Predicting the color of sandy soils from Wisconsin, USA

Abstract

Color is an important soil property and often used to infer soil properties and delineate soil horizons. We investigated the effect of soil texture, carbon, total elements, and particle-size fractions on the color of sandy soils using five color models. In total, 915 soil samples were collected to a maximum depth of 220 cm at 400 locations in the Wisconsin Central Sand Plains, and the soils were sandy throughout. The samples were scanned using a visible-near infrared spectrometer, with soil color models (HSV, RGB, CIE L*a*b*, CIE L*u*v*, and redness index (RI)) extracted from the reflectance spectra. Cubist models were used to predict each of the soil color coordinates from the soil properties. The models showed high prediction accuracy for V, R, SRGB, L*, a*, b*, u*, and v* color coordinates for both calibration and validation. The CIE L*a*b* color model was generally better than other color models. Silt and Fe were used in all of the Cubist models, while sand, clay, carbon, Al, Si, Mn, and Zn were used in most models. The 45–125 µm, 125–250 µm, and 1000–2000 µm fractions affected the soil color as opposed to the 250–500 µm and 500–1000 µm fractions. A clustering analysis on the CIE L*a*b* color model showed that soil lightness was higher with higher sand content, but lower with an increase in silt and clay content, carbon, Fe, Al, Zn, and Mn. The LUCAS dataset has approximately 20,000 soil samples and was used to explore the relationships between soil color and soil properties in sandy soils and to test the robustness of the model established with the Central Sands dataset. The Cubist models developed from either dataset (Central Sands, LUCAS) were not useful to predict CIE L*a*b* color using the other dataset.