Machine learning for improved irrigation representation and understanding within Earth system models
Project summaryIrrigation is a critical component of land and water resource management, leading to anthropogenic changes in the water cycle and energy balance that affect the Earth system at multiple scales . The understanding of irrigation impacts is mainly implemented through the use of process-based models due to the lack of high-resolution spatiotemporal datasets for irrigation parameters. However, the climate feedback from irrigation remains not fully understood and there is often disagreement about its effects. A major source of uncertainty lies in the static or oversimplified irrigation parameterizations in Earth system models, which often fail to consider the complex and adaptive nature of human decision-making in irrigation practices . On the other hand, machine learning is able to capture the complex interactions between crop responses and their interplay with environmental conditions. For example, its branch, reinforcement learning, is increasingly used to provide dynamic, optimal irrigation strategies tailored to different scenarios.
In this project, we aim to bridge the gap between realistic human irrigation practices and their representation in Earth system models. By harnessing the power of machine learning, our goal is to develop a more realistic irrigation parameterization in the Earth system model. The machine learning model will serve as a dynamic replacement or augmentation to the conventional irrigation representation to incorporate a multitude of environmental factors that may influence irrigation practices. Key methodological challenges include overcoming the limitations of detailed data on irrigation practices and reflecting bounded rationality in human decision-making. Rules or constraints from domain expertise can be encoded into the machine learning model to guide the learning process in sparse data. If this new parameterization strategy can better simulate and replicate realistic practices that humans may use in specific environments, the resulting improved Earth system model will be used to investigate the hydroclimatic effects of varying irrigation practices. The prospective PhD student is encouraged to explore their own ideas within the hybrid modeling framework  for understanding the interactions between sustainable irrigation and climate feedback.
Working groupThe successful PhD candidate will work in the Department of Biogeochemical Integration at the Max Planck Institute for Biogeochemistry. The working group offers long-standing expertise and experience in the various fields relevant to this project: Machine Learning (Shijie Jiang, Nuno Carvalhais, Markus Reichstein), Irrigation and Hydrology (Anke Hildebrandt, Shijie Jiang), Earth System Models (Nuno Carvalhais, Markus Reichstein, Shijie Jiang), Agroecosystems (Markus Reichstein, Anke Hildebrandt). The PhD candidate will benefit from the European Lab for Learning and Intelligent Systems (ELLIS) and will work closely with the ELLIS Unit Jena. For further information, please contact Shijie Jiang [firstname.lastname@example.org].
RequirementsWe welcome applications from highly motivated and curious students from any country who have
References Yang, Y. et al. Sustainable irrigation and climate feedbacks. Nature Food 4, 654–663 (2023).
 Pokhrel, Y.N. et al. Recent progresses in incorporating human land–water management into global land surface models toward their integration into Earth system models. WIREs Water 3, 548-574 (2016).
 Reichstein, M. et al. Deep learning and process understanding for data-driven Earth system science. Nature 566, 195–204 (2019).
>> more information about the IMPRS-gBGC + application