Evapotranspiration (ET) is a fundamental component of the water and energy balance, strongly influencing crop growth and productivity. Accurate ET estimation is critical in semi-arid regions, where water scarcity requires optimized management. Among the available approaches, the Penman-Monteith (PM) model is the most widely used for this purpose, its performance strongly depends on the accurate characterization of surface resistance (r(c)), a key parameter controlling ET estimations. In this study, two approaches for estimating r(c) were evaluated for winter wheat cultivated in the Haouz plain (Morocco) under contrasting irrigation regimes (full and deficit) during the 2016-2017 and 2017-2018 growing seasons. The first is a mechanistic formulation, based on the Jarvis model, which incorporates vapor pressure deficit (VPD) and soil water content (theta) to capture stomatal responses. The second is an empirical approach, using a thermal stress index (SI) derived from land surface temperature (LST), providing a rapid indicator of crop water status. Both approaches were integrated into the PM model and calibrated with eddy covariance data collected over a deficit-irrigated field in 2016/2017, then validated across both irrigation regimes and seasons. Results showed that the mechanistic approach reproduced ET dynamics under full irrigation (R-2 >= 0.73; RMSE < 0.6 mmday(-)(1)), but underestimated fluxes under severe stress. Conversely, the empirical approach, being more sensitive to short-term water status, outperformed under deficit irrigation (R-2 >= 0.79; RMSE < 0.6 mmday(-)(1)). Moreover, a critical SI threshold of 0.5 was identified, which could serve as a practical guideline for irrigation scheduling to reduce water losses. Overall, the results highlight the robustness and complementarity of both approaches and suggest the potential of hybrid models combining physiological realism with thermal sensitivity to improve irrigation management in water-limited areas.