Upcoming Events
Notification of Dissertation Defense
Apr 11, 2023, 1:30 - 3:00 PM
Candidate: Haoteng Zhao
PhD of Science in Geographic and Cartographic Sciences
Department of Geography and Geoinformation Science
Date: Tuesday, April 11th, 2023
Time: 1:30 PM – 3:00 PM
Place: Commerce building, room 2006 or join via Zoom
TITLE: A data-driven intelligent decision-making model for irrigation scheduling
Committee:
Dissertation Director: Dr. Liping Di
Committee Members: Dr. Ruixin Yang, Dr. Daniel Tong, Dr. Taylor Anderson
ABSTRACT:
Agriculture dominates global water use. While irrigation improves crop production, it has become the largest consumption of freshwater. This means the already scarce freshwater resources due to urban expansion and climate change are facing even more pressure. On the other hand, excessive irrigation in the field would also increase agriculture run-off, pollutes surface and ground water, depletes water sources, and soil nutrition, and salinizes soils. Therefore, optimizing irrigation management to sustain crop yield while eliminating water wastes in irrigation is vital to agriculture sustainability, environmental quality, and the national economy. In this research, a data-driven intelligent irrigation scheduling model is proposed to improve water use efficiency. This study contributes to the field of water management in agriculture by demonstrating the validity of High-Resolution Land Data Assimilation System (HRLDAS) derived products in irrigation scheduling, and optimizing irrigation scheduling based on deep reinforcement learning. Soil moisture and evapotranspiration (ET) products simulated by HRLDAS are first evaluated and validated in Nebraska, the largest irrigation state in the U.S. ET- Water Balance (ET-WB) and soil-moisture based irrigation methods are used to schedule the irrigation events base on the validated model output in order to demonstrate its usefulness in irrigation management. The water-saving effect of integrating forecasted rainfall in irrigation scheduling was also analyzed. Based on deep reinforcement learning, an irrigation scheduling model was built and validated in Nebraska, to optimizing the irrigation. Yield estimations from crop growth model (AquaCrop) and soil moisture/ET simulations from HRLDAS are integrated in this model. Results show a total 20-40% water-saving was achieved, and highest economic return can be obtained compared to other methods. As all utilized data (e.g., simulated soil moisture and ET) were published and visualized on WaterSmart Data Information Portal (DIP), the application of this model would cost no fee on data service or installation of any sensor.