Emerging climate–yield re-coupling in overexploited date palm oases: satellite evidence from a 22-year temporal decoupling index in southern Tunisia

Tarek Gasmi , Ramzi Guesmi, Slim Ben Abdelbari, Soura Boulaares, Sajeda Albarghati

Southern Tunisia's date-palm oases present a paradox: production has grown roughly three-fold in 22 years (four-fold in Kébili), fuelled by aquifer extraction at more than twice the sustainable rate. Combining the 2002–2024 record of four governorates with GRACE/GRACE-FO terrestrial water storage, reanalysis climate products, ONAGRI production, and CRDA aquifer accounting, we report three findings. First, satellite-based yield-anomaly forecasting fails against a naive per-governorate trend baseline: every learned model returns negative out-of-sample R² under leave-one-year-out cross-validation, and a persistence baseline fails because biennial bearing makes lag-1 anomalies actively misleading. Second, terrestrial water storage falls 16.6 cm between 2002–2005 and 2020–2024, with a sup-Wald structural-break test placing the onset of monotonic depletion at 2007 (95% CI 2006–2008) and a post-break rate of −1.12 cm yr⁻¹. Third, we introduce a temporal decoupling index: the rolling-window slope of the absolute correlation between fruit-development-window vapour pressure deficit (VPD) and yield-per-hectare anomaly. The index identifies Kébili — the most overexploited governorate (229% of renewable resources) — as a locus where VPD–yield coupling is rising (Δ|r| = +0.444; rolling slope p < 0.001 under HAC correction). Tozeur shows the same signature robustly; Gafsa shows it under linear detrending but the two-period contrast does not survive quadratic detrending, so Gafsa is reported tentatively. The Kébili and Tozeur signals survive area normalisation and are robust across break-year, placebo, and four of five detrending alternatives. The focal VPD result emerges from an exploratory scan and does not survive FDR correction in the full grid, but its convergence with GRACE depletion, well-proliferation data, and placebo specificity supports its use as a monitoring signal. The findings reframe satellite earth observation in aquifer-dependent oases: from satellite/climate-based anomaly forecasting (not tractable here) to buffer monitoring.