Influence of synoptic patterns (NAO vs. WeMO) on rainfall isotopic composition in SE Iberia: A machine learning approach

Abstract. Atmospheric circulation controls moisture transport across the western Mediterranean, but its seasonally resolved imprint on precipitation isotopes in many areas remains poorly understood, limiting the integration of circulation diagnostics with modern isotope monitoring. Southeastern Iberia constitutes an outstanding natural laboratory to identify and quantify the roles and isotopic footprint of the main modes of atmospheric variability in the region (i.e., the Western Mediterranean Oscillation (WeMO) and the North Atlantic Oscillation (NAO)) and other local factors, such as altitude and precipitation amount. Here, we combine a multi-altitude (560–1800 m a.s.l.) precipitation-isotope network in Sierra de Segura, a mountain range located in SE Spain with explainable machine-learning methods to quantify how large-scale circulation and precipitation regime control rainfall isotopic composition.

First we extend the WeMO index forward from 2020, when instrumental measurements end, to 2025 using a physically constrained XGBoost reconstruction based on regional sea-level pressure predictors from San Fernando (Spain) and multiple northern Italian stations, reproducing the published index over 2010–2020 with robust cross-validated performance (R² = 0.85 ± 0.05; RMSE = 0.41 ± 0.07, n = 129 months). Then we analyze 448 rainwater samples collected between 2017 and 2023, aggregated into 154 precipitation-weighted monthly observations of δ¹⁸O, δ²H and δ¹⁷O, together with d-excess and ¹⁷O-excess. Bulk isotope ratios co-vary strongly (ρ > 0.95), allowing δ¹⁸O to represent the dominant network-scale signal.

Random Forest models interpreted using SHAP reveal a clear seasonal reorganization of controls. During the wet season (October–March), δ¹⁸O variability is primarily circulation-driven, with the North Atlantic Oscillation acting as the dominant control and the WeMO providing a consistent secondary modulation. In contrast, during the dry season (April–September), δ¹⁸O is governed by precipitation regime, with precipitation amount overwhelmingly predominating over circulation indices and exhibiting a strongly non-linear, amount-effect response. Back-trajectory clustering of regionally coherent isotope-sampling events supports these contrasting seasonal influences, indicating a small number of recurrent synoptic transport regimes in winter and weaker synoptic organization in summer.

Overall, these results provide a transferable, seasonally calibration that links synoptic circulation and precipitation regime to rainfall δ¹⁸O (and excess parameters) in southeastern Iberia. This innovative framework enables robust interpretation of Mediterranean isotope-based paleoclimate archives by distinguishing circulation-driven wet-season signals from precipitation-regime-driven dry-season variability. It is readily applicable to other climatically complex areas influenced by multiple modes of atmospheric variability and pronounced seasonal contrasts. Furthermore, our results provide a basis for assessing the sensitivity of cave and lake systems to future changes in circulation persistence and rainfall intermittency under ongoing climate change.