Srishti Gaur, a postdoctoral scholar in The Ohio State University's Department of Food, Agricultural and Biological Engineering, recently published an article in Agricultural and Forest Meteorology, a peer-reviewed journal covering research on the relationships between meteorology and the fields of plant, animal, and soil sciences, ecology, and biogeochemistry.
Gaur's article, titled "Explainable machine learning for predicting stomatal conductance across multiple plant functional types" will appear in the May 2024 edition of the journal.
Her research focuses on stomal conductance, which is a central plant function that regulates carbon dioxide uptake (photosynthesis) and water use, and the exchange of energy between plants and the atmosphere. This process is a critical component of modern land surface models, including models used to predict the weather, crop productivity and climate change impacts.
Despite the importance of stomatal conductance, the current approaches to simulate it rely on models that are difficult to use. In Gaur's paper, titled “Explainable Machine Learning for Predicting Stomatal Conductance Across Multiple Plant Functional Types,” she demonstrates the power of machine learning to predict stomatal conductance for a wide range of plant types. Importantly, the models she developed rely primarily on standard climate and remote sensing variables, making them widely applicable. A key aspect of this work was the use of explainable machine learning in interpreting the outcomes of machine learning models, allowing researchers to gain an understanding for what are often thought of as “black-box” models.
"This paper is very special for me as this is the first one from my postdoctoral research. My Ph.D. work focused on physical hydrology, and it is exciting to broaden my research portfolio into the area of ecohydrology with this paper," said Gaur. "I would like to thank my postdoctoral advisor Dr. Darren Drewry for his supervision and dedication to our research and my development as a scientist."
Going forward, the work in her paper has laid the foundation for building the next generation of ecosystem models, so-called “hybrid” models that integrate machine learning and biophysical principles to improve our ability to predict and understand terrestrial ecosystems. Gaur is now training and evaluating a new suite of hybrid ecosystem models using a gas exchange dataset collected for eight soybean genotypes and three water stress treatments under field conditions outside of Ames, Iowa during the summer of 2023.
Read the article online