Soil spectral characteristics and their predictive value in relation to spatial and temporal variability in wheat yield and soil quality within a long-term field trial

Ngejane, Nompumelelo (2019-12)

Thesis (MScAgric)--Stellenbosch University, 2019.

Thesis

ENGLISH ABSTRACT: Soil is a heterogeneous growing medium, with complex processes and mechanisms that are not easy to be fully understood. Soil spatial variations may be encountered within short distances, and these variations, directly and indirectly, affect plant production and crop yields. The field of agriculture is facing an escalating demand of databases from a regional to a worldwide scale that will help agriculturists understand and be able to mitigate the impact of spatial variations in the field (soils and crops). However, to make such data available is expensive and involves tedious and labor-intensive methods. Rapid and cost-effective tools to measure variations in soil properties and crop yields for large areas are required. Soil spectroscopy appears to be a fast, nondestructive, cost-effective, environmental-friendly, reproducible, and repeatable analytical technique. The study aims at evaluating the use of soil spectroscopy in predicting common selected soil properties and wheat yield, as well as exploring its potential in explaining the spatial variations in the field, both in soil properties and in wheat yield. The experiment was conducted as a long-term ongoing trial at the Langgewens research farm, Western Cape Department of Agriculture. The trial was laid out in an incomplete block design structure, across a 12 ha area made up of three cropping systems with varying degrees of crop diversity and four replicates allocated in 120 plots. Archived soil samples (for the year 2015) from all the 120 plots were used for the analysis of selected common soil properties and scanned to acquire the near-infrared (NIR) spectral signatures using a spectrophotometer (Bruker Multi-Purpose Analyzer). The NIR spectral signatures were pre-processed following two procedures that are de-noising (removal of the fringe bands with large noise) and data transformation (first derivative and straight-line subtraction) before performing the multivariate data analyses. The partial least squares regression (PLSR) method was used to develop chemometric models to establish the relationship that the NIR spectral signatures have with wheat yield and soil properties. The prediction results of the PLSR models were fairly accurate and falling within the acceptable ranges. For the selected models, most correlation coefficients (R2) ranged between 0.80 and 0.60 with the ratios of performance to deviation (RPD) ranging between 2.38 to 1.6 for wheat yield and selected soil properties (CEC, SOC, pH, Ca). In 2019, the soil core samples at 0-5 cm depth (120 in total) were analyzed for some key soil parameters and were scanned to acquire the NIR spectral signatures. This was done to assess the temporal variations in wheat yield and changes in 5 cm soil spectral characteristics in the field trial area after four years (2015 to 2019). An overall significant difference was obtained between the averaged spectral absorbance for the years 2015 and 2019 (p<0.05), an increase in absorbance for the year 2019 was observed. When assessing the changes that have occurred in some selected soil properties, bulk density was observed to have significantly decreased across the field (p<0.05). A decrease in soil organic carbon (p<0.05) was observed as well as in soil organic carbon stocks (p<0.05) within a fixed depth. However, no significant change was observed in soil carbon stocks when the depth was adjusted. Results obtained in this study show that to a certain extent, the spectral characteristics in the NIR region might be a good indicator of not only soil properties but also plant responses to the changes in soil properties across the field.

AFRIKAANSE OPSOMMING: Die grond is 'n heterogene groeimedium, met komplekse prosesse en meganismes wat nie maklik verstaanbaar is nie. Ruimtelike variasies in die grond kan binne kort afstande voorkom, en hierdie variasies beïnvloed plantproduksie en oesopbrengste direk en indirek. Die landbouveld het 'n vinnig groeiende vraag na databasissevan 'n plaaslike tot 'n globale skaal, wat landboukundiges sal help om die impak van die ruimtelike variasies in die veld (grond en gewasse) te verstaan en te verminder. Om sulke data beskikbaar te stel, is egter duur en behels tydsame, arbeidsintensiewe metodes. Vinnige en kostedoeltreffende instrumente om variasies in grondeienskappe en oesopbrengste vir groot gebiede te meet, is nodig. Grondspektroskopie bied 'n vinnige, nie-vernietigende, koste-effektiewe, omgewingsvriendelike, en herhaalbare analitiese metode. Die studie het ten doel om die gebruik van grondspektroskopie te evalueer om algemene geselekteerde grondeienskappe en koringopbrengste te voorspel, asook om die potensiaal daarvan te ondersoek om die ruimtelike variasies in die veld, sowel in grondeienskappe as koringopbrengs, te verklaar. Die eksperiment is uitgevoer as 'n langtermyn proef op die Langgewens-navorsingsplaas, van die Wes-Kaapse Departement van Landbou. Die proef is uitgevoer in 'n onvolledige blok ontwerpstruktuur , oor 'n oppervlakte van 12 ha wat bestaan uit drie teeltstelsels met wisselende gewasverskeidenheid en vier replikasies toegeken in 120 erwe. Grondmonsters vanuit die argief (die jaar 2015) van al 120 plotte is gebruik vir ontleding van geselekteerde gemeenskaplike grondeienskappe en geskandeer om die naby-infrarooi (NIR) spektrale handtekeninge te verkry, met behulp van 'n spektrofotometer (Bruker Multi-Purpose Analyzer). Die NIR-spektrale handtekeninge is vooraf verwerk volgens twee prosedures wat agtergrond geraas verwyder (die prosses waarby onnodige data bande verwyder word) en datatransformasie (eerste afgeleide en reguitlyn aftrek) voordat die multiveranderlike data-ontleding uitgevoer is. Die metode van gedeeltelike kleinste kwadraatregressie (GKKR) is gebruik om chemometriese modelle te ontwikkel om die verwantskap wat die NIR-spektrale handtekeninge het met koringopbrengs en grondeienskappe te bepaal. Die voorspellingsresultate van die GKKR-modelle was redelik akkuraat en het binne aanvaarbare reekse geval. Vir die geselekteerde modelle het die meeste korrelasiekoëffisiënte (R2) tussen 0,80 en 0,60 gewissel, met die verhoudings van prestasie tot afwyking (RPD) tussen 2,38 en 1,6 vir koringopbrengs en geselekteerde grondeienskappe (katioonuitruilkapasiteit, organiese koolstof, pH, uitruilbare kalsium). In 2019 is die grondkernmonsters op 0-5 cm diepte (120 in totaal) geanaliseer vir enkele belangrike grondparameters en is geskandeer om die NIR-spektrale handtekeninge te bekom. Dit is gedoen om die tydelike variasies in koringopbrengs en veranderinge in 5 cm grondspektrale eienskappe in die veldproefgebied na vier jaar (2015 tot 2019) te bepaal. 'N beduidende verskil is verkry tussen die gemiddelde spektrale absorbansie vir die jare 2015 en 2019 (p <0,05), 'n toename in absorbansie vir die jaar 2019 is waargeneem. By die ondersoek na die veranderinge wat plaasgevind het in sekere geselekteerde grondeienskappe, is daar waargeneem dat die massa digtheid oor die erwe beduidend afgeneem het (p <0.05). 'N afname in organiese koolstof (p<0.05) sowel as in organiese koolstofvoorrade in die grond (p<0.05) is binne 'n vaste diepte waargeneem. Daar is egter geen noemenswaardige verandering in die grondstofkoolstofvoorraad waargeneem toe die diepte aangepas is nie. Resultate wat in hierdie studie verkry is, toon dat die spektrale eienskappe in die NIR-band tot 'n sekere mate 'n goeie aanduiding kan wees van nie net grondeienskappe nie, maar ook plantresponse op die veranderinge in grondeienskappe oor die hele veld.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/107116
This item appears in the following collections: