Browsing by Author "Muller, Sybrand Jacobus"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- ItemIndirect soil salinity detection in irregated areas using earth observation methods(Stellenbosch : Stellenbosch University, 2017-03) Muller, Sybrand Jacobus; Van Niekerk, Adriaan; Stellenbosch University. Faculty of Arts and Social Sciences. Dept. of Geography & Environmental Studies.ENGLISH ABSTRACT: Excessive accumulation of salt in the plant root zone has a deteriorating effect on vegetation growth, resulting in reduced crop yield and rendering fertile soil barren, and ultimately leads to decreasing food production. This problem motivates the critical need for active salinity monitoring, with an aim to implement rehabilitation and preventive measures. Conventional salinity monitoring methods, such as regular field visits and laboratory analyses of soil samples, are ineffective for frequent salt accumulation monitoring over large areas. Earth observation techniques can complement conventional methods and potentially improve the cost- and time-efficiency of the regular salt accumulation monitoring. A review of literature identified a number of direct and indirect approaches for detecting accumulated salts using remote sensing. Given that salt accumulation in South Africa is most prevalent in irrigation schemes, the indirect approach, which mainly focusses on salt induced vegetation response, was identified as the preferred approach. Little is known about the optimal combinations of very high resolution satellite imagery and image classification techniques in South African irrigation schemes, where accumulated salt generally occurs in small localized patches. Consequently, two experiments were carried out to: identify suitable spectral and spatial resolutions of imagery; test the value of various image-derived features and classification techniques; and evaluate the spatial scale at which salt accumulation is best identified. The first experiment, applied on an agricultural field scale, analysed WorldView-2 imagery with statistical (regression) and machine learning (decision trees) techniques, and found clear relationships between salt accumulation and image transformations (vegetation indices and image texture). Spatial resolutions of six metres or higher were found to be most suitable. A higher spectral resolution marginally improved classification accuracies, but the increases were insignificant. The second experiment, applied on an irrigation scheme level, assessed SPOT-5 imagery with statistical (regression) and machine learning (various algorithms) techniques in two irrigation schemes. The failure to highlight any consistent image transformation or classification techniques for both irrigation schemes emphasised the negative effect of varying salinity tolerances of crop types and growing phases in identifying salt accumulation. Knowledge gained from the two experiments aided the development of a field level object-based monitoring system that showed sufficient transferability. The quantitative experiments answered key research questions and will serve as a point of departure for future research regarding indirect methods for detecting salt accumulation in agricultural fields. This work will be instrumental in the establishment of a South African salinity monitoring system – with the aim of rehabilitation – and will help to maximize agricultural production and ultimately contribute to sustainable food production.