Grain yield data is frequently used for precision agriculture management purposes and as a parameter for evaluating agronomy experiments, but unexpected challenges sometimes interfere with harvest plans or cause total losses. The spatial detail of modern grain yield monitoring data is also limited by combine header width, which could be nearly 14 m in some crops.  Remote sensing data, such as multispectral imagery collected via satellite and unmanned aerial systems (UAS), could be used to estimate yield at finer spatial resolutions to extract value from data that may be too spatially coarse for research or ambitious precision ag management goals. Yield estimates at finer spatial resolutions could allow practitioners to more easily match data used to generate variable rate prescriptions with the full hardware capabilities of machinery such as sprayers with individual nozzle control. There is further potential to derive yield estimates that spatially conform to experimental unit sizes independent of combine header width constraints. Therefore, the objectives of this study were to 1) examine the relationship between maize grain yield monitoring data and multispectral UAS imagery, 2) determine the accuracy of simple linear regression and machine learning (ML) yield estimation models, and 3) evaluate whether including covariates such as soil apparent electrical conductivity (EC_a) mapping with UAS data improves yield estimation accuracy. UAS imagery was collected with a RedEdge-3 (MicaSense, Seattle, WA), which has five spectral bands centered at 475, 560, 668, 717, and 840 nm, and processed with Pix4Dmapper software to generate calibrated reflectance orthomosaics as well as Normalized Difference Vegetation Index, Normalized Difference Red Edge (NDRE), and Red Edge Chlorophyll Index. A Veris MSP3 (Veris Technologies, Salina, KS) was used to create shallow and deep EC_a, cation exchange capacity, and soil texture maps calibrated with grid soil samples. With cleaned grain yield monitor data as the centroid points, rectangular polygons were generated to characterize the field area harvested by the combine during the yield monitor logging interval. Finally, the polygons were used to extract mean values for reflectance, vegetation indices, and calibrated EC_a data within the spatial extent (area) of each yield monitor observation. There was a strong linear relationship between grain yield and Normalized Difference Red Edge (R² = 0.8). Comparison of linear regression and ML regression (Random Forest Regression - RFR) indicated that RFR better fused the data to estimate grain yield (average R² = 0.92 ± 0.03). RFR reduced mean absolute error in both the training and test data to 9.9 and 17.8 bu/ac, respectively. Vegetation indices and reflectance were the most important features and layers such as deep EC_a provided minimal improvements for yield estimation at this site.