Short-term forecasting of reference evapotranspiration using limited area models and time series techniques

Reference evapotranspiration (ETo) forecast can be obtained using analytical models (Penman Monteith, Penman, etc.) and meteorological data from numerical weather forecasts. The use of both ETo equations and weather forecast makes possible to obtain forecast on hourly base, usually for up to 72 hours, and with different spatial resolution. Spatial resolution of data is a key factor for many applications in regions with high variable landscape characteristics, in particular in case of complex terrain. Evapotranspiration forecast can be also based on time series analysis of ETo and meteorological variables related to evapotranspiration process. The autoregressive integrated moving average (ARIMA) models are the most frequently used time series models. Moreover, artificial neural networks (ANN) were recently applied in time series modeling and forecasting. The main aims of this study were to analyze and compare the performances of the above-mentioned techniques in short-term prediction of hourly and daily ETo. Hourly values of ETo were calculated using the Penman-Monteith equation and hourly weather data from two versions of a mesoscale forecast model (BOLAM) characterized by different spatial resolution (5 and 20 km) and forecast range (36 and 72 hours). Both ARIMA and ANN models were developed using eight years of meteorological data from five meteorological stations of Sardinian meteorological network. Each method was validated using a one-year data set from five locations. The accuracy of prediction from each method was evaluated comparing the forecasts with ETo values calculated using observed weather data from weather stations. The 12, 24, and 36 hour prediction of hourly ETo values from weather variables forecast gave a root mean square error value of about 0.15 mm h-1 with an overestimation ranging from 3% to 6%. Time series models (ARIMA and ANN models) performed less well than the method based on weather forecast.