Rice Yield Monitoring Using Satellite Data and AI for Food Security and Sustainable Landscapes in Myanmar

• Main Authors: Phalke, Aparna R., Dutta, Rishiraj, Jayasinghe, Susantha, Quintero, Diego, Limaye, Ashutosh S., Griffin, Robert
• Format: Conference Proceeding
• Language: English
• Subjects: Artificial neural network; Artificial neural networks; Computational modeling; Cropland monitoring; Data models; Food security; Production; Reliability; Rice yield model; Satellite remote sensing; Satellites; Yield estimation
• ISSN: 2153-7003
• DOI: 10.1109/IGARSS53475.2024.10642947

Rice production in Myanmar is severely affected by the uncertain weather changes and political unrest especially due to the conflicts of coup after February 2021. Although recent improvements in satellite remote sensing have helped national to field-scale yield estimation and forecasting, the challenges through quality of reference data questions the quality of forecasts and estimation. In Myanmar, agricultural production data is publicly available at regional scale (equivalent to province level in USA) from 2012 to 2020. The consistent, timely and granular information on yield estimates at lower to higher admin unit level is crucial need to support sustainability of livelihood. In support to these efforts to improve food insecurity analysis and climate resilient livelihood formation, SERVIR - a joint USAID and NASA initiative - has implemented rice yield model using machine learning and satellite data to guide the decision makers in the region. This study used time-series (2012-2023) information on satellite derived vegetation indices, weather variability, topography and water use to predict the rice yield at regional to field scale using artificial neural network, random forest and boosted decision tree based artificial intelligence (AI) modeling. The model was tested spatially as well as temporally and overall average R 2 observed varied from 0.6 to 0.7. The presented rice yield model in this study provides carefully evaluated rice yield estimations at regional to field-scale across Myanmar rice-growing region. The consistent rice estimates from this study will help decision makers in the region. This study is the first to model rice yield at this high-resolution with locally tuned information. The model and algorithm developed in this study have the potential of operational use for further needs.