Accurate estimation of aboveground biomass (AGB) in agroforestry systems is critical for assessing carbon sequestration potential and supporting climate mitigation strategies. This study evaluates the performance of public (Sentinel-2, Landsat) and commercial (Pleiades, PlanetScope) satellite datasets in estimating AGB for olive (Olea europaea) and cypress (Cupressus sempervirens) trees within the INNO4CFIs Living Hub in Santa Paolina, Follonica, Italy. Field data, including tree diameter and height, were collected across eight plots and converted to AGB using species-specific allometric equations. Satellite-derived spectral bands and indices were analyzed using machine learning models (Linear Regression, SVM, Random Forest, Gradient Tree Boost) to predict AGB density (tons/ha). Results revealed that no single satellite or model universally outperformed others. Pleiades (0.5 m resolution) achieved the highest accuracy for olive trees (R2 = 0.442, Gradient Tree Boost), while Sentinel-2 excelled for low-AGB cypress (R2 = 0.785, SVM). PlanetScope showed mixed performance, and Landsat’s lower resolution limited its utility. Species-specific AGB density distributions further influenced model efficacy, with cypress requiring separate models for high- and low-biomass clusters. The findings emphasize that satellite selection must consider vegetation type, density, and structural characteristics. This study underscores the value of integrating multi-source satellite data and machine learning for agroforestry monitoring, offering actionable insights to enhance carbon farming initiatives and advance remote sensing strategies in sustainable land management.