Graph Neural Networks for Predicting Wastewater Service Type At A Land Parcel Level

America lacks a unified, parcel-scale map of wastewater service-type infrastructure; the last nationally comprehensive collection was in 1990, and while newer sources exist, they are decentralized and often non-spatial, complicating integration. In this paper, we present an empirical study evaluating Graph Neural Networks (GNNs) for node classification of buried sanitation infrastructure mapping. Using one million parcel-level sanitation records from Florida, we construct sampled graphs of 1k, 5k, and 10k nodes, where each land parcel is modeled as a graph node with spatial and functional edges. We evaluate three classical GNNs in both transductive and inductive settings. The findings demonstrate that graph structure captures meaningful spatial dependencies in wastewater infrastructure.