Skip to main content
ARS Home » Plains Area » Temple, Texas » Grassland Soil and Water Research Laboratory » Research » Publications at this Location » Publication #421967

Research Project: Development of Enhanced Tools and Management Strategies to Support Sustainable Agricultural Systems and Water Quality

Location: Grassland Soil and Water Research Laboratory

Title: Leveraging spectral neighborhood information for corn yield prediction with spatial-lagged machine learning modeling: Can neighborhood information outperform vegetation indices?

Author
item NOA-YARASCA, EFRAIN - Texas Agrilife Research
item OSORIO-LEYTON, JAVIER - Texas Agrilife Research
item Adhikari, Kabindra
item Hajda, Chad
item Smith, Douglas

Submitted to: Artificial Intelligence International Joint Conference
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/20/2025
Publication Date: N/A
Citation: N/A

Interpretive Summary: Crop yield prediction is an important aspect of precision agriculture. This study evaluates an innovative approach of using ground truthed crop yield data to predict yield at specific points through evaluating the nearest neighbors to the point of interest. This study was condicted at a corn field located near Temple, TX. During the growing season, five types of images were collected from an unmanned aerial vehicle. These were assessed using geospatial modeling techniques coupled with machine learning. Inclusion of ground-truthed data of nearest neighbors outperformed strict remote sensing for yield prediction. The optimal number of nearest neighbors was between 4 and 8, with yield predictions diminished outside of this range. This study provides a novel methodology to other precision agriculture researchers to more accurately predict yield when coupling remote sensing with ground truthed data.

Technical Abstract: This study introduces an innovative approach to crop yield prediction by incorporating spatially lagged spectral data through the Spatial-Lagged Machine Learning model, an enhanced version of the Spatial Lag X Model. The research aims to show that spatially lagged spectral data improves prediction compared to traditional vegetation index based methods. Conducted on a 19-hectare in Temple, TX during the 2023 growing season, the study used Five-band multispectral image data and 8,581 yield measurements ranging from 1.69 to 15.86 Mg/Ha. Four predictor sets were evaluated using various combinations of spectral bands, neighborhood data visible spectra. These were evaluated using the Spatial Lag X model and four decision-tree-based Spatial-Lagged Machine Learning models, with performance assessed using correlation coefficient and root-mean squared error. Results showed that incorporating spatial neighborhood data outperformed vegetation index -based approaches, emphasizing the importance of spatial context. Spatial-Lagged Machine Learning models performed best with 4–8 neighbors, while excessive neighbors slightly reduced accuracy. Visible spectral bands improved predictions, but a smaller subset (10–15 indices) was sufficient for optimal yield prediction. Key predictors included spatially lagged spectral bands and visible spectral bands, highlighting the value of integrating neighborhood data for improved corn yield prediction.