Files
Abstract
Accurate assessment of spatial variability of soil texture is a significant component of agriculture and environmental modeling. Current soil maps lack detail necessary for intensive management like precision agriculture. Determining optimal sample sizes for creating detailed soil maps is challenging because it is cost and labor prohibitive. In this work, random forest models of soil texture were developed using an 80/20 split for training and testing data, respectively, for 50 iterations of sample sizes between 10-65. Sixty-nine samples were taken from a 40-acre crop field in July 2020 and May 2021 at 0-10, 10-40, 40-70, and 70-100 cm and combined with topographic covariates, electromagnetic conductivity (EM31), and spectral reflectance data as predictors. R2 and root mean square error (RMSE) varied by soil property and depth. A sample size of 35-45 samples represented the variability of soil texture most depth increments based on the trends in R2 and RMSE.