Large-Ensemble Simulations Reveal Links Between Atmospheric Blocking Frequency and Sea Surface Temperature Variability

TL;DR

The study investigates how sea surface temperature (SST) variability influences atmospheric blocking in the Northern Hemisphere by using century-scale AMIP-style simulations from two deep-learning GCMs (DLESYM and NGCM) and a HiRAM reference. It demonstrates that DL-based ensembles reproduce blocking climatology with skill comparable to or exceeding traditional models, and that large ensemble means effectively isolate the SST-forced component, yielding stronger correlations with reanalysis data. The analysis reveals physically interpretable teleconnections, including a North Atlantic SST tripole linked to Greenland blocking and an El Niño–like pattern in the tropical Pacific, alongside SST-driven, regionally varying trends in blocking. Overall, the results show that large DL-enabled ensembles can separate boundary-forcing signals from internal atmospheric noise and illuminate mechanistic SST–blocking links, offering a cost-effective path for future forcing-sensitivity experiments.

Abstract

Atmospheric blocking events drive persistent weather extremes in midlatitudes, but isolating the influence of sea surface temperature (SST) from chaotic internal atmospheric variability on these events remains a challenge. We address this challenge using century-long (1900-2010), large-ensemble simulations with two computationally efficient deep-learning general circulation models. We find these models skillfully reproduce the observed blocking climatology, matching or exceeding the performance of a traditional high-resolution model and representative CMIP6 models. Averaging the large ensembles filters internal atmospheric noise to isolate the SST-forced component of blocking variability, yielding substantially higher correlations with reanalysis than for individual ensemble members. We identify robust teleconnections linking Greenland blocking frequency to North Atlantic SST and El Niño-like patterns. Furthermore, SST-forced trends in blocking frequency show a consistent decline in winter over Greenland, and an increase over Europe. These results demonstrate that SST variability exerts a significant and physically interpretable influence on blocking frequency and establishes large ensembles from deep learning models as a powerful tool for separating forced SST signals from internal noise.

Figures (10)

• Figure 1: Seasonal atmospheric blocking frequency patterns and model evaluation from 1980 to 2010. Spatial pattern of blocking frequency (fraction of blocked days) for different models (DLESYM, NGCM, HiRAM) and ERA5 reanalysis, shown for (A-D) annual mean, (F-I) December--February (DJF), and (K-N) June--August (JJA). Shading indicates the fraction of blocked days, with contour intervals of 0.2. Panels (E, J, O) show Taylor diagrams comparing model performance against ERA5 in each season, using standard deviation ratios and pattern correlations to quantify spatial agreement, comparing with another 3 CMIP models: CESM2, FGOALS and GISS's AMIP experiments. Models closer to ERA5 indicate better pattern reproduction.
• Figure 2: Temporal correlation between simulated blocking frequency and reanalysis data. Spatial distribution of temporal correlations (1900--2010) between simulated blocking frequency ensemble mean and the 20CR reanalysis ensemble mean for (A-C) annual, (D-F) December--February (DJF), and (G-I) June--August (JJA). Results are shown for 100-member DLESYM (A, D, and G), 100-member NGCM (B, E, and H) and 5-member HiRAM (C, F, and I). Shading indicates local correlation coefficients, with warm (cold) colors representing positive (negative) correlations. Black contours represent climatological blocking frequency from each model. Central circle values denote the domain-averaged correlation, weighted by climatological blocking frequency and grid-point area.
• Figure 3: Greenland blocking frequency and temporal correlation with reanalysis. (A, C, E) Time series of DJF Greenland blocking frequency from DLESYM (A), NGCM (C), and HiRAM (E) models, compared with 20CR (red) and ERA5 (blue). Thin gray lines show individual ensemble members; black lines indicate ensemble means. Dashed black lines represent the 5--95% ensemble spread. (B, D, F) 40-year rolling correlation between model blocking frequency and 20CR. Thin gray lines show correlations from individual ensemble members; the black line shows the ensemble-mean correlation. Dashed lines mark the 5--95% spread across members. The red line shows the rolling correlation of the ensemble mean against 20CR. Horizontal dotted red lines indicate full-period correlations of ensemble means.
• Figure 4: Correlation between Greenland blocking frequency and global sea surface temperatures (SSTs). Spatial correlation patterns between DJF Greenland blocking frequency ensemble mean and detrended SST anomalies during 1900--2010 for (A) DLESYM, (B) NGCM, (C) HiRAM, and (D) 20CR reanalysis. Colors indicate correlation coefficients, with warm (cool) shading showing regions where blocking frequency increases with warmer (cooler) SSTs. Stippling marks regions where correlations are statistically significant ($p < 0.01$).
• Figure 5: Trends in blocking frequency from 1900 to 2010. Spatial distribution of linear trends in ensemble-mean blocking frequency (% per 100 years) for (A-D) annual, (E-H) DJF, and (I-L) JJA seasons. Results are shown for DLESYM (n = 100), NGCM (n = 100), HiRAM (n = 5), and 20CR (n = 1). Shading indicates the magnitude and sign of the trends, with red (blue) areas representing increasing (decreasing) blocking frequencies. Black contours show climatological blocking frequency patterns. Green stippling marks areas where trends are statistically significant ($p < 0.01$).