Soil is an invaluable finite resource, and there is considerable interest in monitoring the status of soil, as well as the direction and degree of any changes in soil attributes. Land use change and agricultural management have the capacity to alter the properties of soil considerably over relatively short time scales, however, it is less clear how recent climatic shifts observed throughout the globe will influence changes in soil condition. In the semi-arid regions of eastern Australia, there has been an expansion of irrigated cotton production, which is considered to be a very intensive land use with vigorous management practices. These regions have also been exposed to significant fluctuations in rainfall patterns in the last decade or so. While these semi-arid areas are agriculturally important, they often possess distinct soil characteristics, such as high levels of alkalinity, salinity, sodicity, inorganic carbon, and low levels of organic carbon. This body of work focuses on monitoring the change in soil condition in the semi-arid irrigated cotton-growing district of Hillston in the lower Lachlan River valley catchment in south-west, New South Wales (NSW), Australia. Data from soil cores extracted to 1.5 m depth from two soil surveys performed in 2002 and 2015 were used to monitor the change in several important soil properties – pH, electrical conductivity, exchangeable sodium percentage, organic carbon, and inorganic carbon. It is anticipated that the significant shifts in land use and rainfall patterns could have altered the condition of soil during this period.

Rather than using traditional digital soil mapping techniques, such as regression kriging or machine learning approaches, this study focuses on using linear mixed models, which are particularly advantageous for monitoring changes in soil properties as they can account for correlation in space and time. In this work, the focus is on using bivariate linear mixed models (BLMMs) and multivariate linear mixed models (MLMMs) to create digital maps of the various soil properties. In the BLMM approach, one model is used to predict a soil property from both time points at a single depth, which results in improved soil maps that have a logical connection through time. The MLMM approach is similar to the BLMM approach, but multiple depths are also be modelled simultaneously in addition, which results in more coherent connections between the different sampling depths. Another strong advantage of using these approaches to monitor soil is that the correlation between the monitoring periods is used to improve the sensitivity of the model to detect statistically significant changes. Traditional laboratory methods of measuring certain soil properties can be expensive and laborious. This study used visible near infrared (VisNIR) spectroscopic techniques to rapidly predict soil exchangeable sodium percentage (ESP), soil organic carbon (SOC) content, and soil inorganic carbon (SIC) content to overcome this.