Agent-based sensor location strategy for smart irrigation of large crop fields

Efficient monitoring of large crop fields is important to ensure the optimal use of resources such as water in irrigation policies, but at the same time represents a challenge to determine the structure of the sensor network. A balance must be accomplished between the acquisition, operation, and maintenance costs of this sensor network with the amount of information that can be collected in real-time to support the optimal use of the resource, e.g., an optimal irrigation policy. In this study, a sensor location strategy is proposed based on an agent-based model (ABM) of the crop-soil system, a state estimation algorithm reconstructing non-measured variables, and an objective function balancing the convergence of the estimation technique and the costs of the sensor network. The ABM model describes the crop-soil dynamics and allows conveniently representing uneven landscapes where water exchanges take place between different portions of the land. Various state estimation techniques can be considered and an extended Kalman filter is implemented in the present study, whose error covariance matrix can be exploited to assess practical observability and observer convergence. Finally, an economic cost function combines the observability measure with the sensor costs in order to select an optimal or suboptimal sensor array. For validation purposes, a numerical simulation case study, corresponding to a rugged land located in Colombia, is used to test various scenarios including the variability of climatic inputs.