Asymmetric ENSO impacts on European climate extremes identified by a kernel-based framework

Teleconnections play a fundamental role in shaping global climate variability and the occurrence of extreme events. The El Niño–Southern Oscillation (ENSO) is one of the most influential large-scale modes, with well-documented impacts on the global climate system. Although ENSO exerts only a modest influence on European seasonal climate, previous studies suggest that a link emerges during late winter. This period is of particular relevance for agriculture, as anomalously warm conditions can trigger early crop development and thereby increase vulnerability to subsequent cold extremes such as spring frosts. As warm winters are projected to become more frequent under future climate change, understanding the large-scale drivers of these conditions is increasingly important for mitigating socio-economic impacts on agriculture. While the general relationship between ENSO and European late-winter climate has been widely studied, the specific role of ENSO in triggering anomalous warm conditions that initiate early-season agricultural risk has not yet been systematically assessed. Establishing this statistical linkage will provide valuable insights for impact assessment and could improve the predictability of climate-related risks.

To assess teleconnection interactions, dimension-reduction techniques such as Empirical Orthogonal Functions and Canonical Correlation Analysis (CCA) are among the most widely used approaches. However, these methods are inherently linear and typically restricted to interactions between two spatial fields, which limits their ability to capture complex nonlinear dependencies. Here, we introduce a novel dimension-reduction framework designed to identify nonlinear interactions among multiple climate variables. The approach integrates kernel generalized CCA with multiple kernel learning and preimages, enabling the extraction of spatially interpretable coupled climate patterns that can serve as a basis for defining teleconnections. By employing an automatic kernel-selection procedure, the framework captures both linear and nonlinear dependencies among the analysed climate variables. We apply this methodology to assess the influence of ENSO on European temperature and water balance anomalies and benchmark the results against a purely linear formulation using the Twentieth Century Reanalysis, version 3 (20CRv3), over the period 1900–2015.

Our results show that the nonlinear framework identifies a substantially larger fraction of Europe being influenced by ENSO than is suggested by linear approaches. The ENSO signal exhibits a pronounced asymmetry across the distributions of temperature and water balance anomalies, with lower and upper extremes responding in different ways. In particular, the upper percentiles of temperature, representing warm and hot extremes over most of Europe, including central Europe, show a clear ENSO-related signal associated with La Niña events that are preceded by El Niño Modoki conditions. The Central European region is of high relevance for agricultural production, suggesting that non-linear ENSO effects may play an important role in shaping early-season climate risk. On the other hand, water balance anomalies primarily respond in the central part of the distribution and are mainly linked to El Niño events. Both signals are significantly weaker or absent in classical linear analyses. Overall, these findings highlight the added value of nonlinear methods for revealing previously hidden teleconnection impacts and point to the benefits for improving climate risk assessment and seasonal prediction.