UFII Fellows Journal Club Seminar Series 

“Guiding Neural Networks with Bio-physical Models for Remotely-sensed Crop Progress Estimation”

by George Worrall
Ph.D. Candidate, Center for Remote Sensing, Department of Agricultural & Biological Engineering

Thursday, February 17, 2022 HYBRID: In-person with lunch at 11:30AM; live and zoom talk at noon

ABSTRACT:

With global crop synchrony increasing, accurate, in-season knowledge of crop progress is key to quantifying crop vulnerability and anticipating harvest shortfall. Satellite remote sensing data is often used to track crop progress, however collection of ground truth is labor-intensive. Moreover, in-season crop progress of spatially contiguous regions is correlated, so data must be separated by growing season. Data availability remains an issue. This limits the use of complex statistical approaches such as neural networks (NNs) that are commonly used in adjacent fields such as crop mapping, where millions of examples are available per year.

Recently, a new paradigm that called Physics-Guided Machine Learning (PGML), has shown that domain-specific physical models are able to provide guidance to NNs that reduce the required amount of training data required to reach a given level of performance. In this seminar, we present results from a recent study where we showed that providing NNs with outputs from simulations conducted using agronomic crop models significantly increases the accuracy of NN-based remotely-sensed crop progress estimation (CPE). Findings are presented for corn growth in Iowa and Illinois, which demonstrate that simulated values of water stress and crop growth based on observed weather data contain complementary information that increases the accuracy of in-season CPE, particularly during years of atypical crop progress.

Bio:

George Worrall is a PhD Candidate at the Center for Remote Sensing in the Agricultural and Biological Engineering Department. His research interests include remote sensing of crop growth, agricultural hydrology, crop modeling, and applications of neural network-based time series regression and prediction methods in agriculture.

Thursday, February 17, 2022
12:00PM – 1:00PM
Lunch will serve at 11:30AM
When you register below, please indicate if you will be attending in-person or via zoom. Zoom invite will be sent out the day before the talk.