The two-layer network structure has been widely adopted in wireless sensor networks (WSNs) for managing sensor nodes. In such a structure, the low layer nodes communicate with their cluster head, followed by the cluster-head nodes communicating with the base station operating in either a one hop or a multi-hop manner. The main focus of node-clustering algorithms is minimizing energy consumption due to strictly limited resources in WSNs. Also, WSNs are data intensive networks with the capability of providing users with accurate data. Unfortunately, data missing is common in WSNs. In this paper, we propose a novel joint design of sensor nodes clustering and data recovery, where the WSNs is organized in a two-layer manner with our developed clustering algorithm, and then the missing data is recovered based on this two-layer structure. Furthermore, in the proposed clustering algorithm, we take both the energy-efficiency and data forecasting accuracy into consideration and investigate the tradeoff between them. This is based on the key observation that the high energy-efficiency of the network can be achieved by reducing the distances among the nodes in a cluster, while the accuracy of the forecasting results can be improved by increasing the correlation of the data stream among the nodes in a cluster. Simulation results demonstrate that our joint design outperforms the existing algorithms in terms of energy consumption and forecasting accuracy.