A comparative study of four color measurement methods for soil color identification and related properties prediction

Soil color serves as a comprehensive indicator of soil composition and intrinsic properties and can be used to predict these properties. The identification of soil color by proximal sensing varies among and even within color measurement methods. In addition, several color space models can be used to quantify color. This poses the question as to the difference between soil color measurement methods in different color spaces, the property predictions from these, and their optimal combination. A total of 200 soil samples from China, 80 from Guizhou Province and 120 from Hunan Province, were used to address this question. The color of these samples was measured by four different methods: subjective visual estimation, a smartphone, a Nix Pro2 colorimeter and an ASD spectrometer. The color difference in CIEL*a*b* color space, defined as (Delta E-ab), between these methods was quantified using the National Bureau of Standard (NBS) unit. Five color space models, including RGB, HIS, CIEL*a*b*, CIEL*c*h*, and CIEL*u*v*, were used to quantify soil color parameters defined by each model. These parameters were then used to predict the color-associated properties soil organic matter (SOM) and free iron concentrations (FI). Stepwise multiple linear regression (SMLR), back propagation neural network (BPNN) and partial least squares regression (PLSR) models were built to make these predictions. The color difference between the Nix Pro2 and smartphone was the smallest (Delta E-ab = 2.66 NBS, considered "noticeable"), while significant systematic differences were found between ASD and the other methods (Delta E-ab 23.48 to 25.51 NBS, "very much"). Using the visible spectra measured by the ASD as input to the PLSR model as a baseline, the ratio of performance to interquartile range (RPIQ) range for SOM prediction by other sensors and models varied from 2.90 to 3.43, while for FI, it ranged from 3.07 to 3.93. Different methods had quite different prediction performances. The ranks were Nix Pro2 > smartphone > ASD > subjective visual estimation, PLSR > BPNN > SMLR for SOM, and smartphone > Nix Pro2 > ASD > subjective visual estimation, PLSR > SMLR > BPNN for FI. Brightness-related color parameters (R, h* and L*) are recommended for SOM prediction, and yellow- and red- related parameters (S and b*) for FI prediction. The Nix Pro2 and smartphone are recommended as portable and cost-effective instruments for proximal soil sensing.