Precision Agriculture General Agricultural and Biological Sciences

AutoPlot grid: development and quantitative evaluation of an RTK-GNSS-based tool for automated breeding-plot delineation and UAV phenotyping data extraction

Dua, Aashvi, Sharda, Ajay
Springer Nature · 2026-10-15 · DOI 10.1007/s11119-026-10457-z

Abstract

Automated tool created plot boundaries from RTK-GNSS points. Converted RTK GNSS data from planter into polygons representing each planted row. A generalized plot-offset method supported flexible rows-per-plot configurations. Quantitative validation was performed using centroid displacement, orientation, IoU, and area agreement. Processing time was reduced by 94.2–95.5% compared with manual polygon drawing. The workflow integrated plot delineation, zonal statistics, and downstream phenotyping-data extraction into user friendly GUI. Purpose: Plant breeders use unmanned aerial vehicles (UAVs) to collect multispectral (MS) imagery for field-based phenotyping; however, plotlevel analysis remains limited by the time and user intervention required to delineate plot boundaries and extract phenotyping data across large breedingtrials. This study aimed to develop and quantitatively evaluate an automated workflow that integrates planter-recorded real-time kinematic global navigation satellite system (RTK-GNSS) data with UAV MS imagery for breeding-plot delineation and phenotyping-data extraction. Methods: Plot boundaries were generated from planter RTK-GNSS points grouped using HDBSCAN, with a generalized plot-offset formulation supporting flexible rowsper-plot configurations and unique plot identification. The workflow integrated batch vegetation index (VI) calculation and zonal-statistics extraction within a graphical user interface. Automated boundaries were evaluated against expert manually delineated polygons for 1,360 two-row soybean plotsusing centroid displacement, orientation, dimensions, intersection-over-union (IoU), and VI agreement. Processing time was compared with manual andgrid-based ArcGIS Pro workflows by two users. Results: Mean centroid displacement was 17.62 cm (RMSE 18.89 cm), orientation differences were negligible, and mean IoU was 0.778. Approximately 99.63% of plots were within 38 cm, one-half of the 0.76-m row spacing. NDVI, NDRE, and GNDVIshowed Pearson correlations greater than 0.90 with manual extraction. Processing time was reduced by 94.2–95.5% relative to manual polygon drawing and by 83.5–87.0% relative to grid-based ArcGIS Pro processing. Conclusion: Planter RTK-GNSS data can provide a consistent and scalable spatial reference for automated breeding-plot delineation and canopy-focused UAV phenotyping, while preserving plot-level VI patterns and substantially reducing manual processing effort.

Read at publisher →