Atmospheric stability regulates wind shear, turbulence, and turbine power output, so its misrepresentation by mesoscale models can bias wind resource estimates. Previous studies assessed stability effects on energy yield from either observations or simulations, mostly at mid-latitude sites. This study applies the same stability-segregated energy calculation to both at Jandaíra, in the semi-arid region of Northeast Brazil, during the dry season (June–December 2024). A 173 m meteorological tower was compared with WRF v4.3 simulations (NCAR TROPICAL suite, YSU boundary layer scheme, 2 km grid). Stability was classified with the bulk Richardson number and the Monin–Obukhov length, and energy production was estimated for the NREL 5 MW and DTU 10 MW reference turbines. Observations show a predominantly stable 10–100 m layer following the diurnal cycle and a predominantly unstable 100–130 m layer. WRF reproduces the lower-layer diurnal pattern but overestimates nighttime stability above 100 m; wind speed RMSE decreases from 2.76 m/s at 10 m to 1.87 m/s at 150 m. Simulated energy production is 35.5–39.5% lower than observed. Stability segregation changes the estimate in opposite directions (−14.3 to −14.7% observed; +10.0 to +11.0% simulated), so a correction based on simulated stability may bias energy yield estimates in semi-arid tropical regions.