Mistake in stretching matrix in layered model

[mpudig]

Hi,

We (in fact, @Empyreal092) noticed a possible mistake in how the reduced gravities are defined in layered_model (https://github.com/pyqg/pyqg/blob/master/pyqg/layered_model.py#L176).

It is our understanding that in the multi-layer QG equations, one should divide by a reference density in the reduced gravities (see Vallis, 2017 p.186), rather than by the density of each layer as is currently done in layered_model (as well as in its pyqg documentation).

Could anyone who was part of model development (@rabernat, @MFJansen, @cesar-rocha, @francispoulin perhaps) comment on whether this is in fact a typo, or if there was a reason behind this choice for the reduced gravities? I'd be happy to push through the changes necessary if required.