self-emission model should use ODR

[gerritholl]

The self-emission model should use ODR rather than OLS. I have, or should have, uncertainty estimates for the temperatures and reference offsets. OLS gives biased results, such as Sams Jupiter Notebook nicely illustrates.

[gerritholl]

But: With no OR INCOMPLETE uncertainty information, ODR too will yield wrong results.

[gerritholl]

I wonder if ODR would resolve biases such as visible in this self-regression test.

[gerritholl]

When there is a state-dependent bias, how can I determine whether this is due to the model (which may be incomplete), or whether this is due to biases in determining the coefficients of the model (OLS, PLS, ODR, EIV)?

[gerritholl]

Elsewhere, such as in the SRF recovery or the calculatino of K (see #183), I have not found that ODR improves the result over LR. But if I use proper uncertainties (#36) ODR should have more advantages over LR, than when I don't.