Related Experiment Videos
A hybrid deep learning approach for winter wheat yield prediction: evidence from leveraging multi-source data.
Manogna R L1, Shaanil Punglia2, Siddhant Tushar Joshi2
1Department of Economics and Finance, BITS Pilani K K Birla Goa Campus, Goa, India. manognar@goa.bits-pilani.ac.in.
Scientific Reports
|June 22, 2026
Summary
Accurate wheat yield forecasting in India is crucial for food security. A new hybrid deep learning model significantly improves predictions, especially during anomalous weather, outperforming traditional methods.
Area of Science:
- Agricultural Science
- Data Science
- Machine Learning
Background:
- Accurate district-level wheat yield forecasts are vital for India's food security, supply chain management, and agricultural policy.
- India is the world's second-largest wheat producer, with seven states accounting for approximately 95% of national production.
Purpose of the Study:
- To benchmark nine different model classes for district-level wheat yield forecasting in India.
- To evaluate the performance of a novel hybrid deep learning model against established methods and a persistence forecast.
Main Methods:
- A 23-year dataset (2001-2023) of 275 districts across India's major wheat-producing states was used.
- Nine model classes were benchmarked, including Random Forest, XGBoost, LightGBM, CNN, LSTM, BiLSTM, Transformer, Parallel CNN-LSTM-Attention, and a hybrid CNN-BiLSTM-Attention model.
- The hybrid model employed modality-specific routing: a 1D-CNN for soil profiles and a BiLSTM with self-attention for time-series meteorological and remote-sensing data.
Main Results:
- The proposed hybrid CNN-BiLSTM-Attention model achieved the best performance with a Mean Absolute Error (MAE) of 273.2 kg/ha and an R-squared of 0.795.
- This represents a 43.8% MAE reduction over Random Forest and a significant improvement over other deep learning models.
- The model's value-add was most pronounced in anomalous years, improving MAE by 7.2% and RMSE by 11.2% over persistence during the dry 2023 sowing season.
Conclusions:
- Persistence-aware, modality-specific deep learning offers a practical framework for stress-sensitive wheat yield forecasting, particularly in data-scarce agricultural regions.
- The hybrid model's ability to capture anomalies highlights its potential for enhancing agricultural planning and policy.
- Attribution analysis identified the grain-filling window, influenced by EVI and NDVI signals, as critical for stress-sensitive forecasting.
Related Concept Videos
01:25Multiple Regression
Multiple regression assesses a linear relationship between one response or dependent variable and two or more independent variables. It has many practical applications.
Farmers can use multiple regression to determine the crop yield based on more than one factor, such as water availability, fertilizer, soil properties, etc. Here, the crop yield is the response or dependent variable as it depends on the other independent variables. The analysis requires the construction of a scatter plot...
Farmers can use multiple regression to determine the crop yield based on more than one factor, such as water availability, fertilizer, soil properties, etc. Here, the crop yield is the response or dependent variable as it depends on the other independent variables. The analysis requires the construction of a scatter plot...
02:16Light Acquisition
In order to produce glucose, plants need to capture sufficient light energy. Many modern plants have evolved leaves specialized for light acquisition. Leaves can be only millimeters in width or tens of meters wide, depending on the environment. Due to competition for sunlight, evolution has driven the evolution of increasingly larger leaves and taller plants, to avoid shading by their neighbors with contaminant elaboration of root architecture and mechanisms to transport water and nutrients.