Is it possible to monitor the entire traffic in Manhattan at a few road intersections? In this paper, we propose a series of novel graph data processing techniques to handle complex and nonsmooth traffic data. Then, we validate our proposed techniques on Manhattan’s taxi pickups during the years of 2014 and 2015. We are able to approximately recover the taxi-pickup activities in Manhattan by taking samples at only 5 selected intersections. We believe that the same techniques can be applied to recover other types of traffic data. The advantages of our methods are (a) quality: we are able to recover the taxi-pickup activities in entire Manhattan with small error from only 5 selected intersections; (b) scalability: we use a tree structure and principle component analysis to make this method efficient for large- scale graphs.