Management zone delineation is a practical strategy for site-specific management. Numerous approaches have been used to identify these homogenous areas in the field, including approaches using multiple years of historical yield maps. However, there are still knowledge gaps in identifying variables influencing spatial and temporal variability of crop yield that should be used for management zone delineation. The objective of this study is to identify key soil and landscape properties affecting yield spatial patterns and yield temporal stability for management zone delineation using machine learning models and to evaluate the consistence of these factors in different prediction models. The study was carried out in a 44 ha corn-soybean rotation field in western Minnesota, USA. Yield maps from six years collected from 2014 to 2020 (excluding 2019, due to hail damage effects on the final yield) were used to create yield spatial trend (YST; average normalized yield map) and yield temporal stability maps (YTS; coefficient of variation map). More than 30 different soil and landscape properties were used as input in the machine learning models including relative elevation, slope, curvature, aspect and topographic wetness index calculated from LiDAR elevation data at 1 m resolution downloaded from the MN TOPO website; soil brightness index calculated from PlanetScope images at 3 m spatial resolution; soil physical properties, and macro and micronutrients predicted using gamma ray sensor data collected with SoilOptix sensor and ground truth data; and apparent electrical conductivity data. All maps were interpolated to a 3 m grid using kriging. Prediction models for YST and YTS were created using random forest, support vector machine and XGBoost algorithms. Boruta algorithm was used for feature selection. Once features were selected based on importance, spearman correlation was used to exclude features that were highly correlated to each other to avoid redundance in the models. Preliminary results showed that while all features were deemed important, relative elevation was the most relevant factor influencing both YST and YTS. The spatial trend was also highly affected by the slope, soil iron, sulfur and calcium concentrations and soil organic matter. For YTS soil sulfur concentration was the second most important factor, followed by organic matter, and soil copper, iron, and calcium concentrations. Further analysis for the variable selection process is being performed and different machine learning models to improve YST and YTS prediction are being tested.  The implications for management zone delineation will be discussed.