Dear all, good afternoon!!

I'm running two numerical experiments (HYCOM model) and compare them to a reference (WOA climatology).

My first experiment (control one) gives a bias when compared to the reference: (expt1 - ref). I want to evaluate how the difference brought by the effect introduced in the second experiment (wave-modified surface stress: expt2 - expt1) improve or worsen the model results.

It seems hard to gather everything in a single map since the bias can be positive or negative and the effect brought can over compensate the original bias (move the SST toward the reference more than the original bias and give a new bias of opposite sign).

All I could figure out is to get rid of the sign and use error instead of bias, where error = abs(bias). I have then the error variation: abs(exp2-ref) - abs(expt1-ref). I can divide it by the original error to get the relative variation and know for instance if the original error is over compensated but it gives noisy maps for the whole domain.

I guess I could focus on specific areas like the Benguela resurgence and the Brazil-Malvinas convergence area. Is there any classic approach to show and evaluate this variation in a nice way?

Have a good day,

Nicolas