Groundwater contamination poses serious threats to public health and environmental sustainability. In this paper, we study the smart groundwater contamination sensing problem that aims at accurately estimating the nitrate concentration in groundwater via a crowdsensing approach. Existing solutions often require professional groundwater collection and high-quality measurement of groundwater properties, making the data collection process time-consuming and unscalable. In this work, we leverage the approximate nitrate concentration measured by crowd sensors (i.e., participants from well-dependent communities) to accurately estimate nitrate concentration in groundwater samples. Two critical challenges exist in developing the crowdsensing-based groundwater contamination estimation solution: i) the spatial irregularity of the crowdsensing groundwater contamination data, and ii) the hidden temporal dependency of groundwater contamination on the anthropogenic context. To address the above challenges, we develop SmartWaterSens, a context-aware graph neural network framework that explicitly models the irregular spatial relations of crowdsensing groundwater contamination data and its relevant anthropogenic context to accurately estimate groundwater nitrate concentration. We evaluate the SmartWaterSens framework through a crowdsensing nitrate contamination dataset collected from a real-world case study in well-dependent communities in Northern Indiana, United States. The evaluation results not only show the effectiveness of SmartWaterSens in accurately estimating nitrate concentration but also demonstrate the viability of crowdsensing for community-level groundwater quality monitoring.