Answers are grounded in this paper and cite their sources.
Study region This study focuses on Guangdong Province in southern China, a typical humid subtropical monsoon region. Study focus Accurate forecasting of reference evapotranspiration (ETo) is essential for irrigation management, yet raw ensemble forecasts suffer from systematic biases that propagate nonlinearly through ETo models. This study systematically evaluates three statistical post-processing methods—Kalman filter (KF), Bayesian model averaging (BMA), and LightGBM (LGBM). The corrected meteorological variables were used to drive three ETo models with distinct physical mechanisms: FAO56 Penman-Monteith (PM), Hargreaves-Samani (HS), and Priestley-Taylor (PT). A multi-level evaluation framework combining deterministic, probabilistic, and sensitivity analyses was employed to decompose the effects of bias correction versus model structure. New hydrological insights for the region Raw ensemble forecasts exhibit systematic biases across meteorological variables, and the effectiveness of correction methods varies. The PM-based forecasts show the closest agreement with the PM-based reference ETo, and the LGBM–PM combination yields the highest accuracy at 50% of stations (mostly coastal); while the KF–HS combination remains a competitive alternative at many inland stations. Sensitivity analysis reveals that bias correction reshapes the error sensitivity of ETo models: radiation becomes the dominant error source in the PM model after temperature and humidity errors are suppressed. A spatially differentiated strategy is proposed. This framework provides improved deterministic and probabilistic ETo forecasts for subtropical monsoon climates.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: