The conventional management approach in agricultural production is applying water and chemicals at a fixed rate throughout the whole field. This practice did not consider the existence of in-field variability in spatial and temporal domains and results in over- and under-applications on crop and serious environment issues such as nitrogen leaking to and contamination of groundwater. Variable-rate technology is the key technology in agricultural production to address the in-field variability, maximize yield and profit, and minimize the agricultural inputs or footprints on the environment. Hence, this project pulls a multidisciplinary effort to address challenges in today's variable-rate technology (VRT) in agricultural production by tightly integrating sensing, networking, AI-based and process-based data analytics, and control systems with classic plant and soil biophysical principles and well-recognized management practices, to provide a generalizable and scalable framework for the real-time in-season variable-rate application. Meanwhile, it also improves the data analytics and decision-making models by turning the massive amount of data generated in daily agricultural production into a dynamic and distributed training process for model self-improving while keeping the farmers' privacy and computational efficiency.