This paper presents the first systematic design of a robust sensing system suited for the challenges presented by soil environments. We describe three soil deployments we have undertaken: in Bangladesh, and in California at the James Reserve and in the San Joaquin River basin. We discuss our experiences and lessons learned in deploying soil sensors. We present data from each deployment and evaluate our techniques for improving the information yield from these systems. Our most notable results include the following: in-situ calibration techniques to postpone labor-intensive and soil disruptive calibration events developed at the James Reserve; achieving a 91% network yield from a Mica2 wireless sensing system without end-to-end reliability in Bangladesh; and the javelin, a new platform that facilitates the deployment, replacement and in-situ calibration of soil sensors, deployed in the San Joaquin River basin. Our techniques to increase information yield have already led to scientifically promising results, including previously unexpected diurnal cycles in various soil chemistry parameters across several deployments.