Daily reference evapotranspiration (ET0) is critical for the agricultural sector, particularly for effective irrigation planning and water resource optimization. ET0 is traditionally calculated using the FAO-56 Penman-Monteith method, but its extensive data requirements make it impractical in many regions, especially in developing areas with limited weather monitoring infrastructure. This study explores machine learning (ML) approaches - Random Forest (RF) and Long Short-Term Memory (LSTM) models - to estimate daily ET0 in Casablanca, Morocco. The research compares model performance using two datasets: a complete set with temperature (Tmax and Tmin), relative humidity (RH), wind speed (WS), and surface pressure (P) measurements, versus a minimal set using only Tmax, Tmin and RH data, the minimum data typically available in most areas. The dataset, spanning four decades (1981–2024), was sourced from the NASA POWER Project, ensuring robust and reliable data for model training and validation. Results demonstrate high accuracy for both models when using complete data, with maintained efficacy even under limited data conditions. These findings suggest ML offers a viable alternative for ET0 estimation in regions with restricted data availability, supporting improved irrigation planning and sustainable water resource management.