Tweak exponential falloff in ScatteringTerm

[ecoskey]

Objective

current exponential falloff parametrization isn't intuitive, strays from previous "scale height" parametrization common in atmospheric scattering literature

Solution

• Switch to a parametrization of the form

          -(1-p)/s    -1/s
         e         - e
f(p,s) = -----------------
                      -1/s
         1.0       - e

This is basically equivalent to the "scale height" parametrization e^-(1-p)/s for small values of scale, but also has the nice properties f(0, s) = 0, f(1, s) = 1 and has a broader domain for values of scale.

Testing

• Ran the example, results looked good

[ecoskey]

I took this directly from your pr @mate-h, but isn't this meant to be 100km?

[ecoskey]

This discontinuity is kind of unfortunate, since it's there because of the division by scale below. I'm fine with it though since it lets us stay close to the "scale height" parametrization.

[alice-i-cecile]

Very clear and a sensible parameterization. Does this need a migration guide if we get it in during 0.18?

[ecoskey]

Very clear and a sensible parameterization. Does this need a migration guide if we get it in during 0.18?

Maybe? Up to you but I'd lean towards improving module docs instead, the main friction is going to be the whole new asset type anyways. This is also much closer to what's in 0.17 than what was merged in the general scattering PR.

[alice-i-cecile]

Maybe? Up to you but I'd lean towards improving module docs instead, the main friction is going to be the whole new asset type anyways. This is also much closer to what's in 0.17 than what was merged in the general scattering PR.

Let's prioritize that then. Better docs are much more durable. I won't block on it in this PR.