- Title
- Mapping soil factors influencing erosion using machine leaerning algorithms in the t35 d-e catchment in the Eastern Cape Province
- Creator
- Du Plessis, Casparus Jacobus
- Subject
- Soils -- Analysis Soil erosion
- Date
- 2019
- Type
- Thesis
- Type
- Masters
- Type
- MSc
- Identifier
- http://hdl.handle.net/10353/17106
- Identifier
- vital:40850
- Description
- The need for detailed spatial soil information is increasing in the fields of environmental studies, agriculture as well as engineering practices. Regarding soil erosion in South Africa, border lines between sensitive soil forms in the Duplex soil group and resistant soil forms in the Oxidic soil group need to be mapped to a high accuracy to prevent/manage erosion and major soil losses. In third world countries like South Africa, digital soil mapping (DSM) with the use of machine learning can provide the solutions to fill numerous gaps in these related fields. Local DSM research has been ongoing at the University of the Free State and related institutions for the past eleven years. It has also been used commercially and can therefore be regarded as fruitful. The National Resource Management unit of the Department of Environmental Affairs required a soil map regarding the susceptibility of the soil to erosion for catchment T35 D-E (87 000 hectares) in the Mzimvubu management area of the Eastern Cape, South Africa. Through a DSM approach, machine learning was used to predict the occurrence of soil forms and soil families in accordance with the South African Soil Classification System. A total of 591 soil observations were made at specific points, pre-determined with the conditioned Latin Hypercube sampling (cLHS) method. These pre-determined points were classified into nineteen soil forms and forty eight soil families. The soils were then further divided into nine soil groups, based on their inherent sensitivity to erosion. Outlier observations were omitted from the dataset to create a reliable framework to predict the soil groups as influenced by the SCORPAN factors. These soil groups were mapped and evaluated with an accuracy of 68% and a Kappa statistic value of 0.42 that compare well with other soil maps, both locally and internationally. Mapping soil with an acceptable accuracy with such a high level of detail on catchment scale would be a tremendous advantage to soil scientists and environmental workforces in southern Africa. Future work should focus on DSM training, to broaden the base of DSM skills available in southern Africa. This will ensure that soil information would be included in addressing an increasing amount of real world problems at catchment scales.
- Format
- I62 leaves
- Format
- Publisher
- University of Fort Hare
- Publisher
- Faculty of Science and Agriculture
- Language
- English
- Rights
- University of Fort Hare
- Hits: 2403
- Visitors: 2454
- Downloads: 77
Collections
UFH Department of Agronomy
| Thumbnail | File | Description | Size | Format | |||
|---|---|---|---|---|---|---|---|
| View Details | SOURCE1 | CJ du Plessis MSc dissertation.pdf | 7 MB | Adobe Acrobat PDF | View Details |