Owing to the rising awareness of environmental protection and health, people put a high premium on air pollution in their living environment. It thus draws considerable attention to air quality monitoring in cities. The paper suggests using a vehicular sensor network (VSN) to tactically monitor metropolitan air quality and develops an efficient data gathering and estimation (EDGE) mechanism on VSN. EDGE has an objective to adaptively change data sampling rates of cars, such that the tradeoff between monitoring accuracy and communication cost is balanced. The monitoring accuracy is measured by the formal air quality index (AQI), whereas the communication cost considers the amount of sampling data and the monetary reward given to drivers. To do so, EDGE proposes dynamic grid partition based on the variation of pollutant concentration and computes the sampling rate by consulting car traffic in each grid. With the help of probabilistic reporting, it allows cars to collect air quality on more different positions and alleviate potential network congestion. Furthermore, EDGE applies the Delaunay triangulation to infer AQIs of the positions without any sensing data. Through simulation of urban mobility and industrial source complex, simulations are conducted based on practical metropolitan traffic and pollutant dispersion models. Experimental results demonstrate the significant effectiveness of the EDGE mechanism, under various scenarios.