Quantifying the uncertainty of soil colour measurements with Munsell charts using a modified attribute agreement analysis

The use of Munsell colour charts is the classical way of determining colour information in soil science. The procedure is well-known and consists of visually comparing soil samples with colour chips contained in the charts. This visual approach has several drawbacks and although the chart-based procedure is routinely used, it is not easy to find systematic studies on the accuracy of this methodology. In this paper, we seek to gain insight into the strengths and weaknesses of using soil colour charts as a colour measurement device. The tool used to conduct our study is a modification to the attribute agreement analysis (AAA) method which consists of finding matches between colour standards and colour designations obtained by several appraisers. In the experiment, standards were obtained using a trichromatic colorimeter coupled to a computer program that implements the k nearest neighbour (k-NN) classification algorithm. In order to do the experiment, 276 soil samples were observed twice by four trained appraisers (2208 data records). The naïve count of matches across all the records in the dataset gave <5% of agreement for all three colour components Hue, Value and Chroma. The modified AAA criterion implemented in the study gave a clear increase in all indicators with values ranging from 82.2% to 100% in the agreement within appraisers, 39.5% in the agreement between appraisers, and 42.8% in the agreement of appraisers vs. standards. Results also show that users of the Munsell charts tend to mostly report correct Hues but higher Values and Chromas than true soil colours.