Advancing food security: Rice yield estimation framework using time-series satellite data & machine learning.

Area of Science:

• Agricultural Science
• Remote Sensing
• Food Security

Background:

• Accurate rice yield data is essential for food security management and agricultural policy in Bangladesh.
• Existing remote sensing methods lack high spatial resolution and integration into national reporting systems.
• There is a need for consistent yearly boro rice yield maps at a sub-district scale.

Purpose of the Study:

• To introduce a workflow for timely and precise rice yield estimation at a sub-district scale (1,000-meter resolution).
• To develop a consistent yearly boro rice yield map for Bangladesh from 2002 to 2021.
• To analyze trends in boro rice yield over two decades.

Main Methods:

• Leveraged MODIS satellite data and annual district-level yield data.
• Trained a random forest model for boro rice yield estimation at 1,000-meter resolution.
• Utilized the modified Mann-Kendall trend test for trend analysis.

Main Results:

• Achieved a mean percentage root mean square error (RMSE) of 8.07% (district yields) and 12.96% (crop-cut data).
• Estimated boro rice yield uncertainty ranged from 0.40 to 0.45 tons per hectare.
• Identified increasing yield trends in 23% of rice areas, decreasing in 0.11%, and no trend in 76.51%.

Conclusions:

• The developed workflow provides scalable, high-resolution boro rice yield estimates crucial for food security in Bangladesh.
• This is the first study to provide two decades of yearly boro rice yield maps at 1,000-meter resolution for Bangladesh.
• The workflow is adaptable for rice yield estimation in other global regions.