Mobile Proximal Sensing with Visible and Near Infrared Spectroscopy for Digital Soil Mapping

Abstract

The objective of this study was to estimate multiple soil property local regression models, confirm the accuracy of the predicted values using visible near-infrared subsurface diffuse reflectance spectra collected by a mobile proximal soil sensor, and show that digital soil maps predicted by multiple soil property local regression models are able to visualize empirical knowledge of the grower. The parent materials in the experimental fields were light clay, clay loam, and sandy clay loam. The study was conducted in Saitama Prefecture, Japan. To develop local regression models for the 30 chemical and 4 physical properties, a total of 231 samples were collected; to evaluate accuracy of prediction, 65 samples were collected. The local regression models were developed using 2nd derivative pretreatment by the Savitzky–Golay algorithm and partial least squares regression. The local regression models were evaluated using the coefficient of determination (R2), residual prediction deviation (RPD), range error ratio (RER), and the ratio of prediction error to interquartile range (RPIQ). The R2 accuracy of the 34 local regression models was 0.81 or higher. In the predicted values for 65 unknown samples, the local regression models could ‘distinguish between high and low’ for 3 of the 34 soil properties, but were ‘not useful’ as absolute quantitative values for the other 31 soil properties. However, it was confirmed that the predicted values followed the transition in measured values, and thus that the developed 34 regression models could be used for generating digital soil maps based on relative quantitative values. The grower changed the ridge direction in the field from east–west to north–south just looking at the digital soil maps.

Keywords

• mobile proximal sensor
• local regression model
• spectroscopy
• partial least squares regression
• chemometrics