A multifunctional matching algorithm for sample design in agricultural plots

Area of Science:

• Agricultural Science
• Geospatial Analysis
• Data Science

Background:

• Efficient crop management relies on accurate, representative agricultural field data.
• Limited grower resources necessitate advanced methods for selecting optimal sampling points.
• Existing methods may not fully capture spatial variability or covariate distributions.

Purpose of the Study:

• Develop a data-driven method for selecting representative agricultural sampling locations.
• Ensure selected points reflect the distribution and spatial variability of field covariates.
• Create an algorithm to determine the minimal number of observations for desired accuracy.

Main Methods:

• Developed the multifunctional matching (MFM) criterion based on matching moments (standard deviation, mean, Kendall's tau) between sample and population.
• Applied MFM to vineyard and peach orchard datasets with covariates like NDVI, soil electrical conductivity, slope, and TWI.
• Validated MFM against crop water stress index (CWSI) and compared it with conditioned Latin hypercube sampling (cLHS) and random sampling.

Main Results:

• MFM algorithm determined optimal sampling locations and number of points for 90% representation accuracy.
• MFM demonstrated superior representation of CWSI distribution compared to cLHS and random sampling.
• Selected MFM locations showed smaller deviations from population mean and standard deviation, capturing spatial variability.

Conclusions:

• MFM is an effective data-driven approach for selecting representative agricultural sampling points.
• The method accurately captures spatial variability and covariate distributions, outperforming other sampling strategies.
• MFM is adaptable for various moments/functionals and applicable across disciplines needing small sample sizes.