Skip to Content

Crop Yield Prediction

 

Sri Lanka’s farmers face increasing uncertainty from shifting weather, rising costs, and climate change. Accurate, timely crop yield prediction is essential for:

  • Ensuring food security.

  • Planning subsidies and resource allocation.

  • Managing loans, insurance, and market supply chains.

  • Supporting exporters with reliable crop forecasts.

The WHY Super App Yield Prediction Solution uses satellite data + machine learning models to provide field- and regional-level yield estimates. Whether it’s a smallholder’s paddy field or a provincial tea estate, the system helps everyone, from farmers to ministries — make better decisions.

Approach and Methodology

Statistical yield prediction model

  • Relies on historical dataset, primarily on past yield data.
  • Trained on historical data and predicts current-season yields by combining past trends with present parameters.

Biophysical yield prediction model

  • Relies on the physical crop parameters and its phenological characteristics (variety, growth stages, water demand, etc.).
  • Can predict yield for a specific date without historical data and can be applied repeatedly throughout the season.

Methodology

90 % Accuracy 

1. Statistical Model.

  • Collects past yield, weather, soil, and satellite data.

  • Trains ML models (Linear Regression, Random Forest, XGBoost).

  • Predicts current-season yield by combining past trends with present conditions.

2. Biophysical Model.

  • Uses phenological data (growth stages, crop characteristics).

  • Simulates biological productivity (total biomass, water use, soil moisture, evapotranspiration).

  • Updates forecasts every 2 weeks with new weather/soil inputs.


90%