About this Abstract |
Meeting |
MS&T24: Materials Science & Technology
|
Symposium
|
Machine Learning and Simulations
|
Presentation Title |
Forecasting Nutrient Flows Using
Terrain Elevation-Aware Spatial-Temporal Graph Neural Networks
|
Author(s) |
Jiaqi Li, Alexander H. Bradley, Olatunde Akanbi, Laura Bruckman, Erika Barcelos, Roger French, Yinghui Wu |
On-Site Speaker (Planned) |
Jiaqi Li |
Abstract Scope |
We propose a class of Terrain Elevation-aware Spatio-Temporal Graph Neural Networks (ST-EGNN) that incorporates terrain elevation data to its edge model to improve the accuracy of time-series forecasting based agricultural nutrient flow analysis. Results were compared against three baseline models: (a) a spatial-temporal Graph Neural Network that only use edges induced from 2D Euclidean distances, and two deep neural networks: (b) sequence to sequence (seq2seq), which is a class of deep learning architectures renowned for their accomplishments in various domain, and (c) LSTM is a type of recurrent neural network architecture designed to overcome the vanishing gradient problem in traditional RNNs, respectively. In addition, the framework leverages data imputation from multiple sensors concurrently to improve the robustness of STEGNN against the impact of missing data. By learning spatial, temporal and elevation patterns, our approach endeavors to offer a more adaptable predictive framework for nutrition analysis in water and river systems. |