Using Gene-Expression Programming method and geographical location information to simulate evapotranspiration in Hunan and Hubei Provinces

Both Hunan and Hubei Provinces are major agricultural regions. Rice production is not only related to food security in the two provinces, but also importantly influences food security in China. Water resources in the two provinces will further decline due to the South-North Water Transfer project. Reference crop evapotranspiration (ET₀) is the key factor for estimating vegetation evapotranspiration. Accurate estimation of ET0 is essential for water resources management and irrigation schedule. The adapted FAO-56 Penman Monteith (P-M) equation has been recommended as the reference equation for estimating ET₀ and for calibrating other ET₀ equations. The main drawback of using the P-M equation is the requirement for a range of meteorological inputs (air temperature, relative humidity, solar radiation and wind speed). However, the number of meteorological stations is limited even in developed countries, where meteorological variables are more accurately measured. As ET₀ is correlated with geographical location, this study investigated the suitability of Gene-Expression Programming (GEP) technique for modeling ET₀ using readily available geographical location information for Hunan and Hubei Provinces. Monthly observation data for 1955 2005 from 46 stations in Hunan and Hubei Provinces were used. The dataset, including monthly maximum temperature, minimum temperature, average wind speed, sunshine duration and relative humidity, were used to model ET₀ based on the FAO-56 P-M equation as the reference equation. While the GEP was trained using latitude, longitude, altitude variables and month count as input, monthly ET₀ was as output. The GEP model proved to have an adequate precision, with the coefficient of determination (R²) and root mean square error (RMSE) for the validation and test analyses of 0.934, 0.951 and 10.050 mm, 8.628 mm, respectively. Through comparison with the Hargreaves and Priestley-Taylor methods, the GEP model had the lowest RMSE values (8.628 9.967 mm). As the GEP technique could produce a simple and explicit mathematical algorithm, irrigation technicians in data-poor regions could use the GEP model to easily estimate ET₀ with adequate precision. It was inferred that ET₀ could be calculated using geographical location information in Hunan and Hubei Provinces. The GEP model could simplify monthly irrigation schedule and vegetation evapotranspiration estimation.