Spatial estimation of surface soil texture using Landsat 8 data
Loading...
Date
2021
Authors
Mtshawu, Babalwa
Journal Title
Journal ISSN
Volume Title
Publisher
University of the Free State
Abstract
The current availability of quantitative soil information does not meet the required quality and coverage for environmental monitoring and modelling of regional- to global-scale soil mapping. Conventional soil sampling and laboratory analysis cannot effectively provide this information because these methods are slow, expensive, and cannot capture all temporal and spatial variability. Remote sensing has in the past three decades shown high potential in the determination of soil characteristics. Different methodologies have been proposed for the estimation of soil parameters based on different remote sensors and technologies. Even with these methods, characterising soil parameters has not worked beyond a scale with sufficient homogeneity (with regard to vegetation, soil, topography, and microclimate) due to local calibration models. For policymaking, management of land resources, and monitoring of environmental impacts, these methods need to be extended beyond this scale.
This study evaluated the possibility of utilising Landsat 8 at surface reflectance (L8SR) Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) data as auxiliary variables for the estimation of surface soil texture using a limited number of legacy samples provided by the Agricultural Research Council: Institute for Soil, Climate and Water (ARC-ISCW), which were taken from the Lejweleputswa District Municipality in the Free State province, South Africa. Ordinary least squares regression (OLSR) and geographically weighted regression (GWR) were utilised to predict and assess the variability of surface soil texture. The data from OLI and TIRS and L8SR were used as explanatory variables and the ARC-ISCW legacy soil texture attributes as response variables. The methodology followed in this research led to the development of a new Optimised Bare Soil Index (OBI). Unlike previously developed bareness indices, the OBI made use of surface reflectance data, which provided a realistic depiction of the elements on the surface of the earth.
The OBI results showed better contrast between bare soil and other elements identified in the study area, which means that the OBI provided better bare soil delineation than previously developed bare soil indices. The OBI results allowed for only bare soil regions of the study area to be sampled for L8SR data. The correlation analysis showed that ARC-ISCW legacy surface sand, silt, and clay contents were significantly correlated with the L8SR digital number (DN) data of the seven bands. Band 6 explained most of the variability in soil, sand, and silt, while Band 5 explained most of the variability in clay contents. The DN of Band 6 for surface sand and silt and the DN of Band 5 for clay were selected as auxiliary data for the estimation of surface soil texture. Regrettably, the key explanatory variables proved to be too complex to be represented by the global OLSR model. Using the same key explanatory variables, GWR significantly improved the surface soil texture estimates. The prediction accuracies provided
by GWR were deemed to be reasonable considering the high variability in land use practices, environmental variables, and the size of the study area. The methodology proposed in this study is promising for future local- to regional-scale digital soil mapping efforts in data-poor regions. The use of L8SR data was shown to reduce soil sampling efforts and therefore reducing soil mapping costs.
Description
Keywords
Landsat 8 at surface reflectance, Soil texture, Spatial variation, Remote sensing, Ordinary least squares regression, Geographically weighted regression, Soil texture -- South Africa -- Free State, Thesis (Ph.D. (Geography))--University of the Free State, 2021