Automatic Calibration of a Farm Irrigation Model: A Multi-Modal Optimization Approach.

In agriculture, plant cultivation requires to take numerous decisions. One of the major problems is irrigation: an adequate irrigation decision must be made accordingly to the hydric status of the plant and soil, and the weather forecasts. In precision agronomy, this leads to the use of hydric sensors combined with a numerical growth plant model. Such models can not often be tuned by experts. We proposed an automatic parameter calibration of the potato growth model based on data collected in several open fields. As these parameter calibration problems are ill-posed, the associated black-box optimization problem is supposed to be multi-modal. We then compare the performances of two state-of-the-art Evolution Strategies which use different restart mechanisms to automatically tune the set of parameters on different crops and shows that multi-modal optimization methods may be recommended for such class of optimization problems.