Winter-Chill Attribution and CMIP6 Projections of ENSO-Driven Olive Yield Collapse on the Hyper-Arid Peruvian Coast

Olive (Olea europaea L.) orchards on the hyper-arid Peruvian coast (Tacna, 18∘ S) suffered >70% yield collapses in the 2016 and 2024 El Niño seasons against a non-failure mean of ∼6 t ha−1 and a 2022 La Niña bumper harvest, raising the question of whether insufficient winter chilling is the binding climate constraint. We combined in situ daily meteorology (2015–2025) with yield records from eleven Sevillana–Ascolana parcels (88 parcel-years over eight seasons) and fitted a year-level log-OLS model with mean chill-window and fruit-growth temperatures, validated by year-block bootstrap, permutation, a closed-form Bayesian posterior, and a parcel-year mixed model. The model achieves Rlog2=0.65, and the chill slope (β=−0.82) is robust across three independent tests: one-sided permutation p=0.036; Bayesian posterior with 99.8 % of mass below zero (Savage–Dickey BF10=15.9); parcel-year mixed model p<10−14. Counterfactual restoration of chill-window temperature to its non-failure climatology recovers the full collapse in both years, whereas restoring fruit-growth temperature recovers nothing. CMIP6 delta-method projections identify a chill-collapse threshold at ΔTwinter≈+1.25∘C; SSP1-2.6 alone reduces mid-century mean yield by ∼52%, and SSP5-8.5 reaches −89% by 2051–2070. Tacna emerges as a chill-sentinel system where winter warmth, not summer heat, is the binding constraint and the transition to the failure regime lies on a near-term adaptation horizon.