Skip to main content Skip to search

YU News

YU News

Dr. Wonjun Lee is Co-recipient of a Major Grant for Soil Moisture Forecasting

Dr. Wonjun Lee Dr. Wonjun Lee

Dr. Wonjun Lee, assistant professor of artificial intelligence at the Katz School for Science and Health, Dr. Sanjiv Kumar (assistant professor in the School of Forestry and Wildlife Sciences at Auburn University) and Dr. Imtiaz Rangwala (Cooperative Institute for Research in Environmental Sciences at the University of Colorado Boulder) received $499,251 from the National Institute of Food and Agriculture, United States Department of Agriculture, for their three-year project, “Interactive Deep Learning Platform and Multi-source Data Integration for Improved Soil Moisture Forecasting.” The grant began on Sept. 1, 2020, and runs until August 31, 2023, and the two researchers will be co-Primary Investigators.

YU News spoke with Dr. Lee about this unique mix of artificial intelligence and agriculture.

Dr. Lee, thank you for taking the time to talk to YU news and congratulations on the grant. Could you explain the problem that your research looks to solve?

The usefulness of a seasonal climate forecast for agricultural planning is limited by two primary factors: 1) large uncertainties in the forecast and 2) the trustworthiness or reliability of the forecast. These uncertainties affect an agriculturalist’s ability to incorporate soil moisture forecasting into making reasonable decisions about inputs and outcomes

What is the solution?

We want to improve the value of the forecast by incorporating both user-provided and Internet of Things (IoT)–based inputs (such as from sensors placed on the land) to calibrate the forecast at the user’s location in real time. To do that, Dr. Weyhenmeyer and I propose to use deep learning (DL) models trained by data sets drawn from the Signals in the Soil network and apply the trained model to forecast soil moisture at the user’s location. We’ll do this by combining the latest advances in earth system observations and DL to create an interactive deep learning IoT platform for next-generation soil moisture forecasting that is smart, agile and adaptable to meet the stakeholders’ needs in a highly accessible manner to users.

Wouldn’t such a platform require a lot of computational power and data storage?

Yes, it would, and to address these requirements, we want to use something called Container Based-Software Defined Storage (CB-SDS), an alternative data storage process that is more flexible than traditional storage devices. CB-SDS will provide data storage to end-users and end-point applications in a dynamic manner while concealing the complexity of resource management.

How does the analysis platform use user data in real time?

Data transmission and analysis and cyber-infrastructure are crucial, and our research will develop novel solutions that utilize DL and IoT to address challenges in dealing with big data sets in soil moisture forecasting. Data from IoT devices is incorporated, and crowdsourcing can positively impact the forecast skill and physics of soil moisture dynamics at multiple scales by reducing the time delay between data collection and its incorporation in the forecast.

What excites you most about this research?

Collaborators from two different fields—artificial intelligence and earth system modeling—will work together to use intelligent analytics methods supported by highly scalable cloud technology and the latest advances in hydrological modeling to minimize agricultural loss due to drought, and we’ll do this by providing accurate and reliable forecast information at the user’s location and on the user’s seasonal time scales. The system we propose will allow users to upload inputs (e.g., location and field-scale soil moisture observations) through IoT devices connected to an app, after which the underlying DL analytics platform will do the rest, including retrieving the forecast and calibrating the product to the user’s location. This real-time soil moisture forecast driven by IoT technology will provide critical information for decision-making as well as improve our understanding of soil moisture.

What are some of the broader implications of your research?

Real-time IoT platform and complemented user will work together to improve the accuracy of soil moisture predictions in broad and impactful areas such as drought planning and soil health. Bringing together the best features of big data and deep learning with earth system modeling can deliver resources to help communities predict and plan for upcoming droughts. Developing an IoT platform with DL and CB-SDS technology will help transfer hydrological understanding to the DL algorithms to automate the calibration and feedback process, thereby improving the understanding of soil moisture processes.

As the number of IoT sensors and devices adopted by users increases, our automated process will provide smarter and more accurate results, more broadly impacting communities around the globe as well as the hydrological science community.

How will you broadcast your research results to the world at large?

We are going to create two “data science summer schools” that will train high school students in cutting-edge data science technology. This research will also create direct impacts through support, training, and mentoring of three graduate students, three undergraduate students, and one data scientist during each year of the project.