We present a new method to estimate the relative abundance of the dominant phyla and diversity of fungi in Australian soil. It uses state-of-the-art machine learning with publicly available data on soil and environmental proxies for edaphic, climatic, biotic and topographic factors, and visible–near infrared wavelengths. The estimates could serve to supplement the more expensive molecular approaches towards a better understanding of soil fungal abundance and diversity in agronomy and ecology.

We used soil parent material, terrain attributes, and geophysical data from the soil surface to test and compare different and unprecedented geophysical sensor combination, as well as different machine learning algorithms to model and predict several soil attributes. Also, we analyzed the importance of pedoenvironmental variables. The soil attributes were modeled throughout different machine learning algorithms and related to different geophysical sensor combinations.

We performed Roth C simulations across Australia and assessed the response of soil carbon to changing inputs and future climate change using a consistent modelling framework. Site-specific initialisation of the C pools with measurements of the C fractions is essential for accurate simulations of soil organic C stocks and composition at a large scale. With further warming, Australian soils will become more vulnerable to C loss: natural environments > native grazing > cropping > modified grazing.

We developed the Swiss mid-infrared spectral library and a statistical model collection across 4374 soil samples with reference measurements of 16 properties. Our library incorporates soil from 1094 grid locations and 71 long-term monitoring sites. This work confirms once again that nationwide spectral libraries with diverse soils can reliably feed information to a fast chemical diagnosis. Our data-driven reduction of the library has the potential to accurately monitor carbon at the plot scale.

In this study, we show that a soil spectral library (SSL) can be used to predict soil carbon at new and very different locations. The importance of this finding is that it requires less time-consuming lab work than calibrating a new model for every local application, while still remaining similar to or more accurate than local models. Furthermore, we show that this method even works for predicting (drained) peat soils, using a SSL with mostly mineral soils containing much less soil carbon.

Using the data from 141 584 whole-soil profiles across the globe, we disentangled the relative importance of biotic, climatic and edaphic variables in controlling global SOC stocks. The results suggested that soil properties and climate contributed similarly to the explained global variance of SOC in four sequential soil layers down to 2 m. However, the most important individual controls are consistently soil-related, challenging current climate-driven framework of SOC dynamics.

The number of samples utilised to create predictive models affected model performance. This research compares the number of samples needed by a deep learning model to outperform the traditional machine learning models using visible near-infrared spectroscopy data for soil properties predictions. The deep learning model was found to outperform machine learning models when the sample size was above 2000.

Soil nitrogen (N) is an essential element for plant growth, but its plant-available forms are subject to loss from the environment by leaching and gaseous emissions. Still, factors controlling soil mineral N concentrations at large spatial scales are not well understood. We determined and discussed primary soil controls over the concentrations of NH₄⁺ and NO₃⁻ at the continental scale of Australia while considering specific dominant land use patterns on a regional basis.

Proximal sensing can be used for soil C accounting, but the methods need to be standardized and procedural guidelines developed to ensure proficient measurement and accurate reporting. This is particularly important if there are financial incentives for landholders to adopt practices to sequester C. We review sensing for C accounting and discuss the requirements for the development of new soil C accounting methods based on sensing, including requirements for reporting, auditing and verification.