Miti, Chawezi
(2025)
Boundary line methodology for yield gap analysis of farm systems.
PhD thesis, University of Nottingham.
Abstract
The growing global food demand necessitates improvements in agricultural productivity to achieve global food security. This can be attained by closing the current yield gaps in cropping systems. The boundary line methodology has emerged as a popular tool for determination of yield gaps and has been widely used in many agronomic studies on yield analysis. However, there is no standard procedure for fitting boundary line models and there is a lack of exploratory analysis tools to ascertain the suitability of data for fitting a boundary line model. Furthermore, whereas the fitting of other types of statistical model is supported by the availability of functions in statistical software no such tool is available for applying boundary line analysis. In this thesis I advance the application of the boundary line methodology in yield gap analysis. First, I undertook a comprehensive review of the use of the boundary line methodology in yield gap analysis. This review revealed inconsistencies in application, particularly in the boundary line fitting procedures. Heuristic methods, though widely used, rely on subjective decisions such as the choice of bin width and often lack a statistical basis for assessing the suitability of the boundary model, underscoring the need for standardized and reproducible approaches. My research addresses these gaps by proposing an exploratory data analysis method using convex hull peels, which assesses whether datasets exhibit boundary-limited responses. This method improves the reliability of boundary line analysis by preventing misapplications to datasets that do not support boundary constraints. I also present a comparative evaluation of multiple boundary line fitting techniques, including binning, boundary line determination technique, quantile regression, and the censored bivariate normal model. Results showed that the censored bivariate normal model provides more stable estimates of critical values than heuristic methods, enhancing the precision of yield-limiting factor identification. However, the original model lacked the ability to accommodate categorical independent variables. This limitation was addressed by developing a framework that extends the censored model to handle categorical data. In this study, data size was found to significantly influence boundary line model fitting. In many cases, smaller datasets did not provide sufficient statistical evidence to support the fitted boundary models. A significant outcome of my research was the development of the BLA R package, which integrates multiple boundary line methodologies, exploratory tools (including those I developed), and interactive functions for enhanced usability. This open-source tool promotes transparency, reproducibility, and accessibility for researchers, aiding in the robust application of boundary line analysis in yield gap assessments. To test the useability of the BLA R package and to evaluate how users engage with various boundary line methods, I held stakeholder engagement workshops in Nairobi and Harare to elicit their opinions on preferred boundary line fitting methods. Although no specific method was favoured, participants emphasized the need for a more objective approach to fitting boundary line models in agronomic research. This research provides a critical foundation for improving yield gap assessments, supporting sustainable agricultural intensification, and contributing to global food security through data-driven decision-making.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
