34370

Prediction of Soil Bulk Density for a set of Scottish Afforested Soils: a Comparison of Approaches

Jean Robertson 1 ; Estefania Perez-Fernandez 1 ; Jackie Potts 2 ; Allan Lilly 1
Favorite this paper

Introduction

Soil bulk density (BD) is required for the calculation of C stocks but has not always been measured during sampling. This problem can be overcome by the development of near infrared (NIR) calibrations to predict soil BD.

To allow the calculation of the changes in C stocks of Scottish soils, NIR calibrations to predict BD were developed for the National Soils Inventory of Scotland (NSIS), a dataset that represent the main soil types in Scotland. In 2013, a set of soils from sites in Scotland that had been afforested, were sampled to quantify changes in their carbon stocks. These sites were previously sampled between 1961 and 1988, but no measurements of the BD were made at the time.

The objectives of this study were to 1) to test the applicability of calibrations already developed for the NSIS dataset for predicting BD on the afforested soils dataset and 2) to determine the best possible approach to predicting the BD values for the afforested soils dataset.

Experimental

A total of 701 horizons from 183 soil profiles at NSIS sites across Scotland (20 km x 20 km grid), were sampled. The afforested soil dataset consisted of 39 profiles (239 soil horizons). Soil BD measurements were taken in triplicate (210cm3 cores), wherever possible, and averaged.

The samples used for NIR analysis were air dried (30°C) and sieved (<2mm). The NIR spectra, in reflectance mode, were recorded at 2nm intervals (1100 to 2500 nm) using a FOSS NIRSystems 5000 spectrophotometer (FOSS NIRSystems, Silver Springs, MD, USA) with a transport module sampling attachment and a quarter cup sample holder. Infrasoft International WinISI III software, version 1.63 (FOSS Tecator AB, Hoganas, Sweden) was used to develop calibration equations to predict the soil BD.

A range of calibrations, to predict soil BD for the afforested soils dataset, were developed:

1) A global calibration for the whole NSIS dataset
2) Separate global calibrations for high and low %C samples for the NSIS dataset
3) A local calibration for the NSIS dataset
4) A global calibration developed for the afforested soils dataset (86 samples for validation)
5) A local calibration combining NSIS and afforested soil spectral libraries (80 samples for validation)

Results and discussion
The spectral variability of the afforested soils fitted well within the NSIS soils population with a small number of outliers which were generally litter layers and those rare soils underrepresented within the NSIS population.
Using the NSIS dataset, the best predictions were achieved when the full dataset was used for horizons with <37% C but, for those with >37% C, it was more accurate to apply a calibration for NSIS samples with > 37% C (Standard error of prediction (SEP) = 0.034 g cm-3 and 0.284 g cm-3 for C content >37% and <37% respectively). The local calibration, incorporating the NSIS samples and the afforested soil samples, gave the best results for calibrations which included the afforested soils, (SEP for the validation set = 0.21 g cm-3).