Maize is Malawi's primary staple crop, cultivated mainly by smallholders on plots under one hectare, where conventional field-based yield assessment is costly and difficult to scale. This study compares Sentinel-2, Landsat 9, and Unmanned Aerial Vehicle (UAV) imagery for maize yield prediction across smallholder fields ranging from 0.06 to 0.89 ha in Zomba District, Malawi. Ground-truth yields were obtained by manual harvesting at the end of the 2024–2025 growing season and standardised to tons per hectare. Sentinel-2 and Landsat 9 scenes were acquired monthly from November to March, while UAV imagery was captured in March 2025 at a ground sampling distance of 3.67 cm. All processing and feature extraction were performed in QGIS using open-source tools. Vegetation indices and spectral band values were extracted per field and used as predictors in separate linear regression models for each sensor, evaluated using R², RMSE, and MAE. UAV imagery achieved the highest accuracy (R² = 0.9991; RMSE = 0.0135 t/ha), followed by Sentinel-2 (R² = 0.4063; RMSE = 0.3537 t/ha) and Landsat 9 (R² = 0.2331; RMSE = 0.4020 t/ha). The results illustrate a trade-off between spatial resolution and operational scalability: UAV data capture within-field variability most precisely but demand greater cost and expertise, whereas freely available satellite imagery supports broader, repeatable monitoring. The study demonstrates a fully open and reproducible workflow for affordable crop yield estimation in Sub-Saharan smallholder systems.