Right now our framework is built on a "state" concept, where each wind
speed and wind direction represent a flow state (could also make
combinations of various other wind resource characteristics).
(response from Erik Q.)
My feeling is that if you have a Cartesian product state space, using
‘iterated expectation’ (which is what I described) should remain possible.
Any joint probability distribution is very often provided decomposed in a
marginal and conditionals (in what we looked at: marginal for direction and
conditionals for wind speed on direction) and this decomposition determines
the efficient calculation of the expectation. So any set of characteristics
and particular decomposition of the joint distribution gives rise to a
single calculation approach.
(Am I making sense?)
If you don't have a Cartesian product space (e.g., different speeds per
direction), this breaks down and working with the joint directly may then
be more efficient than filling up the holes to create a Cartesian product
structure.
Excerpt from email from Erik Q.
(first some comments by Jared T.)
(response from Erik Q.) My feeling is that if you have a Cartesian product state space, using ‘iterated expectation’ (which is what I described) should remain possible. Any joint probability distribution is very often provided decomposed in a marginal and conditionals (in what we looked at: marginal for direction and conditionals for wind speed on direction) and this decomposition determines the efficient calculation of the expectation. So any set of characteristics and particular decomposition of the joint distribution gives rise to a single calculation approach. (Am I making sense?)
If you don't have a Cartesian product space (e.g., different speeds per direction), this breaks down and working with the joint directly may then be more efficient than filling up the holes to create a Cartesian product structure.