Viet Nam2025Machine learning for rice yield forecasting using climate and soil data in the Mekong Delta
[Tên tác giả]
[Đơn vị công tác]
- Published in
- [Tên tạp chí hoặc hội nghị]
- Year
- 2025
- DOI
- 10.48550/clmv.2026.002
- Original language
- Tiếng Việt
- Topic
- Smart agriculture, Digital economy and transformation
- Published by
- [Đơn vị công tác]
Abstract
This study develops a machine learning model to forecast rice yield based on climate, soil and cropping history data from several provinces in the Mekong Delta. The model achieves higher forecast accuracy than traditional statistical methods. Findings suggest potential for wider application in other rice-growing areas across CLMV.
Background
Early and accurate rice yield forecasting is important for production planning and food security, particularly as saltwater intrusion and climate change increasingly affect the Mekong Delta.
Traditional statistical forecasting methods often struggle to capture the non-linear relationships among multiple input factors such as rainfall, temperature, soil salinity and cropping history.
Methods
The study uses climate, soil and rice yield data collected from several Mekong Delta provinces between 2015 and 2024 to train and validate a range of machine learning models.
Models were compared on forecast accuracy, including a linear regression baseline and non-linear machine learning models that jointly process multiple input variables.
Key findings
The non-linear machine learning model produced significantly more accurate yield forecasts than the linear regression baseline, particularly in areas affected by seasonal saltwater intrusion.
Soil salinity and early-season rainfall were identified as the factors with the greatest influence on forecast accuracy in the model.
Policy recommendations
The study recommends that local agricultural agencies pilot integrating the forecasting model into decision-support systems for farmers, helping adjust planting schedules to yearly conditions.
It also recommends expanding standardised soil and climate data collection in other rice-growing areas across CLMV to assess the model's scalability.