Chagumaira, Christopher
(2022)
Geospatial modelling of soil geochemistry at national-scale for improved human nutrition.
PhD thesis, University of Nottingham.
Abstract
Mineral micronutrient deficiencies (MND), prevalent in Ethiopia and Malawi among most countries in sub-Saharan Africa, are linked to soil type. Dietary mineral intake is influenced by mineral content of edible portions of crops, and there is strong evidence that cereals grown in these regions have low uptake of micronutrients. The low nutrient uptake is attributed to soil conditions. Spatial information on soil and crop properties is therefore required to improve local estimates of MND risk in order to implement targeted and efficient interventions. Obtaining spatial information on soil micronutrients status and other relevant properties that affect their uptake requires substantial effort, and there are uncertainties in the resulting information, which depend, in part on the methods used for prediction and the sampling design. Therefore, it is necessary to use robust and efficient methods for spatial prediction which characterise the uncertainty of the predictions reliably. Furthermore, it is necessary that these uncertainties can be communicated effectively to stakeholder groups so that they can account for them at all stages from commissioning the survey through to making decisions based on the information.
In this study, it was important first to understand how uncertain spatial information can be communicated to stakeholders (e.g., those in public health or nutrition and agronomy or soil science) through a systematic evaluation in the forms of maps. Evaluation of the test methods were done through a structured elicitation of the opinions of members of a stakeholder group about the usefulness of the methods. Stakeholders found that general measures of uncertainty, such as prediction error variances (e.g., kriging variance) were less clear than measures which integrated the uncertainty explicitly with the decision—e.g., the probability that the true value of a variable at a site if interest falls below a critical threshold. There was no evidence that they found verbal phrases these (e.g., “very uncertain”) clearer than numerical values (i.e., a probability in the interval [0,1]).
Following on this finding, it was necessary to examine how stakeholders interpret such probability information in more detail. Specifically, is it possible to estimate a probability threshold which a stakeholder group would choose to intervene, reflecting their assessment of the costs attached to errors of commission and omission? Further does this probability depend on framing of the problem (e.g., probability that a threshold is exceeded or that it is not exceeded) and does it depend on professional background of the stakeholder? In a designed experiment, stakeholders were presented with uncertain information on micronutrient supply from a crop, with the uncertainty expressed as a probability with positive framing (probability of adequate supply) or negative framing (probability of insufficient supply). The results showed that probabilities presented in a negative framing led to more conservative decisions, i.e., deciding to intervene at a much smaller probability of deficiency than if the equivalent probability of sufficiency were presented. The elicited probability threshold is prone to framing effects (i.e., how the question is posed), and that this effect interacts with professional group.
The two components of this thesis described above showed how uncertain information can be effectively communicated to stakeholders to support decisions. The next task was to develop a framework for the planning, execution, and evaluation of surveys to address specific requirements of these stakeholders. This was based on a decision-theory approach to analyse the particular task, to identify the key uncertainties and their implications and so to enable stakeholders to ensure that an approach to survey would meet their needs. A particular task, based on research practices within the GeoNutrition project (Bill & Melinda Gates funded) was identified —the selection of the study sites to evaluate agronomic biofortification strategies for MND at selected sites based on soil soluble Se. The information required were analysed, and then the outputs of spatial prediction at national-scale of soil soluble Se by ordinary kriging, indicator kriging, linear mixed models and random forest were evaluated. There were substantial uncertainties by all three methods, and challenges with dealing with a complex statistical distribution. This work showed the importance of validation—internal and independent for understanding the suitability of spatial prediction to support decision making.
Uncertainty should be considered when planning sampling for a geostatistical survey. It is important to consider how stakeholders can assess the implications of uncertainty in spatial predictions to determine appropriate sampling grid space for a geostatistical survey. Four approaches (offset correlation, prediction intervals, conditional probabilities and implicit loss functions), that can be used to assess the implications of uncertainty in spatial predictions using prior information on variability of the target properties, were presented to a diverse group of stakeholders in order to determine an appropriate grid spacing. There were variations in the selection made by each method. Some were not well understood. The one which stakeholder favoured, offset correlation, is not directly linked to decision making. More work is needed to develop sound but accessible ways to engage stakeholders with uncertainty consistent across planning and interpretation. Findings from this research will help in better understanding of uncertainties in the data obtained in the GeoNutrition projects thereby facilitating improved use and uptake of that information by decision makers in Ethiopia and Malawi. Better decisions will be made on sampling for such surveys in other countries which decide to undertake those using better methodologies for national-scale surveys of soil properties or similar environmental variables.
| Item Type: |
Thesis (University of Nottingham only)
(PhD)
|
| Supervisors: |
Lark, R. Murray
Milne, Alice E.
Broadley, Martin R.
Nalivata, Patson C.
Chimungu, Joseph G. |
| Keywords: |
Micronutrient deficiencies, Uncertainty, Decision-making, Interpret probability, Spatial prediction |
| Subjects: |
Q Science > QE Geology |
| Faculties/Schools: |
UK Campuses > Faculty of Science > School of Biosciences |
| Item ID: |
71710 |
| Depositing User: |
Chagumaira, Christopher
|
| Date Deposited: |
14 Dec 2022 04:40 |
| Last Modified: |
14 Dec 2022 04:40 |
| URI: |
https://eprints.nottingham.ac.uk/id/eprint/71710 |
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
