Potential of Soil and Crop Spatial Properties for Depicting Within-Field Nutrient Variability and Delineating Management Zones for Precision Soil Sampling

Abstract

Limited information currently exists on the potential of various soil and crop spatial properties for delineating management zones (MZs) for precision soil sampling in the southeastern United States; therefore, two studies were conducted to evaluate the effectiveness of different spatial data layers for characterizing within-field nutrient variability and generating MZs for zone-based soil sampling as a cost-effective alternative to high-density grid sampling. The first study assessed relationships between spatial data layers including shallow and deep soil electrical conductivity (EC), elevation, normalized difference vegetation index (NDVI), normalized yield, and soil survey data, and soil pH, phosphorus (P), and potassium (K) variability. This study was performed across nine fields in Alabama and Georgia ranging in size from 9.6 to 37.8 ha. High-density grid soil samples (0.1 ha) were collected across all fields to establish baseline nutrient variability. Pearson correlation analysis and Random Forest (RF) modeling were performed to assess relationships between spatial layers and soil nutrient levels on a field-by-field basis. Soil P and K exhibited considerable spatial variability across all nine fields, with CV values reaching up to 46% and 39%, respectively, while soil pH showed comparatively little variability, with CV values not exceeding 7%. No single spatial layer showed consistent relationships across all fields; however, deep EC demonstrated significant correlations with all three nutrients in seven or more fields, and elevation showed significant associations across five to seven fields depending on the nutrient, while soil series, NDVI, and yield showed more limited and field-specific performance, with soil survey data demonstrating the weakest relationships between all three nutrients. RF modeling resulted in R² values between 0.04 and 0.78, with soil K models showing the greatest consistency and soil pH models performing the weakest, reflecting the low inherent spatial variability of pH across the study fields. The second study evaluated four spatial data layer combinations for MZ delineation across four fields (11.8–30.1 ha) in the Coastal Plain region of Georgia, using k-means clustering at k = 3, 4, and 5. Delineation strategies included EC Only (S1), EC + Elevation (S2), EC + Elevation + NDVI (S3), and EC + Elevation + Yield (S4), with zone performance assessed via variance reduction (VR) of soil pH, P, and K relative to the 0.1-ha grid soil sampling baseline. Results indicated that S1 performed worst across all fields and zone counts, while S2 showed competitive performance with S3 and S4 across most fields. S3 and S4 produced notably higher VR in one field containing an area of recently cleared land, where vegetative indices responded to strong spatial differences, with S4 reaching a mean VR of 32.54% at 4 MZs in that field. Four management zones consistently provided the best balance between VR improvement and practical implementation, with five zones offering only marginal gains over four. Soil K showed the greatest zone differentiation across all fields, while soil pH was the most difficult soil property to differentiate due to its inherently low spatial variability. Overall, S2 was identified as the most broadly applicable and practically efficient delineation strategy, offering improved performance over EC alone while minimizing data requirements.