doc: write up the polygonisation logic/algo

[chrisbc]

whilst we still remember

[voj]

Poligonisation of spatial PDF gridded data

Inputs:

STEPS : resolution of up-sampled grid
EXPONENT

1. Up-sample pdf grid from 10 steps per degree to STEPS steps per degree.
   • Values for the new grid are assigned by the nearest neighbour from the original grid. No new gradients are introduced.
   • The grid is then normalised to 1
   • From now on we operate on the up-sampled grid unless noted otherwise
2. For each sub section:
   • For each grid point inside the sub section polygon.
     • polyWeight = distance_of_point_to_fault_trace / distance_of_trace_to_poly_border
       • This means that polyWeight is 1 if it's on the polygon border and 0 if on the fault trace.
       • These distances are calculated parallel to the direction of the fault dip.
     • d = polyWeight^EXPONENT
     • new_value_at_grid_point = old_value * d / sum_of_all_d_in_polygon
3. Where a grid point is in more than one polygon, we average over all new values
4. Grid points not in polygons are left unmodified.
5. Down-sample to 10-step grid resolution.
   • No averaging or normalisation is performed.
   • From now on we operate on the normal grid resolution again
6. Mmins are calculated per grid point:
   • Add up all mMin values from sub sections where the grid point is inside the polygon
     • mMin values are normalised based on how much the grid bin overlaps with the fault section polygon

[voj]

Updated the write-up based on feedback by Matt G and Chris DC.