Fig. 1: Spatiotemporal features of the leading vapor pressure deficit (VPD) mode and its predictability.

A Spatial pattern of the leading VPD mode and corresponding PC time series. The black box indicates the southwest US area analyzed in this study. B Observed and predicted time series for the area-averaged VPD in the southwest US, NINO3.4, and the PDO. Black curves represent observations. Grey curves represent the forecast trajectories of individual forecast members with forecast leads up to 12 months, and colored curves represent the forecast trajectories of the ensemble mean forecasts with different colors denoting different initialization months. For instance, the dark green color represents the forecast trajectory initialized in January with forecast leads up to 12 months. C Ensemble-mean anomaly correlation (AC) skill for area-averaged VPD in the southwest US, NINO3.4, and the PDO, respectively, with forecast leads up to 12 months. The shading indicates the 5-95% confidence interval based on the bootstrapping significance test. D Ensemble-mean AC skill for area-averaged VPD in the southwest US, NINO3.4, and the PDO as a function of the initialization month (vertical axis) and forecast lead times up to 12 months (horizontal axis). AC skill is computed based on the entire 1992–2022 period using verification datasets from ERA5 reanalysis. Forecasts were generated from the SPEAR seasonal forecasting system with 15 ensemble members. Hatching indicates correlation coefficients not statistically significant at the 95% confidence level following ref. ⁶³.