Human deaths due to natural disasters like earthquakes and tsunamis in recent times could have been avoided if seismological sensors were deployed at prone areas to help forecast the catastrophic events. With electronic sensors becoming very cheap it will become feasible to deploy large scale sensor networks by simply spreading thousands of disposable sensors on ocean-beds and other earthquake prone areas. However, sensor node populations must be replenished due to their limited battery lifetimes and natural damage. The decision to replenish is driven by the need to maintain a minimum population size for gathering sufficient sensor data for a reliable and timely forecast. It depends on an estimate of the alive sensor node population which is definitely hard to get in this scenario. We obtain a novel design of a first order, time-invariant Wiener filter that robustly estimates sensor node population with very minimal live packet feed from the active sensors. The sensor network is modeled with a G/GI/∞ queue. Our filter can reduce the estimation error by up to 42.7% as compared to an existing approach for a similar estimation problem.