The propagation of coherent light in multimode optical fibers results in a speckled output that is both complex and sensitive to environmental effects. These properties can be a powerful tool for sensing, as small perturbations lead to significant changes in the output of the fiber. However, the mechanism to encode spatially resolved sensing information into the speckle pattern and the ability to extract this information are thus far unclear. In this paper, we demonstrate that spatially dependent mode coupling is crucial to achieving spatially resolved measurements. We leverage machine learning to quantitatively extract the spatially resolved sensing information from three fiber types with dramatically different characteristics and demonstrate that the fiber with the highest degree of spatially dependent mode coupling provides the greatest accuracy.