Variational Quantum Algorithms (VQAs) are a promising approach to achieve near-term quantum advantage. VQAs provide a hybrid classical and quantum computing approach to solving problems by employing a classical optimizer to train a Parameterized Quantum Circuit (PQC). However, VQAs still face several challenges, including their trainability, efficiency, and accuracy. The trainability, for example, is highly dependent on the structure of the associated cost landscape, which is induced by the cost function of a problem. The optimizer navigates the cost landscape to find local and global optima. A major challenge during optimization is regions in the cost landscape where gradients exponentially vanish. These so-called barren plateaus can be caused by many aspects of a VQA, such as a high degree of entanglement in the training data. To detect such challenging regions, it can be helpful to characterize regions within the cost landscape using roughness metrics such as curvature. This work introduces and implements Total Absolute Scalar Curvature (TASC) for VQA cost functions, which describes the curvature of a region within the cost landscape instead of just at a singular point. We then perform experiments for several different cost landscapes, defined by different training data structures, and evaluate the effectiveness of using TASC to characterize them.