Many early sensor networks in environmental science application domains have been limited to test beds and have focused on hardening the sensor, networking, and power management strategies needed to survive harsh conditions and remote locations. If many cases, applications researchers have necessarily let the technology guide the deployment, with scientific payoff being a secondary concern. This presentation will focus on the budding reversal of this undesirable paradigm. The use of environmental process models to develop optimal sampling strategies in the context of two examples, one in a river domain and the other in soil-plant systems. In the first case, simple environmental models will be used to illustrate how a complex deployment might be orchestrated for maximum scientific payoff. In the second, real-time model calibration using sensor feedback is used to control irrigation behavior in an agricultural system.