The Ocean's Long Reach: Accounting for Teleconnections in Spatial Models ✎

In the mountains of Colorado, snowfall varies from winter to winter, and the reason is not always local weather. Thousands of kilometers away, in the tropical Pacific Ocean, water warms and cools, and this pulsation extends an invisible hand that alters precipitation in the Rockies. The connection between distant regions is called a teleconnection, and it is usually described by preselected indices, such as El Niño. But the appropriate index depends on the region and may not be known before analysis. The authors propose a model that does not choose an index; instead, it directly uses ocean temperature and finds on its own which areas matter most for Colorado precipitation.

To understand how this works, imagine an interpreter who translates a conversation between the ocean and the mountains. A typical model asks the interpreter: "Do you know El Niño? Pass along only what he says." The new model gives the interpreter the entire recording of the ocean's conversation and lets the interpreter decide which phrases matter for the mountains' reply. To do this, sea surface temperature is decomposed into the main patterns of variability, called empirical orthogonal functions. The first such pattern explains 30% of all temperature fluctuations and exactly mirrors the El Niño pattern: warm water in the center and east, cool water in the west. The model does not know this in advance; it identifies this pattern from the data on its own.

The result agrees with the known picture: during El Niño, northern and western Colorado receive less precipitation, while the east receives more. But now the model estimates this influence not through a single index but through the entire ocean map. This is akin to using thousands of thermometers across the ocean instead of one at a single point, and letting the data show which regions matter. Such an approach adds physically meaningful structure to statistical climate models.

The key difference from related studies is that the model does not require teleconnection indices to be specified in advance. Many works predict the indices themselves or apply preexisting patterns; here, the model itself identifies significant remote influences. This is useful for downscaling, when forecasts from global climate models are translated into local precipitation. For example, for forecasting water supply in regions dependent on El Niño, such a model can provide a more honest estimate of uncertainty.

The ocean's hand reaching toward the mountains of Colorado is no longer guessed from a single index but is felt across the entire water surface. And if the model is extended to multiple oceans or multivariate data in the future, it could capture even subtler long-range connections. For now, it has already shown that remote influences can be incorporated directly into statistics without losing local data structure. This is a step toward models that see climate as a single system rather than a collection of isolated regions.