Evapotranspiration (ET) is a fundamental variable of the hydrological cycle. Estimation of ET at regional scale from remote sensing data provides spatially distributed information that can improve ground water assessment. In particular, remote sensing can provide information on vegetation phenology which is strongly related to the capacity of vegetation to extract water from the ground over long period of time. We tested the interest of including such information in the modelling of the stream flow at the outlet of a karst system. The test site is the impluvium of Fontaine de Vaucluse karst aquifer, located in the South-East France. The spring is the only outlet of a 1162 km² catchment area. The hydrologic regime of the system is described using the Karst Recharge Model (Ollivier et al., submitted) based on the assumption that karst recharge is mainly controlled by climate and surface and sub-surface catchment heterogeneities. We have developed a semi-distributed model based on lumped structure in order to compute water transfer thought the karst aquifer. In the original version, the epikarst is represented as simple buckets with known water holding capacity (spatially variable as a function of soil type) and a simple water balance is performed to compute drainage from rain and evapotranspiration. The latter is computed using the reference ET based on the Penman-Monteith equation (ET-PM). In the present work we replace ET-PM by a new calculation (ET-RS) accounting for the amount of vegetation through a vegetation index derived from MODIS data (Kc-ETref). The rain regime is also accounted for better describe the impact of evaporation from the superficial soil layers. The new ET assessment based on ET-RS yields more consistent ET temporal dynamics than ET-PM-based approach. The calculated discharge was also improved (Nash Index rises from 0.58 to 0.69).