Unraveling volcanic impacts in semi-arid climates: machine learning and CMIP6 insights from Krakatau and Santa Maria

Abstract

Volcanic eruptions are among the strongest natural climate forcings, yet their regional impacts remain poorly resolved, particularly in semi-arid hotspots. Here, we investigate the climatic fingerprints of the Krakatau (1883) and Santa Maria (1902) eruptions on surface temperatures over Iran by integrating CMIP6 multi-model simulations with Random Forest machine learning. This dual framework enables the first systematic assessment of Iran-specific volcanic teleconnection dynamics. Both eruptions triggered prolonged and spatially extensive cooling episodes, characterized by biphasic responses: an intense initial cooling wave followed by a weaker secondary phase linked to aerosol dispersion and decay. Spatial patterns reveal that localized anomalies in northern and western Iran eventually expanded to cover nearly the entire country. Teleconnection analysis highlights the Pacific Warm Pool and Tropical North Atlantic as dominant modulators, with additional contributions from ENSO and the North Pacific, including an anomalous negative TNA phase that amplified Santa Maria’s impact. Our results demonstrate how volcanic cooling is regionally mediated through ocean–atmosphere teleconnections, and reveal systematic model–observation mismatches in intensity and persistence. By bridging physical climate modeling with machine learning, this study provides a novel framework for attributing volcanic impacts in teleconnection-sensitive, semi-arid regions, with implications for climate adaptation, agricultural planning, and regional resilience under future extreme events.