Convolutional LSTM for Planetary Boundary Layer Height (PBLH) Prediction
Loading...
Files
Permanent Link
Author/Creator ORCID
Date
2021-03-22
Type of Work
Department
Program
Citation of Original Publication
Dorsa Ziaei, Jennifer Sleeman, Milton Halem, Vanessa Caicedo, Ruben Mann Delgado, and Belay Demoz,Convolutional LSTM for Planetary Boundary Layer Height (PBLH) Prediction, https://ebiquity.umbc.edu/paper/html/id/971/Convolutional-LSTM-for-Planetary-Boundary-Layer-Height-PBLH-Prediction
Rights
This item is likely protected under Title 17 of the U.S. Copyright Law. Unless on a Creative Commons license, for uses protected by Copyright Law, contact the copyright holder or the author.
Subjects
Abstract
We describe new work that uses deep learning to learn temporal changes in Planetary Boundary Layer Height (PBLH). This work is performed in conjunction with a deep edge detection method that identifies edges in imagery based on ceilometer backscatter signal from LIDAR observations. We implement a convolutional Long Short Term Memory (LSTM) to predict small temporal changes in PBLH estimates. In the presence of rain, clouds, and other unfavorable conditions, PBLH heights are challenging to estimate. The convolutional LSTM acts as an internal state representation of the external partially observable environment, supplementing the deep edge detection method, providing a prediction of PBLH in the absence of a reliable estimation. Convolutional LSTMs trained on image-based frames that define the movements of artifacts in the images, such as Moving MNIST digits, have been used to predict the movement of these artifacts for a set of frames in a sequence. We show how a similar network could be extended to learn more complex movement across frames and learn new information introduced at each frame. Utilizing the convolutional LSTM model with our proposed augmentation methodology applied to ten-minute frames, we predicted the change of the movement of edges identified as the PBL over time with favorable accuracy. We show the result of the prediction of PBL-based edges and evaluate the performance using three different metrics.