Earth observation technologies for improved agricultural decision support systems in South Africa
| dc.contributor.author | Mpakairi, Kudzai Shaun | |
| dc.date.accessioned | 2025-11-26T09:45:08Z | |
| dc.date.available | 2025-11-26T09:45:08Z | |
| dc.date.issued | 2025 | |
| dc.description.abstract | This work investigates the application of remotely sensed data and advanced machine learning techniques in enhancing sustainable agricultural practices in South Africa, focusing on crop monitoring, water use efficiency, and land management. Firstly, a systematic review of remote sensing applications in Southern African agriculture, evaluating key advancements, challenges, and opportunities was conducted to document the key scientific knowledge gaps that then informed the focus of this study. The findings of the review revealed that the adoption of remotely sensed data and machine learning algorithms in agriculture remains in its infancy. Building on these insights, this study proposed a methodological framework for delineating irrigated and rainfed croplands in South Africa. By leveraging high-resolution Sentinel-2 satellite imagery and advanced machine learning techniques—including Deep Learning Neural Networks (DNN) and Random Forest (RF)—the study demonstrated the effectiveness of these models in generating accurate, large-scale agricultural land-use maps, achieving an overall classification accuracy of 0.71. Further, a novel approach for crop classification was also introduced by integrating unsupervised learning techniques and spectral matching algorithms, enabling accurate identification (OA = 0.84, p-value = 0.01) of major crop species across South Africa’s diverse agricultural landscapes. Additionally, the study employed multi-temporal MODIS satellite imagery to quantify annual crop water use (CWU) and crop water productivity (CWP), revealing substantial spatiotemporal variations between irrigated and rainfed croplands. Irrigated croplands generally had higher annual CWP (>0.002 kg/mm3/yr), while rainfed croplands consistently showed low CWP in forestry (0.001 kg/mm3/yr) and sugar (0.0012 kg/mm3/yr) agricultural regions. | |
| dc.identifier.citation | N/A | |
| dc.identifier.uri | https://hdl.handle.net/10566/21475 | |
| dc.language.iso | en | |
| dc.publisher | University of the Western Cape | |
| dc.relation.ispartofseries | N/A | |
| dc.subject | Climate change | |
| dc.subject | Cropland | |
| dc.subject | Crop-type mapping | |
| dc.subject | Crop water productivity | |
| dc.subject | Food insecurity | |
| dc.title | Earth observation technologies for improved agricultural decision support systems in South Africa | |
| dc.type | Thesis |