<p>Accurate estimation of reference evapotranspiration (ET₀) is critical for enhancing sustainable agricultural development of water-limited regions, such as Tabriz, Iran. The conventional meteorological models often fail in data-sparse situations due to insufficient input parameters. To overcome this limitation, an original and biologically inspired ensemble machine learning approach, Biological Ensemble Support Vector Regression (BE-SVR), was developed to synergistically fuse terrestrial meteorological data with remote-sensing indices derived from MODIS including land surface temperature (LST), normalized difference vegetation index (NDVI), leaf area index (LAI), fraction of photosynthetically active radiation (FPAR), and enhanced vegetation index (EVI). In order to handle missing data and generate a continuous time series from the multi-day data dataset provided by MODIS, Kalman filtering and cubic spline interpolation was used. The BE-SVR model’s performance was evaluated against the standard FAO-56 Penman–Monteith (FPM) equation for ET₀. The results show that BE-SVR achieved exceptional accuracy (RMSE = 0.159 mm day⁻¹, R² = 0.998), outperforming a Firefly Algorithm-optimized SVR (FFA-SVR) with RMSE = 0.226 mm day⁻¹, R² = 0.995 and a conventional SVR (RMSE = 0.467 mm day⁻¹, R² = 0.980). BE-SVR integrates different SVRs trained on separate feature subsets in which its outputs in the kernel space are combined, without an explicit incorporation of combining the SVRs outputs, which is a novelty that improves generalization while avoiding weighting bias. These findings demonstrate that incorporating remote sensing data with novel ensemble architecture substantially improves ET₀ prediction. The proposed approach provides an accurate, robust and data efficient modeling framework to support irrigation planning and climate adaptation strategies in water-limited agricultural regions.</p>