Comparison of simple statistical model and Brunt-type model to predict daily minimum temperature for spring frost protection

Abstract

We constructed a simple statistical model to predict the minimum temperature in spring from evening dew point temperature (T_d) and relative humidity (H) data, and then compared it with a Brunt-type semi-empirical heat balance model. In general, statistical models are considered to have only local validity and to relate well only to calm, clear skies. However, there has been a lack of detailed analyses on the universal validity of these statistical models and the effects of weather conditions on them. We used hourly meteorological data collected in March, April, and May in Aomori, Yamagata, Fukushima, and Matsumoto to construct both models. The minimum temperature on calm, clear nights (T_m) was expressed as T_m = T_d-10.29 ln (H)+42.0. The equation was accurate at all 4 locations, with a low root mean square error (RMSE) of 1.39°C. On all calm nights, the RMSE was 1.80°C. Under all weather conditions, the RMSE was 2.16°C. Two important factors in the Brunt-type model-the net longwave radiation and the thermal inertia of soil-were empirically determined from data collected on calm, clear nights. The Brunt-type model gave less accurate results, with an RMSE of 1.88°C. Applying the statistical model with all-weather parameters at 18 locations nationwide in April 2001, when severe frosts occurred, gave RMSE values of 1.5 to 2.7°C. These results indicate that the statistical model has good universal validity, with more accurate estimates of minimum temperatures than the Brunt-type model under calm, clear skies, and relatively good results under all-weather conditions.