Improvement of evapotranspiration (ET) estimates using remote sensing (RS) products based on multispectral and thermal sensors has been a breakthrough in hydrological research. In large-scale applications, methods that use the approach of RS-based surface energy balance (SEB) models often rely on oversimplifications of the aerodynamic resistances. The use of these SEB models for Seasonally Dry Tropical Forests (SDTF) has been challenging due to incompatibilities between the assumptions underlying those models and the specificities of this environment, such as the highly contrasting phenological phases or ET is mainly controlled by soil–water availability. We developed a RS-based SEB model from a one-source bulk transfer equation, called STEEP. Our model uses the Plant Area Index to represent the woody structure of the plants in calculating the moment roughness length. In the aerodynamic resistance for heat transfer, the parameter kB-1 was included, correcting it with RS soil moisture. Besides, the remaining λET in endmembers pixels was quantified using the Priestley-Taylor equation. We implemented the STEEP algorithm on the Google Earth Engine platform, using worldwide free data. Four sites with eddy covariance data located in the Caatinga, the largest SDTF in South America, in the Brazilian semiarid region, were used to evaluate the STEEP model. Our results show that STEEP based on the specific characteristics of the SDTF increased the accuracy of ET estimates without requiring any additional climatological information. This improvement is more pronounced during the dry season, which in general, ET for these SDTF is overestimated by traditional SEB models, as happened in our research with the SEBAL. The STEEP model had similar or superior behaviour and performance statistics relative to global ET products (MOD16 and PMLv2). This work contributes to an improved understanding of the drivers and modulators of the energy and water balances at local and regional scales in SDTF.