Sub-seasonal to seasonal climate prediction is vital for agricultural planning. In this study, we evaluate the contribution of land sources to the soil moisture forecast skill using the Community Earth System Model version 2 (CESM2) SubX forecast experiments. The soil moisture forecast skills in the control experiment that incorporated all three components’ initializations: Ocean, Land, and Atmosphere, are compared with three sensitivity experiments: (1) ocean and land initializations with climatological atmosphere conditions, (2) land only initializations along with climatological ocean and atmospheric conditions, and (3) ocean and atmospheric initializations with climatological land conditions. The root zone (0-1m) soil moisture forecast is compared with ERA5 soil moisture, the reanalysis dataset based on numerical integrations of the ECMWF land surface model. The soil moisture anomaly correlation shows that land-only initialization contributes 77.1% of the total skill in the control experiment. The soil moisture forecast skill is greatest in Great Plain and central southern US. Furthermore, we verify the land sensitivity experiment with a counterfactual experiment where the land's initial condition is fixed to its climatological condition, which (control minus experiment) shows the major contribution of land sources to the soil moisture forecast skill in the agricultural regions. Finally, we compare CESM2 soil moisture forecast skills with two other SubX climate models: ESRL-FIM and RSMAS-CCSM4, and identify soil moisture to precipitation feedback as a potential area for improving the CESM2 forecast skill.