Learning to Irrigate - A model of the plant water balance
Matthias Maszuhn, Frerk Müller-von Aschwege, Susanne Boll-Westermann, Jan Pinski
Nowadays, a wide range of soil sensor systems is cheaply available for tree nurseries and thus allows them to monitor the condition of their plants. An automated, sensor-based irrigation and fertilization system promises many advantages over the manual irrigation such as a better plant growth rate and a lower water consumption. This paper proposes a workflow to create a plant water balance model based on recorded soil sensor data, weather data and weather forecasts and to bring it in relation with the plant’s growth model. The model was implemented and used to train a reinforcement learning algorithm to control the irrigationso that the soil humidity would stay within appropriate thresholds. After the training, we were able to keep a stable soil humidity for 19 of the 23 tested plants and react to changing weather conditions and soil temperatures.
Februar / 2023
misc
PPP
Predictive Plant Production
Predictive Plant Production
