The Ocean Remembers Randomness: How Noise Memory Strengthens Climate Links ✛ ✎

Original: "A Liang-Kleeman Causality Analysis based on Linear Inverse Modeling"
· Justin Lien
arXiv:2409.06797 · 2024-09-10 · arXiv non-exclusive · independent analysis · 1 min
Random disturbances in the ocean don't vanish right away; they remember themselves for months, and this amplifies El Niño's influence on the Indian Ocean Dipole.

Ocean currents intertwine like a sticky web: every disturbance leaves a trace that doesn't disappear instantly. Scientists used to think that random jolts in this web were forgotten right away, but new research shows that in the Niño 3 region, noise remembers itself for more than two months. This memory strengthens the link between El Niño in the Pacific and the Indian Ocean Dipole.

Two models were compared: one treats random disturbances as memoryless, the other gives them memory. The second model matches real data from 1958–2010 more closely.

The method works like this: build a simple linear model where random forcing can be with or without memory. Then tune the parameters so the model reproduces observed correlations. Next, compute the information flow between variables: it shows who influences whom. It turns out the Pacific stabilizes the dipole, while the dipole excites the Pacific. Accounting for noise memory makes these flows stronger, especially in the Niño 3 region.

The sticky web of ocean connections turned out to be even clingier than thought: a disturbance that hits the Niño 3 node lingers there longer than a seasonal cycle. Noise memory, like a viscous thread, stretches El Niño's influence on the Indian Ocean Dipole and back. Now climatologists can see where the web doesn't break but merely sags slightly under the weight of randomness.

Noise memory in the Niño 3 region exceeds two months, meaning a random ocean perturbation “remembers” itself longer than the average seasonal cycle lasts.

T_{j\to i} = A_{ij} \frac{C_{ij}}{C_{ii}}
Causality is proportional to the dynamical coupling and the correlation normalized by the variance.
K_w(s) = e^{sA_w} C
Correlation decays exponentially with time, with the rate set by the dynamics matrix.
K_c(s) = e^{sA_c} C + e^{sA_c} \int_0^s e^{-s'(A_c + \frac{1}{\tau})} ds' Q_c B^T
An additional contribution arises from noise memory, depending on the correlation time τ.
Scientists
Jascha Sohl-Dickstein
Concepts
linear inverse model causality nonlinear laplacian spectral analysis Multivariate ENSO Index granger causality Jin-Timmermann model ENSO causal set theory linear regression chaos theory TAO/TRITON moorings El Niño index inverse problem interannual variability independent component analysis natural variability noise cancellation Southern Oscillation Indian Ocean Dipole information criterion