Comparing local vs. global visible and near-infrared (VisNIR) diffuse reflectance spectroscopy (DRS) calibrations for the prediction of soil clay, organic C and inorganic C

Abstract

a b s t r a c t a r t i c l e i n f oLocal, field-scale, VisNIR-DRS soil calibrations generally yield the most accurate predictions but require a substantial number of local calibration samples at every application site. Global to regional calibrations are more economically efficient, but don't provide sufficient accuracy for many applications. In this study, we quantified the value of augmenting a large global spectral library with relatively few local calibration samples for VisNIR-DRS predictions of soil clay content (clay), organic carbon content (SOC), and inorganic carbon content (IC). VisNIR models were constructed with boosted regression trees employing global, local + global, and local spectral data, using local samples from two low-relief, sedimentary bedrock controlled, semiarid grassland sites, and one granitic, montane, subalpine forest site, in Montana, USA. The local + global calibration yielded the most accurate SOC predictions for all three sites [Standard Error of Prediction (SEP) = 3.8, 6.7, and 26.2 g kg ), which also had the largest number of local calibration samples (N = 210). Using only samples from calcareous soils in the global spectral library combined with local samples produced the best SOC and IC results at the more arid of the two semiarid sites. Global samples alone never achieved more accurate predictions than the best local + global calibrations. For the temperate soils used in this study, the augmentation of a large global spectral library with relatively few local samples generally improved the prediction of soil clay, SOC, and IC relative to global or local samples alone.

Transferability of a Visible and Near-Infrared Model for Soil Organic Matter Estimation in Riparian Landscapes

“…Although there have been extensive studies of the VNIR estimation of SOM content, it is generally agreed that most of the models are location-dependent and data-specific [13,19,20]. A frequently asked question by the potential users of VNIR models for SOM estimation is: will a model calibrated from samples in a certain location work in other locations?…”

Proximal sensing for soil carbon accounting

“…Vasques et al, 2010;Sankey et al, 2008a). Shi et al (2015) proposed the use of both spectral similarities and geographically constrained local calibrations to predict soil organic C content.…”

“…They are 'spiking', which uses several local spectra to augment 10 the calibration made with an ESL (e.g. Guerrero et al, 2010;Sankey et al, 2008b;Viscarra Rossel, R.A. et al, 2009), and spiking with extra-weighting (Guerrero et al, 2014a), which uses multiple copies of the local samples to improve their leverage in the calibrations. Guerrero et al (2014a) showed that the approach improved on spiking and suggested it might be more appropriate with larger spectral libraries.…”

Transferability and scaling of soil total carbon prediction models in Florida

“…It is interesting to note that Model 3 (developed in Pineland and Psamments), which had the highest variability in TC among field sites, performed poorest in terms of transferability to other Spiking of a spectral model with local samples has been suggested to improve soil predictions (Sankey et al, 2008;Wetterlind and Stenberg, 2010) though its success is highly dependent on multiple factors including the ratio between the 'number of spike samples' and 'number of samples in the spectral library', characteristics of the soil attribute and spectral domain spaces, and the methods used to develop spectral-based soil prediction models. Although the same constraints, mechanisms, and effects impact spiking and scaling of chemometric models, the aims are inherently different.…”