Comparison of Vis–NIR on in situ, intact core and dried, sieved soil to estimate clay content at field to regional scales

We evaluated the in situ performance of a newly designed visible and near‐infrared (Vis–NIR) penetrometer to predict clay content for 10 fields, 151 profile locations and 485 samples that span four orders of the USDA Soil Taxonomy across the Pacific Northwest region of the USA. The clay mineralogy of the soil families is mixed and heterogeneous. To minimize the effects of soil moisture, in situ Vis–NIR measurements were taken during the summer in a Mediterranean climate when most soil is at or near wilting point. Partial least squares regression (PLSR) and leave‐one‐core‐out cross‐validation were used to assess the different soil Vis–NIR measurements: (i) standard error of prediction (SEP) = 5.7, residual product differential (RPD) = 1.69 for in situ, (ii) SEP = 5.3, RPD = 1.83 for laboratory measurements of intact cores and (iii) SEP = 4.7, RPD = 2.0 for laboratory measurements of dried, sieved soil. The two primary factors contributing to larger in situ errors of prediction relative to laboratory‐based measurements of dried, sieved soil were (i) scanning an intact soil surface at field moisture compared with dried, sieved‐homogenized soil and (ii) problems with instrument drift and referencing in the field. For four of the 10 fields with similar loess soil and 24 cores per field, we found that for all soil preparations (in situ, intact core and dried, sieved) field‐specific calibrations (SEP = 5.1, RPD = 1.29 for in situ; SEP = 3.3, RPD = 2.0 for dried, sieved samples) out‐performed both a single calibration for the four fields and a calibration that included all 10 fields in this study. More research is required to understand better the effects of examining intact rather than crushed‐sieved soil material, a potentially important factor for in situ spectroscopy. In situ predictions could probably be improved with a more stable and internally referenced field spectrometer, but even with these challenges the results show that penetrometer‐derived in situ spectroscopy can be used to map the variation in soil profile clay content semi‐quantitatively. HIGHLIGHTS: We evaluated a new Vis–NIR penetrometer foreoptic to predict clay content in situ. We compared sources of error from in situ and laboratory Vis–NIR. Sample preparation markedly affected (intact soil or dried, sieved) Vis–NIR model performance. We identified major differences in clay content with penetrometer‐based Vis–NIR spectroscopy.