Predicting rice productivity for ground data-sparse regions: A transferable framework and its application to North...

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Assessing crop productivity in data-sparse regions is challenging.
• Observation-dependent methods often fail due to limited data availability.

Purpose of the Study:

• To develop a transferable framework for assessing rice productivity in North Korea.
• To utilize climate similarity, machine learning, and multi-source data for accurate yield prediction.

Main Methods:

• Extracted dynamic rice distributions using phenological stages and Moderate Resolution Imaging Spectroradiometer (MODIS) products.
• Compared four machine learning models (Linear Regression, Neural Network, Support Vector Machine, Random Forest) for simulating rice productivity.
• Integrated an optimal model with agricultural management practices for transferability and prediction.

Main Results:

• Random Forest demonstrated superior performance, achieving R-squares of 0.87 (whole growth period) and 0.83 (seeding-heading period).
• Solar-induced chlorophyll fluorescence, maximum temperature, and evapotranspiration were key yield determinants (approx. 40%).
• Planting area significantly influenced production (over 42%), and the framework explained substantial yield and production variations compared to FAO data.

Conclusions:

• Phenological identification via remote sensing accurately captures rice growth and distribution.
• The proposed transferable framework enhances crop productivity assessment and early warning systems for food security.
• The framework shows potential for scalability to other crops and data-sparse regions worldwide.