REFERENCE EVAPOTRANSPIRATION IN THE MUNICIPALITY OF INHAMBANE, MOZAMBIQUE: EMPIRICAL METHODS AND SUPPORT VECTOR MACHINE Evapotranspiration Section AGRICULTURE

Abstract

Reference evapotranspiration (ETo) is useful for calculating crops water demand and irrigation scheduling. Thus, eleven empirical methods for ETo estimation (four based on temperature, three on mass transfer and four on solar radiation) and sixteen on Support Vector Machine models: SVM (eight with the Radial Basic Kernel function: RBFK and eight with the Kernel polynomial: PK) were evaluated against the Penman Montheith FAO 56 method (PM FAO 56) based on the following statistical indicators: MBE (Mean Bias Error), RMSE (Root Mean Square Error), R² and t test. The RMSE was used as the main selection criterea of methods/ models. Monthly meteorological data of Inhambane Municipality (maximum, minimum and average temperature: T, relative humidity: RH, wind speed and sunshine hour: n), from 1985 to 2009 were used. The results showed the following classification: RBFK7 = PK7 > PK7 > PK6 > PK6 > RBFK8 = PK8 > Makm > PK3> Turc = RBFK3 > RBFK1 = RBFK5 = PK5 > BenL = RBFK4 = 24 > Ham > Lin > Pen > Mah > Wmo. The RBFK7 (MBE = 0.26 mm d^-1, RMSE = 0.36 mm d^-1 and R² = 0.96) and PK7 (MBE = 0.26 mm d^-1, RMSE = 0.36 mm d^-1 and R² = 0.96) models require the measurement of T, RH and n. In the absence of n, which is common, the PK3 (MBE = -0.02 mm d^-1, RMSE = 0.42 mm d^-1 and R² = 0.83) and RBFK1 (MBE = -0.01 mm d^-1, RMSE = 0.52 mm d^-1 and R² = 0.74) models can be used as alternatives, which require T and RH data; and T, respectively. Contrary to SVM, the empirical methods were statistically different from the PM FAO 56 method at 5% significance level.