Analyzing the Use of Publicly Available Multispectral Imagery to Guide the Creation of Soil Sampling Schemes

Abstract

As site-specific nutrient management has grown in popularity, the need for accurate soil fertility data has increased. Unfortunately, the cost of detailed soil sampling has prohibited many farmers and consultants from collecting samples at the proper resolution. It is necessary to develop techniques using easy-to-access ancillary data to guide the creation of soil sampling strategies. Ancillary data acquired from three publicly available sources (Landsat 7, Landsat 8, Sentinel 2a) and Soil Electrical Conductivity (EC) was used to determine the strength of relationship to commonly amended soil fertility variables (phosphorus (P), potassium (K), and soil pH). Ancillary data relevancy was determined by a comparison of spatial bi-correlation and fitted semi-variogram ranges. Additionally, principal component analysis (PCA) was performed to allow for unsupervised clustering of ancillary data. These clusters were used to predict zones of nutrient sufficiency. This method resulted in an average class accuracy of 67% (K), 78% (P), and 46% (pH), indicating that classification by clustering may be used to delineate soil fertility distributions to guide the creation of soil sampling schemes.