Implement Jacobian determinant

What should we add?

GEMPAK provides a JCBN function, so to support the transition to MetPy, it would be good to provide a direct equivalent. The GEMPAK code indicates:

 * This subroutine computes the Jacobian determinant of two scalar	*
 * grids:								*
 *									*
 *     JCBN ( S1, S2 ) = DDX(S1) * DDY(S2) - DDY(S1) * DDX(S2)		*
 *									*
 * Map scale factors are included implicitly.				*

Looks like a pretty easy thing to implement (famous last words!)

Reference

No response

To have "map scale factors [...] included implicitly," taking inspiration from #1389 (comment), I think this could be implemented something like

def jacobian(field_0, field_1):
    grad_field_1_y, grad_field_1_x = gradient(field_1, axes=('y', 'x'))
    return vorticity(field_0 * grad_field_1_x, field_0 * grad_field_1_y)

Since

@jthielen

Our current implementation of vorticity won't work with @jthielen suggestion since we check for units of speed in a decorator with the function. Doing two gradient calls doesn't seem like it would be too much with a function like

def jacobian(field_0, field_1):
    grad_field_1_y, grad_field_1_x = gradient(field_1, axes=('y', 'x'))
    grad_field_0_y, grad_field_0_x = gradient(field_0, axes=('y', 'x'))
    return grad_field_0_x * grad_field_1_y - grad_field_0_y * grad_field_1_x

Otherwise, we could develop a more generic curl function that might help both vorticity and this calculation.

@kgoebber

@kgoebber That approach was my first thought, but I avoided it because my prior impression from #893 was that scalar gradients (as used there) wouldn't have any scale factor corrections, and so this in turn would lack them, compared to the curl-based approach which would.

However, this mismatch between formulations that should otherwise be equivalent worried me a bit, so I went back to https://en.wikipedia.org/wiki/Orthogonal_coordinates and the implementations in #893 to figure out the proper/expected behavior. Long story short, I'm pretty sure we need to apply scale factor corrections in gradient:

so that the Jacobian (via either curl of product with gradient or cross product of gradients, the former results in a lot of cancellations so the latter is probably better computationally) ends up as:

Ah, yes. Good catch.

Implement Jacobian determinant

Navigation Menu

Repository navigation

Implement Jacobian determinant #2356

What should we add?

Reference

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Implement Jacobian determinant #2356

Description

What should we add?

Reference

Activity

jthielen commented on Feb 22, 2022

kgoebber commented on Feb 22, 2022

jthielen commented on Feb 23, 2022

kgoebber commented on Feb 23, 2022

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions