The North Australian Tropical Transect (NATT) is one of the research transects for tropical savannas in the world. It crosses about 1000 km with decreasing annual rainfall from northern 1700 mm to southern 300 mm, and is of great significance for global carbon sink anomaly. Across NATT, drought constrains the biomass of plants by regulating transpiration and respiration. In recent years, drought has been evaluated by many indexes considering evapotranspiration. In this research, we adopted eddy covariance data and meteorological observations from Australia Water Availability Project in five research sites, respectively Howard Springs, Daly River Uncleared, Dry River, Sturt Plains and Alice Springs to estimate potential evapotranspiration based on Penman–Monteith, Thornthwaite, and Priestley-Taylor methods. Then we calculated Climatic Water Deficit (CWD) which is the difference between the potential evapotranspiration and the actual evapotranspiration, and compared the relationship between CWD and soil moisture from flux tower measurements. The results showed that the Pearson correlation coefficients between Priestley-Taylor estimations and Penman–Monteith reference evapotranspiration were 0.84~0.98 in the five sites, which were higher than that between temperature-based Thornthwaite and Penman–Monteith references. The average monthly estimation errors of Priestley-Taylor were 18.9 ~26.2 mm comparing with Penman–Monteith references, that Priestley-Taylor method can effectively estimate the potential evapotranspiration across NATT with less meteorological arguments. Meanwhile, the average monthly estimation errors of Thornthwaite method increased with the decreasing precipitation, from 29.4 mm in Howard Springs to 78.8 mm in Alice Springs with annual rainfall 1714 mm and 318 mm respectively, which illustrated that the Thornthwaite method is not suitable for arid areas. Furthermore, Penman-Monteith based and Priestley-Taylor based CWD had the significantly negative correlation with soil moisture, especially in mesic areas where the correlation coefficients reached -0.8 that the CWD can be used for drought monitoring.