Date of this Version
Wilkening, E. J., Heeren, D. M., Shi, Y., Katimbo, A., Amori, P. N., Balboa, G. R., Puntel, L. A., Zhang, K., & Rudnick, D. R. July 9-12, 2023. Development of a scalable edge-cloud computing based variable rate irrigation scheduling framework. ASABE Annual International Meeting, Omaha, Nebr. Poster presentation.
Currently, variable-rate precision irrigation (VRI) scheduling methods require large amounts of data and processing time to accurately determine crop water demands and spatially process those demands into an irrigation prescription. Unfortunately, irrigated crops continue to develop additional water stress when the previously collected data is being processed. Machine learning is a helpful tool, but handling and transmitting large datasets can be problematic; more rural areas may not have access to necessary wireless data transmission infrastructure to support cloud interaction. The introduction of “edge-cloud” processing to agricultural applications has shown to be effective at increasing data processing speed and reducing the amount of data transmission to remote processing computers or base stations. In irrigation in particular, edge-cloud computing has so far had limited implementation. Therefore, an initial logic flow concept has been developed to effectively implement this new processing technique for VRI. Utilizing edge-cloud computer nodes in the field, autonomous data collection devices such as center pivot-mounted infrared canopy thermometers, soil moisture sensors, local weather stations, and UAVs could transmit highly localized crop data to the edge-cloud computer for processing. The edge computer Following the implementation of an irrigation strategy created by the edge-cloud computer with a machine learning model, data would be transmitted to the cloud (requiring transmission of only minimal model parameters), resulting in a feedback loop for continual improvement of the global model on the cloud (federated learning). VRI prescription maps from the SETMI model were used as the training data for training the machine learning model.