initial commit

[magpowell]

I'm proposing using the skipping boundary crossings in x and y as our compression algorithm over the previous manual one. Skipping boundary crossings in x and y works to "compress" the tilted field as the number of layers out is always equal to the number in. Rather than merge vertical layers from the top down, we only record values and create new path segments when a z boundary is crossed. This distributes the compression randomly throughout the column rather than starting at the top.

The key change is here:

--   if ((std::abs(l - lx) < epsilon) ||  (std::abs(l - ly) < epsilon) || ( (std::abs(l - lz) < epsilon || zp >= zh[k+1]))){
++   if ( (std::abs(l - lz) < epsilon || zp >= zh[k+1])){
            // Create a new path segment after crossing boundary
            tilted_path.push_back({i, j, k});
            dz_tilted.push_back(0.0);
}

It would skip the boundary crossings and values for the pixels highlighted. Those segments still count toward the path length of the previous value.

I was motivated to look into this after than errors we saw in the aerosol PR:

Now, with this change we don't have higher percent differences with the 3D direct surface field for some solar zenith angles:

In fact, the simplified TICA has lower errors overall for runs with and without aerosols. It also has the added benefits of:

• simpler, more readable code (100s of lines fewer)
• runs faster

Finally, the simplified TICA can be run for theta greater than 70 degrees, but percent absolute errors increase dramatically after ~75 degrees: