Lead-time and cumulative-error evaluation of NBM-derived reference evapotranspiration forecasts using CIMIS records in California

Reliable reference evapotranspiration (ETo) forecasts can inform planning for atmospheric water demand. This study evaluated National Blend of Models (NBM)-derived daily ETo against California Irrigation Management Information System (CIMIS) station-based ETo records at 132 stations from 1 January 2021 through 11 June 2026. The analysis combined lead-dependent daily performance, seasonal and spatial cumulative bias, and rolling cumulative fixed-lead errors, using 714,041 station–lead–date pairs at leads of 1, 3, and 7 days. Pooled RMSE was 0.70, 0.79, and 0.95 mm day⁻¹ at 1-, 3-, and 7-day leads, respectively, while R² decreased from 0.91 to 0.88 and 0.83. Station-clustered confidence intervals supported the increase in absolute error with lead time. Compared with a prospective persistence benchmark based on the most recent fully available CIMIS ETo, NBM reduced RMSE by 36%, 36%, and 27% at 1-, 3-, and 7-day leads, respectively. Relative to leave-one-year-out station-month climatology, RMSE reductions were 32%, 24%, and 8%. Errors were smallest in winter and largest in summer. Cumulative station bias varied substantially among complete years, but the overall ordering of forecast performance with lead time remained unchanged when the highest-error year was excluded. For 3-day rolling windows, absolute cumulative ETo error exceeded 5 mm in 1.89%, 2.54%, and 4.35% of windows at 1-, 3-, and 7-day leads, respectively; for 14-day windows, the corresponding frequencies were 36.18%, 42.37%, and 46.36%. Because each rolling window combined same-lead forecasts for consecutive target dates from different initializations, these values describe accumulated fixed-lead forecast error rather than a single multi-day forecast sequence. NBM-derived ETo provided useful prospective guidance on atmospheric water demand, with the strongest performance at shorter lead times, and can serve as an input to local irrigation planning when combined with crop- and field-specific information.