How much organic carbon could the soil store? The carbon sequestration potential of Australian soil.

Soil is a huge carbon (C) reservoir, but where and how much extra C can be stored is unknown. Current methods to estimate the maximum amount of mineral-associated organic carbon (MAOC) stabilized in the fine fraction (clay + silt, < 20 µm $$ <20\;\upmu \mathrm{m} $$ ) fit through the MAOC versus clay + silt relationship, not their maxima, making their estimates more uncertain and unreliable. We need a function that 'envelopes' that relationship. Here, using 5089 observations, we estimated that the uppermost 30 cm of Australian soil holds 13 Gt (10-18 Gt) of MAOC. We then fitted frontier lines, by soil type, to the relationship between MAOC and the percentage of clay + silt to estimate the maximum amounts of MAOC that Australian soils could store in their current environments, and calculated the MAOC deficit, or C sequestration potential. We propagated the uncertainties from the frontier line fitting and mapped the estimates of these values over Australia using machine learning and kriging with external drift. The maps show regions where the soil is more in MAOC deficit and has greater sequestration potential. The modelling shows that the variation over the whole continent is determined mainly by climate, linked to vegetation and soil mineralogy. We find that the MAOC deficit in Australian soil is 40 Gt (25-60 Gt). The deficit in the vast rangelands is 20.84 Gt (13.97-29.70 Gt) and the deficit in cropping soil is 1.63 Gt (1.12-2.32 Gt). Management could increase C sequestration in these regions if the climate allowed it. Our findings provide new information on the C sequestration potential of Australian soils and highlight priority regions for soil management. Australia could benefit environmentally, socially and economically by unlocking even a tiny portion of its soil's C sequestration potential.

Automated spectroscopic modelling with optimised convolutional neural networks.

Convolutional neural networks (CNN) for spectroscopic modelling are currently tuned manually, and the effects of their hyperparameters are not analysed. These can result in sub-optimal models. Here, we propose an approach to tune one-dimensional CNN (1D-CNNs) automatically. It consists of a parametric representation of 1D-CNNs and an optimisation of hyperparameters to maximise a model's performance. We used a large European soil spectroscopic database to demonstrate our approach for estimating soil organic carbon (SOC) contents. To assess the optimisation, we compared it to random search, and to understand the effects of the hyperparameters, we calculated their importance using functional Analysis of Variance. Compared to random search, the optimisation produced better final results and showed faster convergence. The optimal model produced the most accurate estimates of SOC with [Formula: see text] (s.d.) and [Formula: see text] (s.d.). The hyperparameters associated with model training and architecture critically affected the model's performance, while those related to the spectral preprocessing had little effect. The optimisation searched through a complex hyperparameter space and returned an optimal 1D-CNN. Our approach simplified the development of 1D-CNNs for spectroscopic modelling by automatically selecting hyperparameters and preprocessing methods. Hyperparameter importance analysis shed light on the tuning process and increased the model's reliability.

Wavelet geographically weighted regression for spectroscopic modelling of soil properties.

Soil properties, such as organic carbon, pH and clay content, are critical indicators of ecosystem function. Visible-near infrared (vis-NIR) reflectance spectroscopy has been widely used to cost-efficiently estimate such soil properties. Multivariate modelling, such as partial least squares regression (PLSR), and machine learning are the most common methods for modelling soil properties with spectra. Often, such models do not account for the multiresolution information presented in the vis-NIR signal, or the spatial variation in the data. To address these potential shortcomings, we used wavelets to decompose the vis-NIR spectra of 226 soils from agricultural and forested regions in south-western Western Australia and developed a wavelet geographically weighted regression (WGWR) for estimating soil organic carbon content, clay content and pH. To evaluate the WGWR models, we compared them to linear models derived with multiresolution data from a wavelet decomposition (WLR) and PLSR without multiresolution information. Overall, validation of the WGWR models produced more accurate estimates of the soil properties than WLR and PLSR. Around 3.5-49.1% of the improvement in the estimates was due to the multiresolution analysis and 1.0-5.2% due to the integration of spatial information in the modelling. The WGWR improves the modelling of soil properties with spectra.