Abstract
Droughts are a natural occurrence in many small Pacific Islands and can have severe impacts on local populations and environments. The El Niño-Southern Oscillation (ENSO) is a well-known driver of drought in the South Pacific, but our understanding of extreme ENSO events and their influence on island hydroclimate is limited by the short instrumental record and the infrequency of ENSO extremes. To address this gap, we developed the South Pacific Drought Atlas (SPaDA), a multi-proxy, spatially resolved reconstruction of the November–April Standardised Precipitation Evapotranspiration Index for the southwest Pacific islands. The reconstruction covers the period from 1640 to 1998 CE and is based on nested principal components regression. It replicates historical droughts linked to ENSO events with global influence, compares well to previously published ENSO reconstructions, and is independently verified against fossil coral records from the Pacific. To identify anomalous hydroclimatic states in the SPaDA that may indicate the occurrence of an extreme event, we used an Isolation Forest, an unsupervised machine learning algorithm. Extreme El Niño events characterised by very strong southwest Pacific drought anomalies and a zonal SPCZ orientation are shown to have occurred throughout the reconstruction interval, providing a valuable baseline to compare to climate model projections. By identifying the spatial patterns of drought resulting from extreme events, we can better understand the impacts these events may have on individual Pacific Islands in the future.