Alpysbay et al. (2026) Beyond Vegetation Indices: Winter Solar Radiation and Soil Properties Drive Wheat Yield Prediction in the Arid Steppes of Kazakhstan Using Gradient Boosting

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Agriculture
• Year: 2026
• Date: 2026-04-01
• Authors: Marua Alpysbay, Serik Nurakynov, Azamat Kaldybayev
• DOI: 10.3390/agriculture16070782

Research Groups

Not explicitly mentioned in the paper text.

Short Summary

This study developed a robust XGBoost-based framework for spatio-temporal spring wheat yield forecasting in rainfed agricultural zones, achieving R² values of 0.69 (interpolation) and 0.65 (extrapolation). It revealed that pre-seasonal agroclimatic drivers, particularly winter insolation and April soil moisture recharge, are more influential on yield than mid-season vegetation indices in arid rainfed systems.

Objective

• To develop a comprehensive analytical framework for the spatio-temporal forecasting of spring wheat yield in risk-prone rainfed agricultural zones.
• To identify and quantify the most influential biophysical drivers of spring wheat yield, including pre-seasonal and in-season factors.

Study Configuration

• Spatial Scale: Regional to sub-regional scale, covering risk-prone rainfed agricultural zones, with specific analysis of foothill and irrigated clusters.
• Temporal Scale: 25-year time series for model training and analysis, focusing on annual spring wheat yield and its drivers across different seasons (winter, spring, grain-filling stage).

Methodology and Data

• Models used: XGBoost algorithm for yield modeling, TreeSHAP framework for model interpretability.
• Data sources:
  • Remote sensing data (including NDWI and NDVI).
  • Meteorological reanalysis products.
  • Soil parameters.
  • 25-year time series integration.

Main Results

• The proposed framework achieved an interpolation accuracy of R² = 0.69 with a Root Mean Square Error (RMSE) of 330 kg/ha.
• Extrapolation to unseen regions yielded an accuracy of R² = 0.65 with an RMSE of 350 kg/ha, demonstrating robustness and transferability.
• TreeSHAP analysis identified April soil moisture recharge and winter insolation (as a proxy for snow cover persistence and surface albedo dynamics) as the dominant agroclimatic drivers.
• NDWI was empirically confirmed to be superior to NDVI for detecting latent water stress during the grain-filling stage.
• Pre-seasonal agroclimatic drivers (winter solar radiation and April moisture recharge) were found to exert a stronger influence on yield than mid-season NDVI in arid rainfed systems.
• Geospatial analysis revealed a pronounced domain shift in foothill and irrigated clusters, attributed to the coarse spatial resolution of climate grids and irrigation-induced decoupling of crop phenology from precipitation regimes.

Contributions

• Development of a comprehensive and robust analytical framework for spatio-temporal spring wheat yield forecasting using the XGBoost algorithm and a two-level validation strategy.
• Demonstration of the framework's high accuracy and transferability across different regions.
• Identification of critical pre-seasonal agroclimatic drivers (winter solar radiation and April moisture recharge) as more influential than mid-season vegetation indices in arid rainfed systems, challenging prior frameworks.
• Empirical confirmation of NDWI's superiority over NDVI for detecting latent water stress during the grain-filling stage.
• Application of the TreeSHAP framework to provide interpretability into the complex biophysical relationships modeled.

Funding

Not explicitly mentioned in the paper text.

Citation

@article{Alpysbay2026Beyond,
  author = {Alpysbay, Marua and Nurakynov, Serik and Kaldybayev, Azamat},
  title = {Beyond Vegetation Indices: Winter Solar Radiation and Soil Properties Drive Wheat Yield Prediction in the Arid Steppes of Kazakhstan Using Gradient Boosting},
  journal = {Agriculture},
  year = {2026},
  doi = {10.3390/agriculture16070782},
  url = {https://doi.org/10.3390/agriculture16070782}
}