments. Again, this is treated as a single inverse problem. The grid overlaps are used to estimate a plane surface for each grid such that, once this plane is subtracted from the grid, the differences in the overlap areas are minimised. Once again, a baseline is required – either control traverses and/or base grids must be specified.

Once grids are leveled, they can then be merged into a single composite grid. GridMerge offers two options for merging:

• A Merge, where grids are re-sampled and patched onto the merged grid in reverse-priority order (i.e. the poorest quality grids are placed at the bottom, the best on top). This is a simple and fast solution used for quality control; and
• A  more sophisticated merge where the grids are Feathered And Merged. With feathering, any residual differences along the grid edges are removed by propagating them into the overlapping grids by distances proportional to their wavelength. This produces a seamless join. Feathering is usually reserved for the final merge as it is quite compute intensive and somewhat slower.

Finally, users should be aware that it is possible to have an almost perfect fit between individual grid pairs, yet a poor fit to the merge! Consider the following situation.

Let us assume: grid A fits grid B along their common border; B fits C; C fits D; and D fits A. However, if just one of the grids is now tilted, the collection becomes inconsistent, and can never be levelled using just a shift adjustment. It might also not be obvious which grid is tilted, as GridMerge tends to distribute any inconsistency between all the overlaps. GridMerge provides the capability to force the inversion to honour particular grid overlaps. This can be used to force any inconsistencies to other parts of the merge where they may be less obvious.