Here is an interesting idea concerning dipoles that I have been playing around with.
The Southern Oscillation of ENSO is defined as the magnitude of the atmospheric pressure dipole between Tahiti and Darwin. See the following figure, where x represents the longitudinal distance between Darwin (to the west) and Tahiti (to the east).
So when Darwin is positive, Tahiti is generally negative (and vise versa).
However the data is fairly noisy, so it is difficult to trust any particular data point. It is possible that one or the other of the data points is more trustworthy than the other. It may be that instead of using Tahiti-Darwin, one could use 2×Tahiti or -2×Darwin, if that reflected the "true" value of the dipole.
My idea is that the SOI model could be compared to either T-D, 2×T, or -2×D, depending on what minimizes the error between data and model. Remember that the model does not have a lot of parameter wiggle room, as the factors are set.
It amounts to a simple Mathematica one-liner algorithm, and shown in the figure below are the results of picking the "best" SOI data and comparing to the SOI model (SOIM) in red.
I am not sure if this is invoking Morton's demon confirmation bias, or is an appropriate kind of learning classifier that is useful for handling uncertain data.
The Southern Oscillation of ENSO is defined as the magnitude of the atmospheric pressure dipole between Tahiti and Darwin. See the following figure, where x represents the longitudinal distance between Darwin (to the west) and Tahiti (to the east).
So when Darwin is positive, Tahiti is generally negative (and vise versa).
However the data is fairly noisy, so it is difficult to trust any particular data point. It is possible that one or the other of the data points is more trustworthy than the other. It may be that instead of using Tahiti-Darwin, one could use 2×Tahiti or -2×Darwin, if that reflected the "true" value of the dipole.
My idea is that the SOI model could be compared to either T-D, 2×T, or -2×D, depending on what minimizes the error between data and model. Remember that the model does not have a lot of parameter wiggle room, as the factors are set.
It amounts to a simple Mathematica one-liner algorithm, and shown in the figure below are the results of picking the "best" SOI data and comparing to the SOI model (SOIM) in red.
I am not sure if this is invoking Morton's demon confirmation bias, or is an appropriate kind of learning classifier that is useful for handling uncertain data.
Version 2.1.8p2
