Prediction of daily reference evapotranspiration by a multiple regression method based on weather forecast data

Abstract

Prediction of daily reference evapotranspiration (ET 0) is the basis of real-time irrigation scheduling. A multiple regression method for ET 0 prediction based on its seasonal variation pattern and public weather forecast data was presented for application in East China. The forecasted maximum temperature (T max), minimum temperature (T min) and weather condition index (WCI) were adopted to calculate the correction coefficient by multilinear regression under five time-division regimes (10 days, monthly, seasonal, semi-annual and annual). The multiple regression method was tested for its feasibility for ET 0 prediction using forecasted weather data as the input, and the monthly regime was selected as the most suitable. Average absolute error (AAE) and root mean square error (RMSE) were 0.395 and 0.522 mm d−1, respectively. ET 0 prediction errors increased linearly with the increase in temperature prediction error. A temperature error within 3 K is likely to result in acceptable ET 0 predictions, with AAE and average absolute relative error (AARE) <0.142 mm d−1 and 5.8%, respectively. However, one rank error in WCI results in a much larger error in ET 0 prediction due to the high sensitivity of the correction coefficient to WCI and the large relative error in WCI caused by one rank deviation. Improving the accuracy of weather forecasts, especially for WCI prediction, is helpful in obtaining better estimations of ET 0 based on public weather data.