Calibrating anomalies improves forecasting of daily reference crop evapotranspiration

Abstract

• We develop a new calibration strategy to enhance ET o forecasting. • This strategy improves the representation of ET o seasonality. • This strategy adds skills to calibrated ET o forecasts. • We validate the effectiveness and robustness of the strategy. Forecasting of short-term reference crop evapotranspiration (ET o ) provides valuable information for hydrological, agricultural, and ecological applications. ET o forecasts can be derived from weather forecasts of Numerical Weather Prediction (NWP) models, but such raw forecasts need to be calibrated to correct errors and improve reliability. This study calibrates the short-term ET o forecasts constructed with weather forecasts from the Australian Bureau of Meteorology's Australian Community Climate and Earth-System Simulator G2 version (ACCESS-G2) model, using the recently developed Seasonally Coherent Calibration (SCC) model. The monthly parameterization of the SCC model will not be able to capture the increasing or decreasing trends in ET o at the submonthly scale, posing a challenge for effective forecast calibration. To address this challenge, we developed a new calibration strategy based on ET o anomalies and climatological mean. We thoroughly evaluated this strategy at both the continental and weather station scales. Results indicate that calibrating ET o anomalies improves the correlation coefficient between calibrated forecasts and observations by up to 10%, and increases forecast skill scores by up to 200%, with more significant improvements found at longer lead times (9-day lead time vs. 1-day lead time). Improvements in forecast quality in calibrations across two spatial scales based on different types of observations (gridded observations vs. weather station observations) validate the effectiveness and robustness of the developed strategy. We anticipate that this strategy will be applicable to other calibration models to enhance NWP-based ET o forecasting.