The authors consider a Wireless Sensor Network (WSN) deployed over a large geographical area, where a querying node wishes to perform a estimation of a localized phenomenon. The authors formulate the problem as a joint optimization of sensor selection and routing structure where they minimize the estimation distortion subject to a total communication power constraint for the WSN. Two scenarios are analyzed: measurement forwarding and estimation-and-forward at the nodes. The authors show that the optimization problems corresponding to these scenarios are both NP-hard and they propose two approximation algorithms.