Short-term sea surface temperature (SST) forecasts are crucial for operational oceanology. This study introduces a specialized Transformer model (U-Transformer) to forecast global short-term SST variability and compares its performance with Convolutional Long Short-Term Memory (ConvLSTM) and Residual Neural Network (ResNet) models. The U-Transformer model forecast consistently outperformed the ConvLSTM and ResNet models, especially in regions with active mesoscale eddies. Globally, the U-Transformer model achieved SST root mean square errors (RMSEs) ranging from 0.2 °C at a 1-day lead time to 0.54 °C at a 10-day lead time during 2020–2022, with anomaly correlation coefficients (ACCs) decreasing from 0.97 to 0.79, respectively. However, in regions characterized by active mesoscale eddies, RMSEs from the U-Transformer model exceeded the global averages by at least 40%, with values in the Gulf Stream region reaching more than twice the global average. Additionally, ACC values in active mesoscale eddy regions declined more sharply with forecast lead time compared to the global averages, decreasing from approximately 0.96 at a 1-day lead time to 0.73 at a 10-day lead time. Specifically, the ACC value dropped to 0.89 in the Gulf Stream region at a 3-day lead time, while maintaining 0.92 globally. These findings underscore the importance of advanced approaches to enhance SST forecast accuracy in challenging active mesoscale eddy regions.

The hurricane, with maximum wind speed over 64 kts, is among the most terrible calamities over the northern Atlantic (NATL). Previous studies identified a poleward migration of tropical cyclone (TC) genesis over the Pacific Ocean, but the shift over the NATL is statistically insignificant. The present study detects a robust southward migration in the genesis latitude of NATL TCs that later reach hurricane strength after 1979, which is consistent with a growth in hurricane frequency in the southern part (10°-20°N) of NATL. This increasing trend of hurricane frequency is intimately attributable to the decreasing vertical shear of zonal wind, resulting from a decreasing north-south temperature gradient. The reduced north-south temperature gradient is primarily caused by greater warming trend in tropospheric temperature in the subtropics, driven by intensified static stability. The present research suggests a potential increase in the hazards confronted by low-latitude islands and coastal nations in Northern America.

Previous studies have shown that year-to-year incremental prediction (YIP) can obtain considerable skill in seasonal forecasts. This study analyzes the mathematical definition of YIP and derives its formula in the nonlinear time series prediction (NP) method. It is shown that the two methods are equivalent when the prediction time series is embedded in one-dimensional phase space. Compared to previous NP models, the new one introduces multiple external forcings in the form of year-to-year increments. The year-to-year increments have physical meaning, which is better than the NP model with empirically chosen parameters. The summer rainfall over the middle to lower reaches of the Yangtze River is analyzed to examine the prediction skill of the NP models. Results show that the NP model with year-to-year increments can reach a similar skill as the YIP model. When the embedded number of dimensions is increased to two, more accurate prediction can be obtained. Besides similar results, the NP method has more dynamical meaning, as it is based on the classical reconstruction theory. Moreover, by choosing different embedded dimensions, the NP model can reconstruct the dynamical curve into phase space with more than one dimension, which is an advantage of the NP model. The present study suggests that YIP has a robust dynamical foundation, besides its physical mechanism, and the modified NP model has the potential to increase the operational skill in short-term climate prediction.