Soil moisture is one of the most important parameters in precision agriculture. While techniques such as remote sensing seems appropriate for moisture monitoring over large areas, they generally do not offer sufficiently fine resolution for precision work, and there are time restrictions on when the data is available. Moreover, while it is possible to get high resolution-on demand data, but the costs are often prohibitive for most developing countries.

Direct ground level measurement can be a viable and economical alternative if one is able to accurately estimate the value of soil moisture over the entire field by using measurement from only a few points. If the number of measurement points, their location, and data are available, then Compressive Sensing (CS) theory may be used to give an estimate of the moisture. This is because although moisture values in a field do not constitute a sparse signal, they are spatially correlated and can be expressed as sparse signals in other domains such as DCT or DFT.

The difficulty in using the CS theory for estimation of moisture values is that the number and location of the sensors must be known a priori.  In reality, this means the optimization problem has to be solved several times for various different network configurations to determine the best layout. Straightforward augmentation of the CS reconstruction optimization problem to include the configuration selection leads to so called MINLP optimization type of problems which are combinatorial and non-polynomial time. Such problems take exceptionally large times to solve for large scale problems (as encountered in PA type of applications).

 In this paper, we propose a new heuristic algorithm to find a sub-optimal set for sensor locations. A data set for numerical experiments is extracted from the simulation of a simple field using the state-of-the-art TIN-based Real-time Integrate Basin Simulator (tRIBS). This data set is used for validation of the optimization results