Evapotranspiration (ET) describes the sum of water transfer from the ground surface through soil evaporation and water loss from leaf stomata into the atmosphere − critical factors linking the global water and carbon cycles. Myriad ET models based on remote sensing data provide spatially continuous estimates of ET; however, leaf photosynthetic information is critical to ensure accurate ET estimates, which are difficult to measure from space. Remotely sensed sun-induced chlorophyll fluorescence (SIF) provides a proxy of stomatal conductance activity with high performance in predicting plant transpiration, which can account for a large proportion of terrestrial and riverine ET. This study aims to improve estimates of tree water use in semi-arid to arid environments. In this study, a fixed stomatal conductance model and three SIF-driven canopy conductance (g_sc) models were applied to model potential ET (PET). The models estimated PET using the Penman-Monteith equation with: (1) a constant leaf stomatal conductance; (2) a transpiration-driven g_sc model; (3) a g_sc model based on electron-transfer rate and vapor pressure deficit, and a (4) Ball-Berry stomatal conductance model. A machine learning model was then applied to scale PET to actual ET (AET) using remote sensing and climate data. Accordingly, four AET models were cross-validated with in-situ measured AET at 52 sites, including 21 eddy covariance flux tower sites, and 31 sap-flow measurement sites (semi-arid and plantation area), for various plant functional types in Australia. This study demonstrated that SIF effectively captured seasonal variations of g_sc, finding that AET models with SIF-driven g_sc models correlated well with in-situ measured AET (R² = 0.64). Modelled AET with dynamic variations of g_sc generated lower prediction error (0.85 mm day⁻¹), while the AET model with fixed stomatal conductance tended to overestimate AET in floodplains and underestimate it in evergreen broadleaf forests, indicating using fixed stomatal conductance results in unstable performance when modelling AET. This study demonstrated that SIF-driven AET models improved broadscale estimation of ET. Our findings provide vital broadscale hydrological data to assist catchment and regional water management, particularly over unmonitored areas at risk of future climate-driven reductions in rainfall.