[A-HW18] Hydrology & Water Environment

[AHW18-04] Optimizing Rice Crop Yield Predictions in the Vietnamese Mekong Delta: An Integrated Crop Modeling and Machine Learning Approach

*Le Dang^1,2, Hiroshi ISHIDAIRA¹, Phung Ky Nguyen³, Kazuyoshi SOUMA¹, Jun MAGOME¹, Bay Thi Nguyen⁴ (1.Interdisciplinary Centre for River Basin Environment, University of Yamanashi,, 2.Faculty of Environment, University of Science, Vietnam National University Ho Chi Minh City, Ho Chi Minh City, Vietnam, 3.Thu Duc People's Committee, Ho Chi Minh City, Vietnam, 4.Faculty of Civil Engineering, University of Technology, Vietnam National University Ho Chi Minh City, Ho Chi Minh City, Vietnam)

Keywords:Machine learning algorithms, DSSAT, Rice yield, Vietnamese Mekong Delta

Rice production holds a crucial role in Vietnam's food security, rural employment, and foreign exchange, with the majority originating from the Vietnamese Mekong Delta (VMD). This study aims to assess the potential of integrating the Decision Support System for Agrotechnology Transfer (DSSAT) model and machine learning algorithms including Random Forest (RF), Support Vector Regression (SVR), Bayesian Ridge (BRR), Gradient Boosting (GB) and Extreme Gradient Boosting (XGBoost) to predict and proactively manage rice productivity factors in Long Phu district, Soc Trang province, VMD. The DSSAT model, applied for simulating rice productivity from 2015 to 2020, utilizes daily meteorological data, soil characteristics, crop varieties, irrigation regime and cultivation management information as input data. The machine learning models incorporates rice yield from DSSAT output along with daily minimum and maximum temperatures, daily precipitation, daily solar radiation, and dailywind speed. With a high coefficient of determination (R2) of 0.92 and normalized root mean square error (nRMSE) of 1.39, the DSSAT model serves as a suitable input for machine learning models.Random Forest, Support Vector Regression, XGBoost, Ridge Regression, and Gradient Boosting algorithms all demonstrating R² values exceeding 0.9. This affirms their efficacy in simulating rice yield, essential for early issue identification and yield optimization.