The amazing invisible floating bookshelf

How To Install Invisible Shelves

I love it! Haven't you always wanted one of these floating bookshelves. Imagine a whole wall of them...

And the video is just so lovably cheesy.

[via Lifehacker]

Gridding errors for potential field data

I regularly have a need to grid gravity and magnetic geophysical data, but don't have much in the way of fancy software to do it, mainly an old version of ER Mapper (6.3) and MATLAB. Recently I made a comparison between the methods I available to me and got some surprising results.

I have an extremely irregularly spaced set of gravity data on a topographic surface. Data spacings range from 25 m to 10 km. I require a 400 m grid. I will do some processing later that will get rid of grid points that are located "far" from any original data points, so my primary criteria for a good grid is:

1. The grid patch near an observed data point reproduces that data point accurately
   
• I judged this by interpolating the recovered grid at the original data points and calculating the difference between the original data points and the grid value at those points.
2. The grid patch near observed data points has a nice shape to it - no rips, tears, jumps, spikes, etc.
• I judged this by looking at the grid image and seeing if I liked how it looked near data points.

The methods I tried were:

• ER Mapper, minimum curvature gridding, with tension = 25 (ER Mapper default for potential field data).
• The grid image had little bumps at each of the datapoints that were widely spaced (like tent poles holding up a tent). The average difference to the original data points was 0.26 mGal or 0.53%
  
  
• ER Mapper, minimum curvature gridding, with tension = 15.
• The grid image still had the little bumps at each of the datapoints, but not as badly as the example above. The average difference to the original data points was 0.25 mGal or 0.51%.
  
  
  
• MATLAB - GRIDDATA, 'v4' spline gridding.
• The grid image looked nice and smooth where data were sparse, and detailed where data was abundant. The average difference to the original data points was 0.17 mGal or 0.34%.
  
  
• MATLAB - GRIDDATA, 'cubic' gridding.
• The grid image looked okay, but in some places there were sharp facets or triangles in the grid where data were sparse. The average difference to the original data points was 0.18 mGal or 0.37%.
  
  
  

This is hardly a robust analysis of gridding methods, and doesn't take into account all the multitude of algorithms out there, but it is enough to tell me that MATLAB's griddata command using 'v4' splines is more than acceptable for these problems (even if there are complaints about how inefficient and unreliable GRIDDATA it is). It will be my method of choice until I find a better one.

But it also gives a good indication of how poor all of these methods are at creating reliable grids. The errors of > 0.17 mGal are much greater than the measurement accuracy (including all sources of error) of < 0.05-0.10 mGal. User beware!