Large Low Velocity Provinces (LLVPs) are hypothesised to be purely thermal features or possess some chemical heterogeneity, but exactly which remains ambiguous. Regional seismology studies typically use travel time residuals and multipathing identification in the waveforms to infer properties of LLVPs. These studies have not fully analysed all available information such as measuring the direction and inclination of the arrivals. These measurements would provide more constraints of LLVP properties such as the boundary velocity gradient and help determine their nature. Here, we use array seismology to measure backazimuth (direction) and horizontal slowness (inclination) of arriving waves to identify structures causing multipathing and wavefield perturbation. Following this, we use full-wavefield forward modelling to estimate the gradients required to produce the observed multipathing. We use SKS and SKKS data from 83 events sampling the African LLVP, which has been extensively studied providing a good comparison to our observations. We find evidence for structures at heights of up to 600 km above the core-mantle boundary causing multipathing and wavefield perturbation. Forward modelling shows gradients of up to 0.7% δVs per 100 km (0.0005 km /s km) are required to produce multipathing with similar backazimuth and horizontal slowness to our observations. This is an order of magnitude lower than the previous strongest estimates of -3% δV per 50 km (0.0044 km /s km). As this is lower than found for both thermal and thermochemical structures, gradients capable of producing multipathing is not necessarily evidence for a thermochemical nature.