A platform to interpret soil attributes to support profitable farming systems.

Abstract

This collection contains data on some estimated and mapped soil physical properties such as: clay, sand, pH, Cation Exchange Capacity (CEC), and organic carbon (OC) generated through predictive models using a developed framework that quantitatively assess the accuracy of data collected with proximal soil sensors and spectroscopic techniques such as visible near-infrared visNIR and portable X-ray Fluorescence (pXRF) spectroscopy.

First, tools were provided to assist in the collation of freely available data such as elevation and satellite derived data as well as on-farm data produced with electromagnetic induction (EM) and gama radiometricts. Second, an automated site selection software was developed to collate and process covariate data to identify 25 samples sites across L'lara, a mixed-farming property located ~11 km Narrabri in NSW in 2020. Fieldwork and example mapping soil properties were conducted at L'lara using visNIR spectrocopy and pXRF spectrometers. A conditioned Latin hypercube sampling design was chosen to sample the distribution of the covariate space under both cropping and pasture on the 1,850 ha property. Covariate data supplied to the software included on-site EM, gamma radiometrics, yield, soil legacy data, plus elevation and satellite derived data.

A soil inference system (SPEC-SINFERS) was developed, using other spectrally active properties through pedotransfer functions (PTFs) to predic further properties such as available water capacity (AWC) from sensor predicted properties.

A large spectral library was construted with > 8,000 pre-existing soil samples predominantly from grain-growing regions of NSW and additional accession from Qld., Victoria and Tasmania and fieldwork data.

Multi-depth mapping of soil properties and attributes (Depth-to pH constraint) was also investigated to provide agronomic interpretations to the produced soil maps and correlations with available yield data. The accuracy of mapped soil properties was tested under data-rich and data-poor scenarios. Calibration and validations of each scenario were made with laboratory data, available covariate data (elevation, satellite image) and with/without on-farm colleted EM and gamma data. RMSE was used in percentage change as reference to other studies. Mapped yield products revealed significant correlations for canola, chickpea and wheat in two paddocks over two growing seasons.

Datasets generated for this project are stored in the RDS - GRDC_US00087 (\\shared.sydney.edu.au\research-data\PRJ-GRDC_US00087). Please contact Prof. Alex McBratney to request access to them.