Irrigation Optimization in Agricultural Fields Using Deep Reinforcement Learning Approaches

In recent years, due to the world population growth, the demand for agricultural products has also increased. Global crises such as climate change, destruction of water resources, pollution and destruction of aquatic ecosystems, have limited the access to clean water to irrigate the agricultural fields in a global scale. Meanwhile, according to the reports of Food and Agriculture Organization of the United Nations (FAO), about 70% of global water consumption is allocated to agricultural activities. Therefore, it is necessary to minimize water consumption during agricultural irrigation while maximizing crop yield by the end of the season. Most IoT frameworks used in irrigation systems collect data from the soil and weather, then send it to the user, leaving the final decision on irrigation to the user. This approach may cause water wastage or even water stress for the plants as the amount of water irrigated in fields may be low. In this paper, we will use deep reinforcement learning approaches to reach this goal. We propose a novel approach to minimize water usage while keeping the yield produced at the highest level. Our approach utilizes DQN and A2C networks. The findings demonstrate that our methods reduce water wastage compared to previous approaches, making it more flexible and practical in varying weather and soil conditions.