Agricultural land-use composition modulates hydroclimatic coupling of cropland water use efficiency under prolonged aridification: sub-watershed evidence from Chile's megadrought

Francisco Zambrano , Francisco Fernández, María Molinos-Senante

Cropland water-use efficiency (WUE; net primary productivity divided by actual evapotranspiration) is widely expected to improve under technological intensification, yet its response to prolonged aridification at regional scales is poorly understood. We analyzed spatiotemporal patterns of WUE across 127 Chilean agricultural sub-watersheds from 2001 to 2020, a period spanning an unprecedented megadrought. WUE responses were spatially heterogeneous: arid and semi-arid sub-watersheds showed predominantly negative $\Delta$WUE, while irrigated Mediterranean sub-watersheds showed positive $\Delta$WUE; negative long-term trends and structural breaks concentrated in the 30--35°S band. Precipitation deficit indices dominated cross-sectional WUE sensitivity; at the 36-month accumulation scale, SPEI (Standardized Precipitation-Evapotranspiration Index) and SPI (Standardized Precipitation Index) showed practically equivalent explanatory power (two one-sided test $p$ < 0.001, 90% CI [−0.004, +0.003]). Crop system composition is associated with WUE drought sensitivity through two contrasting patterns, each constrained by a different form of confounding. Annual-crop prevalence is robustly and negatively associated with drought sensitivity across all latitude-control specifications (spatial error model [SEM] $\beta$ range: −0.341 to −0.591, $p$ < 0.01); whether this reflects structural buffering or drought-driven reduction of annual cultivation cannot be established from the observational data alone. Irrigation infrastructure prevalence shows a positive association with WUE-drought coupling that is not distinguishable from the latitudinal aridity gradient: it is significant in the full sample (SEM $\beta$ = +0.438, $p$ < 0.001) but attenuates to non-significance when latitude is explicitly controlled ($\beta$ = +0.117, $p$ = 0.309; $\Delta$AIC = 19.8). Geographic cross-validation confirms that the quantitative crop-composition model does not transfer across latitude bands within Chile (per-block out-of-sample $R^2$ range: −1.25 to −0.29), every held-out block being predicted worse than its own mean, indicating that the coefficients capture a spatially contingent configuration rather than generalizable structural parameters. Both crop-composition associations are correlational; causal inference is precluded by the observational design and spatial confounding. These findings are consistent with crop-portfolio flexibility reducing structural drought exposure under prolonged aridification, and with WUE-based efficiency metrics being insufficient indicators of agricultural resilience where irrigation supply is hydrologically coupled to the same signals that suppress crop productivity. These results carry implications for water governance and agricultural planning in snowmelt-dependent dryland regions globally.