Abstract
It is important to understand the real-time irrigation demands of wetland plants in areas experiencing ecological water shortages. The Levenberg-Marquardt algorithm (L-M) in an Artificial Neural Network (ANN) was used to train and forecast daily reference crop evapotranspiration (ET0), using a polynomial fuzzy daily precipitation function, based on short-term meteorological predictions. This method allowed simulation of a real-time irrigation schedule based on field water balance, and was applied to a variety of wetland plants including reeds, Typha orientalis and paddy. The results showed that the determinant coefficient of daily ET0 forecast was 0.945. More than 75% of the sampling error was below 10% and 96% of the sampling error was below 20%. The mean-error (ME) and root mean square error (RMSE) for the daily field water level simulated with reeds, T. orientalis and paddy were less than 6.6 mm, while the value of Index of agreement (IA), and RMSE were greater than 0.986 and 0.946, respectively. The difference in ME, RMSE, IA, and Nash-Sutcliffe efficiency factor (NSE) for reeds and T. orientalis was not significant, but it was twice as high for paddy. For real-time irrigation for paddy in the Dianchi Basin, the values of ME, RMSE, IA, NSE, and R were 0.36–0.91mm, 2.85–4.92mm, 0.986–0.996, 0.946–0.985, and 0.938, respectively. The method can meet the needs of regional water resources allocations and operations.
2010 Mathematics Subject Classification: