Intelligent Fertilizer Optimization in Precision Agriculture using Deep Reinforcement Learning for Sustainable Crop Production

Precision farming is paramount in the context of responding to world food security amidst the climatic conditions, optimization of farm input like the fertilizers to achieve the maximum harvest and minimization of damages to our environment. The standard practices of controlling nitrogen that follows rigid cycles or easy rules, are likely to lead to draining of nutrients, soil erosion and poor harvests which enhances pollution and loss of farm owners. In the framework proposed in this paper, a combination of DRA and the Gym-DSSAT crop simulation interface is suggested, and nitrogen scheduling is presented as a series of decisions in the MDP framework. Three variants of DRL are PPO-based and SAC variants along with DQN-based variants, which are trained and tested against conventional fixed-schedule baselines in various experimental settings of growing maize. SAC performed better as it recorded a 13% positive growth in yield, a 32.4% reduction in nitrogen leaching as compared to a normal practice. This means that SAC can help farmers to achieve more production of maize at the cost of the environment that is more efficient than the conventional vehicles. Such a solution of DR will introduce self-sustaining farming systems that will make farm places more resilient, profitable, and eco-friendly.