The relationships between extreme precipitation and rice and maize yields using machine learning in Sichuan Province, China

• Main Authors: Jun Fan, Attachai Jintrawet, Chanchai Sangchyoswat
• Format: Journal
• Published: 2020
• Subjects: Agricultural and Biological Sciences; Biochemistry, Genetics and Molecular Biology; Environmental Science
• Online Access: https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85087555116&origin=inward; http://cmuir.cmu.ac.th/jspui/handle/6653943832/69968
• Institution: Chiang Mai University

© 2020, King Mongkut's Institute of Technology Ladkrabang. All rights reserved. Rice and maize are two staple crops that play a critical role in food security in the southwest of China. They are sensitive to extreme climate events such as drought and flood. Therefore, the assessment of the future production of these crops in an era of climate change is essential. However, current assessment tools are time consuming and require extensive input datasets and expert knowledge. This study is an attempt to provide an alternative tool for prefecture level users using an aggregate Z-index which uses precipitation as an input. A machine learning algorithm (Random Forest) was introduced to build and train the model. Finally, future precipitation projected from Global Climate Models (GCMs) was used as an input to assess the future rice and maize yield variations. This tool outperformed other conventional statistical tools and was especially suitable for assessing extreme cases. Crop yields are significantly affected by drought in the study area. Z-indexes derived from three GCMs on decadal time scales were used for assessing future yield variations. Under the lower emission Representative Concentration Pathway (RCP) 4.5, average maize yields and rice yields are likely to be reduced by-0.58% and-1.49%, respectively, over the next three decades in Mianyang prefecture compared to the baseline period. Similarly, under the higher emission (RCP) 8.5, maize yields and rice yields may decrease by-0.75% and-1.30%, respectively. This assessment tool can be applied in other locations, providing that datasets are available to meet the user’s needs.