Ground-based active optical sensors (GBAOS) have been successfully used in agriculture to predict crop yield potential (YP) early in the season and to improvise N rates for optimal crop yield. However, the models were found weak or inconsistent due to environmental variation especially rainfall. The objectives of the study were to evaluate if GBAOS could predict YP across multiple locations, soil types, cultivation systems, and rainfall differences. This study was carried from 2011 to 2013 on corn (Zea mays L.) in North Dakota, and in 2017 in potatoes in Maine. Six N rates were used on 50 sites in North Dakota and 12 N rates on two sites, one dryland and one irrigated, in Maine. Two active GBAOS used for this study were GreenSeeker and Holland Scientific Crop Circle Sensor ACS 470 (HSCCACS-470) and 430 (HSCCACS-430). Rainfall data, with or without including crop height, improved the YP models in term of reliability and consistency. The polynomial model was relatively better compared to the exponential model. A significant difference in the relationship between sensor reading multiplied by rainfall data and crop yield was observed in terms of soil type, clay and medium textured, and cultivation system, conventional and no-till, respectively, in the North Dakota corn study. The two potato sites in Maine, irrigated and dryland, performed differently in terms of total yield and rainfall data helped to improve sensor YP models. In conclusion, this study strongly advocates the use of rainfall data while using sensor-based N calculator algorithms.