|Abstract: ||This thesis is concerned with practicable methodologies for delivering comprehensive spatial soil information to end-users. There is a need for relevant spatial soil information to complement objective decision-making for addressing current problems associated with soil degradation; for modelling, monitoring and measurement of particular soil services; and for the general management of soil resources. These are real-world situations, which operate at spatial scales ranging from field to global scales. As such, comprehensive spatial soil information is tailored to meet the spatial scale specifications of the end user, and is of a nature that fully characterises the whole-soil profile with associated prediction uncertainties, and where possible, both the predictions and uncertainties have been independently validated. ‘Practicable’ is an idealistic pursuit but nonetheless necessary because of a need to equip land-holders, private-sector and non-governmental stakeholders and, governmental departments including soil mapping agencies with the necessary tools to ensure wide application of the methodologies to match the demand for relevant spatial soil information. Practicable methodologies are general and computationally efficient; can be applied to a wide range of soil attributes; can handle variable qualities of data; and are effective when working with very large datasets.
In this thesis, delivering comprehensive spatial soil information relies on coupling legacy soil information (principally site observations made in the field) with Digital Soil Mapping (DSM) which comprises quantitative, state-of-the-art technologies for soil mapping. After the General Introduction, a review of the literature is given in Chapter 1 which describes the research context of the thesis. The review describes soil mapping first from a historical perspective and rudimentary efforts of mapping soils and then tracks the succession of advances that have been made towards the realisation of populated, digital spatial soil information databases where measures of prediction certainties are also expressed. From the findings of the review, in order to deliver comprehensive spatial soil information to end-users, new research was required to investigate: 1) a general method for digital soil mapping the whole-profile (effectively pseudo-3D) distribution of soil properties; 2) a general method for quantifying the total prediction uncertainties of the digital soil maps that describe the whole-profile distribution of soil properties; 3) a method for validating the whole-profile predictions of soil properties and the quantifications of their uncertainties; 4) a systematic framework for scale manipulations or upscaling and downscaling techniques for digital soil mapping as a means of generating soil information products tailored to the needs of soil information users. Chapters 2 to 6 set about investigating how we might go about doing these with a succession of practicable methodologies.
Chapter 2 addressed the need for whole-profile mapping of soil property distribution. Equal-area spline depth functions coupled with DSM facilitated continuous mapping the lateral and vertical distribution of soil properties. The spline function is a useful tool for deriving the continuous variation of soil properties from soil profile and core observations and is also suitable to use for a number of different soil properties. Generally, mapping the continuous depth function of soil properties reveals that the accuracy of the models is highest at the soil surface but progressively decreases with increasing soil depth.
Chapter 3 complements the investigations made in Chapter 2 where an empirical method of quantifying prediction uncertainties from DSM was devised. This method was applied for quantifying the uncertainties of whole-profile digital soil maps. Prediction uncertainty with the devised empirical method is expressed as a prediction interval of the underlying model errors. The method is practicable in the sense that it accounts for all sources of uncertainty and is computationally efficient. Furthermore the method is amenable in situations where complex spatial soil prediction functions such as regression kriging approaches are used.
Proper evaluation of digital soil maps requires testing the predictions and the quantification of the prediction uncertainties. Chapter 4 devised two new criteria in which to properly evaluate digital soil maps when additional soil samples collected by probability sampling are used for validation. The first criterion addresses the accuracy of the predictions in the presence of uncertainties and is the spatial average of the statistical expectation of the Mean Square Error of a simulated random value (MSES). The second criterion addresses the quality of the uncertainties which is estimated as the total proportion of the study area where the (1-α)-prediction interval (PI) covers the true value (APCP). Ideally these criteria will be coupled with conventional measures of map quality so that objective decisions can be made about the reliability and subsequent suitability of a map for a given purpose. It was revealed in Chapter 4, that the quantifications of uncertainty are susceptible to bias as a result of using legacy soil data to construct spatial soil prediction functions. As a consequence, in addition to an increasing uncertainty with soil depth, there is increasing misspecification of the prediction uncertainties.
Chapter 2, 3, and 4 thus represent a framework for delivering whole-soil profile predictions of soil properties and their uncertainties, where both have been assessed or validated across mapping domains at a range of spatial scales for addressing field, farm, regional, catchment, national, continental or global soil-related problems. The direction of Chapters 5 and 6 however addresses issues specifically related to tailoring spatial soil information to the scale specifications of the end-user through the use of scale manipulations on existing digital soil maps. What is proposed in Chapter 5 is a scaling framework that takes into account the scaling triplet of digital soil maps—extent, resolution, and support—and recommends pedometric methodologies for scale manipulation based on the scale entities of the source and destination maps. Upscaling and downscaling are descriptors for moving up to coarser or down to finer scales respectively but may be too general for DSM. Subsequently Fine-gridding and coarse-gridding are operations where the grid spacing changes but support remains unchanged. Deconvolution and convolution are operations where the support always changes, which may or may not involve changing the grid spacing. While disseveration and conflation operations occur when the support and grid size are equal and both are then changed equally and simultaneously.
There is an increasing richness of data sources describing the physical distribution of the Earth’s resources with improved qualities and resolutions. To take advantage of this, Chapter 6 devises a novel procedure for downscaling, involving disseveration. The method attempts to maintain the mass balance of the fine scaled predictions with the available coarse scaled information, through an iterative algorithm which attempts to reconstruct the variation of a property at a prescribed fine scale through an empirical function using environmental or covariate information. One of the advantages associated with the devised method is that soil property uncertainties at the coarse scale can be incorporated into the downscaling algorithm.
Finally Chapter 7 presents a synthesis of the investigations made in Chapters 2 to 6 and summarises the pertinent findings. Directly from the investigations carried out during this project there are opportunities for further work; both in terms of addressing shortcomings that were highlighted but not investigated in the thesis, and more generally for advancing digital soil mapping to an operational status and beyond.|