The random distribution of sensors and the irregularityof routing paths lead to unordered sensory datawhich are difficult to deal with in Wireless Sensor Networks(WSNs). However, for simplicity, most existing researches ignorethose characteristics in the designs of Compressive Sensingbased Data Aggregation Schemes (CSDAS). Since conventionalsparsification bases (e.g., DCT, Wavelets) are inefficient to dealwith unordered data, performances of CSDAS with conventionalbases are inevitably constrained. In this work, a novel CSDASwhich adopts Treelet transform as a sparse transformation toolis proposed. Our CSDAS is capable to exploit both spatialrelevance and temporal smoothness of sensory data. Moreover,our CSDAS contains a novel correlation based clustering strategywhich is realized with the localized correlation structure ofsensory data returned by Treelets and facilitates energy saving ofCSDAS in WSNs. Comparative results show the reconstructionerror rate with adopting Treelet transform in CSDAS is about18% lower than that of conventional ones when the normalizedenergy consumption is 0.3. Even larger performance gain willbe obtained at higher energy consumption level. Meanwhile,simulations results further show that our novel correlation basedclustering strategy is of great potential. Specially, there is a gain ofroughly 35% for total energy savings with our proposed clusteringstrategy.