Statistical issues. Comment on “proximal sensor data fusion for tropical soil property prediction: Soil fertility properties”

Abstract

dos Santos Teixeira et al. evaluated the effectiveness of three different sensors, portable X-ray fluorescence (pXRF) spectrometry, visible near-infrared (Vis-NIR), and NixProTM color sensor, in predicting six soil properties including sum of bases (SB), effective cation exchange capacity (eCEC), potential cation exchange capacity (CEC), base saturation (BS), soil organic matter (SOM) content, and soil pH, and compared their performance when used alone or in combination. Preprocessing methods and other variables were also analyzed. In this comment paper, some misuses of Kappa statistic in the original paper by dos Santos Teixeira et al. are discussed. The authors calculated Cohen's kappa coefficient to make categorical predictions for soil fertility classes. However, using Cohen's Kappa to estimate the intra-rater reliabilities (IRRs) is not appropriate on the condition of ordinal categorical variables in four or five categories in the original paper. We briefly introduce the Kappa statistic for evaluation of IRR between two raters and propose to use weighted Kappa to compute IRRs under such conditions. To conclude, we recognize that this does not significantly alter the conclusions of this paper, but it is necessary to ensure that the appropriate statistical tools are used.