Applicability of meteorological ensemble forecasting to predict summer cold damage in rice growth

Abrupt temperature drops pose serious concerns for rice production in northern Japan. Previous early warning systems have been based on projected temperature tendencies, and alerts have announced for the occurrence of low temperatures. The rice crop has low-temperature-sensitive stages; however, previous systems have not considered them because of the difficulty of simulating rice growth at the local scale. The forecast system would be more valuable by considering both the rice growth stage and current meteorological forecast techniques. In this study, we synthesized ensemble numerical weather prediction and a cultivar-based rice growth model to forecast 14-day cold damage risk. The ensemble mean forecast with nine members predicted surface air temperatures more skillfully for seven days with lower root-mean-square errors (RMSEs) (1.3-1.9°C) than those of the climatological forecast (2.1-2.4°C) that is derived from historical observations over 30 years. The single deterministic forecast predicted the temperatures better for five days with 1.3-2.0°C of RMSEs, showing the extension of the predictable period by two days with ensemble forecasting. For the cooling degree-days, both the single and ensemble mean forecasts showed lower RMSEs than the climatological forecast throughout the forecast period of 14 days (4.1, 3.8, and 5.2°C at the forecast time = 14 day for single deterministic, ensemble mean, and climatological forecasts, respectively). Although the climatological forecast estimated the rice growth stages reasonably, the performance for cooling degree-days was inferior to the ensemble mean and single deterministic forecasts. The meteorological mean state is sufficient to estimate the rice growth stage, but an accurate temporal pattern of the surface air temperature provided by numerical weather forecast is essential for reliable cold damage forecasting. Moreover, ensemble forecasting is more effective than the single deterministic forecast to reduce prediction errors for both the surface air temperature and cold damage.