Truace
Subscribe

Wheat yield prediction using machine learning, proximal sensing, nitrogen and climate data for Oklahoma

PURPOSE: Accurate in-season yield prediction is a practical prerequisite for informed nitrogen (N) management decisions in winter wheat production, yet no rigorous framework has been established and tested against the tools producers currently rely on. Our study developed and evaluated machine learning (ML) models for pre-application wheat yield prediction to support in-season top-dress N management in Oklahoma. METHODS: Four ML algorithms were evaluated using 57 site-years of multi-source data from rainfed Okla…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1399· cite & verify
Wheat yield prediction using machine learning, proximal sensing, nitrogen and climate data for Oklahoma
Dry Spring in Eastern Australia (5073) by NASA image of NDVI anomaly by Jesse Allen using MODIS data provided by Inbal Reshef as part of the Global Agricultural Monitoring Project involving NASA, USDA’s Foreign Agricultural Service (FAS), and the University of Maryland. Assaf Anyamba from the Goddard Earth Sciences Technology Center, and James Crutchfield with FAS, contributed to the caption as part of the same partnership. More data and information about this joint project is available at Satellite Information for Agricultural Monitoring.. Public domain

In brief

Researchers tested four machine learning algorithms on 57 site-years of data from rainfed Oklahoma wheat environments. Inputs included proximal-sensing NDVI at several growth stages, pre-emergence nitrogen rate, soil texture and monthly climate variables. The models were compared with the INSEY tool and with classical statistical methods.

Better yield forecasts before the top-dress window could help producers decide how much nitrogen to apply, supporting productivity and nitrogen use efficiency. Because the climate inputs come from the free Oklahoma Mesonet, the approach could fit existing recommendation systems. The abstract reports model accuracy, not field outcomes from farmers actually using it to adjust nitrogen.

Main points

  1. The best model, CatBoost, reached R² of 0.81, RMSE of 0.62 t per hectare and MAE of 0.42 t per hectare.
  2. The INSEY benchmark managed an R² of only 0.12 under the same pre-application data constraint.
  3. The final 11-feature model relied most on February–March temperatures, evapotranspiration, water balance, pre-plant nitrogen rate and early-season NDVI.
  4. All climate inputs are freely available via the Oklahoma Mesonet, so no extra data infrastructure is required.

The gain

A machine learning model using sensor, soil, nitrogen and free climate data predicted wheat yield before nitrogen application with R² of 0.81, versus 0.12 for the INSEY benchmark.

The rundown

The study was framed around a gap: accurate in-season yield prediction is a prerequisite for nitrogen decisions in winter wheat, but no rigorous framework had been tested against the tools producers currently use.

Data came from rainfed Oklahoma environments over 57 site-years. Besides the INSEY benchmark, the authors compared ordinary least squares, a linear mixed model and a generalized additive model as classical references.

The authors conclude the framework can be integrated into existing nitrogen recommendation systems, combining remote sensing, climate and management data to predict yield within the application window.

Sources

  1. Peer-reviewedPrecision agriculture2026-10-01

The debate