Use of geospatial techniques to improve bee farming and bee health across four main agroecological zones in Kenya.
dc.contributor.advisor | Mutanga, Onisimo. | |
dc.contributor.advisor | Odindi, John Odhiambo. | |
dc.contributor.advisor | Abdel-Rahman, Elfatih Mohamed. | |
dc.contributor.author | Makori, David Masereti. | |
dc.date.accessioned | 2023-06-30T13:09:15Z | |
dc.date.available | 2023-06-30T13:09:15Z | |
dc.date.created | 2023 | |
dc.date.issued | 2023 | |
dc.description | Doctoral Degree. University of KwaZulu-Natal, Pietermaritzburg. | en_US |
dc.description.abstract | Amid augmented climate change and anthropogenic influence on natural environments and agricultural systems, the global socioeconomic and environmental value of bees is undisputed. Bee products such as honey, pollen, nectar, royal jelly and to a lesser extent bee venom are important supplemental sources of income generation especially in the underdeveloped rural African areas. Moreover, bee farming is an important incentive for forest conservation, biodiversity and ecosystem services in terms of pollination services. Bee pollination services play a vital role in crop production, hence directly contribute to food and nutritional security for African smallholder farmers. Nevertheless, bee farming and bee health in general are under threat from climate change, agricultural intensification and associated habitat alteration, agrochemicals intensification, bee pests and diseases. Therefore, there is need to establish spatial distribution of bees, their food substrates, floral cycle and biotic and abiotic threats, especially bee pests. Bee pests devastate bee colonies through physical injury and as vectors of pathogens, hence causing a considerable reduction in bee colony productivity. Thus, this study sought to establish geospatial techniques that could be used to improve bee farming and bee health in Kenya. Firstly, this study aimed to determine the spatial and temporal distribution of stingless bees in Kenya using six machine learning ecological niche approaches and non-conflating variables from both bioclimatic, vegetation phenology and topographic features. All machine learning algorithms used herein performed at an ‘excellent’ level with a true skills statistics (TSS) score of up to 0.91. Secondly, the study assessed the suitability of resampled multispectral data for mapping melliferous (flowering plants that produce substance used by bees to produce honey) plants in Kenya. Bi-temporal AISA Eagle hyperspectral images, resampled to four sensors’ (i.e., WorldView-2, RapidEye, Spot- 6 and Sentinel-2) spatial and spectral resolutions, and a RF classifier were used to map melliferous plants. Melliferous plants were successfully mapped with up to 93.33% overall accuracy using WorldView-2. Furthermore, the study predicted the distribution of four main bee pests (Aethina tumida, Galleria mellonella, Oplostomus haroldi and Varroa destructor) in Kenya using the maximum entropy (MaxEnt) model and random forest (RF) classifier. The effect of seasonality on the abundance of bee pests was apparent, as indicated by the Wilcoxon rank sum test, with up to 6.35 times more pests in the wet than the dry season. Furthermore, bioclimatic variables especially precipitation contributed the most (up to 77.8%) to all bee pest predictions, while vegetation phenology provided vital information needed to sharpen the prediction models at grain level due to their higher spatial resolution and seasonal and phenological features. Moreover, topography had a moderate influence (14.3%) on the distribution of bee pests. Also, there was a positive correlation between bee pests’ abundance, habitat suitability and high altitude. Anthropogenic influence (as depicted by human footprint data) on the distribution of bee pests was relatively low (1.2%) due to the availability of a variety of bee food substrate from the mixed land use/land cover (LULC) classes, especially farmlands. Using the Pearson correlation coefficient, the prediction models for all bee pests scored at an excellent level (0.84), except for the G. mellonella prediction model, which was ranked ‘fair’ (0.55). Due to the relatively high accuracy for models developed herein to map stingless bees’ distribution, melliferous plants and bee pests’ occurrence and abundance, this study concluded that the models developed could reliably be used to indicate high suitability areas for bee farming. They could also be used to predict high bee pests risk areas for mitigation and management purposes, hence improving bee health and hive productivity. | en_US |
dc.identifier.uri | https://researchspace.ukzn.ac.za/handle/10413/21695 | |
dc.language.iso | en | en_US |
dc.subject.other | Stingless bees. | en_US |
dc.subject.other | Honeybee pests. | en_US |
dc.subject.other | Climate change. | en_US |
dc.subject.other | Machine learning. | en_US |
dc.subject.other | Remote sensing. | en_US |
dc.title | Use of geospatial techniques to improve bee farming and bee health across four main agroecological zones in Kenya. | en_US |
dc.type | Thesis | en_US |
Files
Original bundle
1 - 1 of 1
Loading...
- Name:
- Makori_David_Masereti_2023.pdf
- Size:
- 9.43 MB
- Format:
- Adobe Portable Document Format
- Description: