Monthly North Atlantic Sea level pressure reconstruction back to 1750 CE using artificial intelligence optimization

Abstract

Main modes of atmospheric variability exert a significant influence on weather and climate at local and regional scales on all time scales. However, their past changes and variability over the instrumental record are not well constrained due to limited availability of observations, particularly over the oceans. Here we couple a reconstruction method with an evolutionary algorithm to yield a new 1° × 1° optimized reconstruction of monthly North Atlantic sea level pressure since 1750 from a network of meteorological land and ocean observations. Our biologically inspired optimization technique finds an optimal set of weights for the observing network that maximizes the reconstruction skill of sea level pressure fields over the North Atlantic Ocean, bringing significant improvements over poorly sampled oceanic regions, as compared to non-optimized reconstructions. It also reproduces realistic variations of regional climate patterns such as the winter North Atlantic Oscillation and the associated variability of the subtropical North Atlantic high and the subpolar low pressure system, including the unprecedented strengthening of the Azores high in the second half of the twentieth century. We find that differences in the winter North Atlantic Oscillation indices are partially explained by disparities in estimates of its Azores high center. Moreover, our reconstruction also shows that displacements of the summer Azores high center toward the northeast coincided with extremely warm events in western Europe including the anomalous summer of 1783. Overall, our results highlight the importance of improving the characterization of the Azores high for understanding the climate of the Euro-Atlantic sector and the added value of artificial intelligence in this avenue.