Allow users to use own TOA as input for erbs

[ramaroesilva]

Is your feature request related to a problem? Please describe.
Often decomposition models use TOA as input. For practicality, (at least some) pvlib models calculate TOA internally in case the user does not have it. However, this should be optional to increase coherence in the cases where users have TOA.

This is done at least for erbs_driesse, but not for erbs nor boland (maybe others as well).

Describe the solution you'd like
Allow user-defined TOA as done for erbs_driesse.

[kandersolar]

Unless I miss it, at least the Erbs paper does not mention how dni_extra should be computed, so +1 to making it a parameter.

[cwhanse]

I disagree with adding a parameter for dni_extra to erbs. The equation for extraterrestrial irradiance used by the authors is built into the correlation they developed. The paper does not provide that equation. It does allude to using TRNSYS, which makes sense given the author's affiliation. TRNSYS documents from 1987 describe the calculation of extraterrestrial irradiance here, in a formula that looks like the ASCE method (not spencer which is the current default). On that same page appears a version of the Erbs correlation for diffuse irradiance, but with rounded coefficients.

I'm inferring a lot from remarks in the 1982 paper. But I think we'd be misrepresenting the Erbs model to present dni_extra as a variable input, which it was fixed (although unknown to us.)

[kandersolar]

Thanks @cwhanse for looking into it. IMHO not knowing which method is correct is more reason to expose dni_extra as an input, not less. Why do we arbitrarily choose a method and take that choice away from the user? I'd say that our current function misrepresents the model, as it hardcodes one method without any basis for doing so.

[cwhanse]

it hardcodes one method without any basis for doing so.

I have to agree with this.

Since I think it is incorrect to suggest that a user should choose a model for extraterrestrial irradiance, but we don't want to make that choice for them, what about adding dni_extra_model='asce'?

That's my best guess at what Erbs et al. actually used.

[ramaroesilva]

Thanks @kandersolar and @cwhanse for the insightful discussion.

I have to admit that now I see my initial remark as being a bit rushed. In fact, at first it seemed logic to me that the TOA, Kt, whatever used within a modelling chain should be always the same (as a sign of coherence and computation efficiency). But, as @cwhanse rightfully points out in any given model the parameters derived in a paper are intrinsically linked to whatever assumptions (including solar position, TOA, etc.) the original author made.

I like the suggestion of having dni_extra_model to make this assumption more explicit. However, it forces TOA to be computed even when the user had already precomputed TOA before calling erbs using the same model defined in dni_extra_model.

If we combine @kandersolar suggestion, we could have both dni_extra_model and dni as inputs (with dni defaulted as None and ensuring that users have one of the two as None - similar to what I did for k and noct in the Ross temperature function) and in the documentation strongly recommend keeping the original setup for maximum coherence with the paper.