Evaluation of Reference Evapotranspiration Estimation Methods under Southeast Australian Conditions

Reference evapotranspiration (${\mathrm{ET}}_o$) estimates are often required for use in water resources planning and irrigation scheduling. Ten ${\mathrm{ET}}_o$ estimation methods ranging from simple temperature-based to data-extensive combination methods, including Hargreaves (HAR), improved Hargreaves (IHA), FAO-24 Radiation (RAD), Ritchi-type (RIT), FAO-24 Class-A Pan with pan coefficients of Doorenbos and Pruitt (PEV) and empirical regression coefficient (SEV), combination methods McIlroy (McI), FAO-Penman with wind functions of Watts and Hancock (W_H) and Meyer (M_PY), and the Penman-Monteith (P_M) were evaluated at three sites, namely, Aspendale, Griffith, and Tatura in the Goulburn-Murray Irrigation Area (GMIA) of southeastern Australia. At Aspendale, 4 out of 10 ${\mathrm{ET}}_o$ methods (McI, M_PY, SEV, and RAD) overestimated the ${\mathrm{ET}}_o$ estimates; at Griffith no method overestimated them, whereas at Tatura only the RAD method overestimated ${\mathrm{ET}}_o$. The overestimations were at Griffith, McI (1%), M_PY (10%), and SEV (4%); at Tatura, RAD (2%). At the Griffith and Tatura sites, almost all methods showed a strong tendency to underestimate daily ${\mathrm{ET}}_o$ estimates throughout the entire range of evaporative demand. Overall, the underestimation ranges observed were McI (12–27%), W_H (7–22%), RIT (6–25%), PEV (19–31%), HAR (18–31%), and IHA (8–11%). The underestimation of daily ${\mathrm{ET}}_o$ estimates by the P_M method ranged from 21 to 29%, raising caution about its use as a base method (without calibration against measured data under local conditions) to evaluate other ${\mathrm{ET}}_o$ methods, as has been advocated in recent literature. The use of the McI method as the top-ranked method at Aspendale and Tatura, and the W_H method at Griffith, indicated that no single daily ${\mathrm{ET}}_o$ estimation method using meteorological data was satisfactory for all three sites. Generally, the combination methods proved to be the most accurate ${\mathrm{ET}}_o$ estimates. At Tatura, the fact that the RAD method was ahead of the W_H and M_P combination methods indicates how a less data-intensive ${\mathrm{ET}}_o$ method, if calibrated, can perform even better than a physically based combination method. All ${\mathrm{ET}}_o$ estimation methods required local calibration against measured lysimeter ${\mathrm{ET}}_o$ data for better performance.