Deep reinforcement learning based climate aware irrigation ccheduling for boro rice cultivation in Mymensingh, Bangladesh

Abstract

An optimized irrigation system leads to better crop yield and reduction in water wastage,

which in broader terms contributes to the growth of a country’s economic sector and im- proves food security. In flood prone countries like Bangladesh[10], farmers manually

access irrigation strategies using finger tests and eye observation, leading to several prob- lems such as over-irrigation, under-irrigation, failure to adapt irrigation strategies with

the weather, etc. Thus, there has been significant focus on developing efficient irrigation systems, integrating various technologies such as IoT, Geographic Information Systems (GIS), Artificial Intelligence (AI), Machine Learning (ML), etc. Within the domain of ML, Reinforcement Learning (RL) has the potential to take irrigation systems to the next level through automating irrigation, however, integrating RL in this field remains under-researched. Moreover, in Bangladesh, neither traditional methods of irrigation nor

smart irrigation systems are capable of adapting to the dynamic climatic and environ- mental conditions of Bangladesh. Thus, this paper proposes an RL-system that utilizes

diverse crop data as well as real time climatic and environmental data to develop a robust irrigation system to optimize irrigation, maximize yield, reduce water wastage and adapt to the dynamic environment of Bangladesh.