Application of indigenous knowledge, machine learning and satellite imagery to optimize cropping decisions by small-scale farmers: case study of uMgungundlovu District Municipality, South Africa

Abstract

This thesis presents the methodology to optimise cropping decisions for small-scale farmers who depend primarily on rainfed agriculture for subsistence and market production. The research was motivated by the significant role small-scale farmers, particularly in sub-Saharan Africa (SSA), play in ensuring food security, despite often being overlooked by many technological developments and advancements. Although the use of chemicals to grow crops seems to expand in Africa, many smallholder farmers continue to use the Indigenous Knowledge (IK) system that by its originality is natural, organic and environmentally friendly, and most importantly food produced by using only IK methods is of high quality. Although IK is a crucial resource for small-scale farmers in food production and managing climate-related challenges, its value, accuracy, and presence have diminished over time due to climate change, market competition, deforestation, pollution, modernisation, and many more. This thesis aims to restore and enhance the dignity and effectiveness of indigenous knowledge by integrating it with scientific data and methods to optimise cropping decisions for small-scale farmers. This integration seeks to improve farmers’ traditional methods to predict rainfall patterns and temperature fluctuations throughout the agricultural season. This integration is further expanded to strengthen farmers’ conventional methods of monitoring the health and growth of the crops. While indigenous and scientific knowledge systems have their strengths and limitations, a significant body of literature highlights their collaboration’s benefits in enhancing the resilience of local communities to climate change. However, not much has been done to incorporate these knowledge domains together to optimise food production given that one is qualitative and locally based while the other is quantitative and generic. In this research project, efforts are made to quantify the IK system farmers use to predict rainy season behaviour and monitor the health and growth of crops, then integrate it with scientific data and approaches (satellite imagery, climate data, and machine learning models) to come up with more robust and local-based information that farmers can use to improve their cropping decisions. These knowledge systems are integrated into a mobile-based technology designed for farmers’ use. This hybrid technology is trained to recommend crops a farmer can grow based on anticipated seasonal behaviour. It is then evaluated using a dataset with unknown inputs, achieving an accuracy level of 83.3%.