Deep learning models for brain tumour segmentation have shown remarkable performance in curated datasets from high-resource settings. However, these models often underperform when applied to magnetic resonance imaging (MRI) scans acquired in low- and middle-income regions due to domain shifts caused by differences in scanner hardware, acquisition protocols, and population-specific characteristics. In this paper, we investigate domain adaptation strategies–specifically transfer learning and the augmentation of training data using MRI characteristics from African data–to improve generalization to scans from Sub-Saharan Africa. Using the BraTS-Africa 2025 dataset as a benchmark, and the nnU-Net v2 and MedNeXt architectures, we develop and compare domain-specific augmentation strategies guided by global image statistics and local artifact patterns. We show that applying transfer learning and augmentation techniques to high-resource training data enables models to better generalize to African MRI datasets. Our results indicate that augmentation can reduce 95% Hausdorff distance (HD95) by up to 11 mm, while transfer learning improves both Dice scores (by up to 5%) and HD95 (by up to 13 mm). These improvements are particularly notable for tumour core and enhancing tumour subregions, which are the most sensitive to domain shift. These findings represent a step toward more equitable AI by addressing performance gaps in glioma segmentation for underrepresented populations. While our approach improves model performance in resource-constrained imaging settings, further validation across diverse African cohorts and clinical contexts is necessary to assess its broader generalizability and clinical utility.