Quantitative model of irrigation effect on maize yield by deep neural network

A rapidly expanding world population and extreme climate change have made food production a crucial challenge in the twenty-first century. Improving crop management could be an effective solution for this challenge. However, due to the associated cost and time to perform field works, researchers are widely rely on agricultural systems modeling to examine the impacts of different crop management scenarios. Meanwhile, due to the complexity of agricultural systems modeling, their applications in producing practical knowledge for producers are limited. Meanwhile, deep learning techniques have been recognized as the preferred method compared to other machine learning techniques, especially when dealing with large datasets. In addition, deep learning techniques are easily adopted by non-experts due to the feature of learning ability that can automatically discover the classifications from raw data. Meanwhile, one of the drawbacks of using deep learning techniques is the training time, which can last anywhere from a couple of weeks to even a few months. Therefore, the goal of this study is to examine the applicability of deep learning techniques to compute a numerical model of crop growth. In this study, an agricultural systems model known as the Decision Support System for Agrotechnology Transfer (DSSAT) is used to evaluate the impacts of irrigation amount and time of application on crop yield. A deep learning network is utilized and trained by incorporating the large amounts of DSSAT models inputs (i.e., precipitation date, precipitation amount, irrigation date irrigation amount) and output (i.e., maize yield at the end of the growing season). However, in order to simplify the process, we combined the amount of irrigation and rainfall together and presented them in the form of the amount of water per day. Experimental results have demonstrated the effectiveness of this proposed deep learning technique in crop yield prediction.