Intra-Field Crop Yield Predictability

TXST STEM Research Conference 2026

Using Machine Learning and Deep Learning at pixel-level precision

Download Poster PDF Explore Research

Research Team

Swipe to explore all co-authors

Affiliated Organizations

Introduction

Background, motivation, and objectives of the study

Problem Statement

Research Objectives

Machine LearningDeep LearningCrop YieldPrecision AgricultureRemote Sensing

Study Area

Geographic overview of the research site

Research Site

This study focuses on the USDA Beltsville Agricultural Research Center (BARC) in Beltsville, Maryland- a 6,600-acre USDA agricultural research facility located between the Atlantic Coastal Plain and the Piedmont Plateau. The site features diverse land cover, including cultivated fields, forests, wetlands, and infrastructure, providing a suitable agricultural landscape for predictive analysis.

Study Site

USDA Beltsville ARC

Location

Beltsville, Maryland

Fields

118

Years Covered

11 years

Crops Studied

🌽 Corn🫘 Soybean🌾 Wheat

Loading map…

Dataset Overview

Multi-source remote sensing and environmental variables

Methodology

Research pipeline, seasonal design, and yield favorability

Models Compared

Classical machine learning vs. deep learning architectures

Random Forest

Classical ML

XGB

XGBoost

Classical ML

CNN LSTM

Attention Based CNN-LSTM

Deep Learning

Methodology Flowchart

Seasonal Date Windows

Year-Wise Favorability Formula

F_{c,y} = \frac{\bar{Y}_{c,y} - \mu_c}{\sigma_c}

where $\mu_c$ and $\sigma_c$ represent the mean and standard deviation of crop $c$ , and $\bar{Y}_{c,y}$ denotes the average yield for crop $c$ in year $y$ across all fields.

F_{c,y}

Favorability score for crop c, year y

\bar{Y}_{c,y}

Mean yield of crop c in year y

\mu_c

Long-term mean yield for crop c

\sigma_c

Standard deviation for crop c

Results

Interactive analyses across three evaluation frameworks

Representative Model Performance Comparisons

Section 1 of 3

Yield Map

Central Farm-4-23A · Corn · multi-year

Key Takeaways

Select Crop:

Deep LearningCNN-LSTM — Corn

Scatterplot

Predicted vs Observed Yield

Mean Temporal Attention

Attention Weights Over Time

Spatial Attention Heatmap

Attention Across the Field

Classical MLRF & XGB — Corn

Random Forest (RF)

Predicted vs Observed Yield

XGBoost (XGB)

Predicted vs Observed Yield

Key Takeaways

Conclusion

Key findings from the study

Future Work

Planned directions and next steps

Ongoing research — more directions to be added

References

Academic sources and citations