Development of multistage crop yield estimation model using machine learning and deep learning techniques.

Area of Science:

• Agricultural Science
• Data Science
• Meteorology

Background:

• Accurate wheat yield estimation is crucial for food security and agricultural planning.
• Meteorological factors significantly influence crop productivity, necessitating data-driven approaches.

Purpose of the Study:

• To apply and compare machine learning techniques for multivariate meteorological time series data analysis.
• To estimate wheat yield across five districts in Punjab at different crop growth stages.

Main Methods:

• Utilized 34 years of wheat yield and weather data.
• Developed and validated models using stepwise multi-linear regression (SMLR), artificial neural network (ANN), support vector regression (SVR), random forest (RF), and deep neural network (DNN).
• Incorporated meteorological variables from specific weeks corresponding to tillering, flowering, and grain-filling stages.

Main Results:

• Random Forest (RF), Support Vector Regression (SVR), and Deep Neural Network (DNN) models demonstrated consistent and promising performance.
• Models achieved overall Mean Absolute Percentage Error (MAPE) and normalized Root Mean Square Error (nRMSE) below 6% during validation.
• RF, SVR, and DNN models showed outstanding validation performance for Faridkot, Ferozpur, and Gurdaspur districts.

Conclusions:

• RF, SVR, and DNN models are highly effective for district-level wheat yield estimation.
• The Random Forest model exhibited superior accuracy compared to SVR and DNN.
• These machine learning approaches offer reliable tools for agricultural forecasting at various crop growth stages.