FogNet

A multiscale 3D CNN with double-branch dense block and attention mechanism for fog prediction.

• Input: raster of atmospheric variables
• Output: binary classification probability

This webpage is a portal the FogNet source code, datasets, and published material.

Publications

Kamangir, H., Collins, W., Tissot, P., King, S. A., Dinh, H. T. H., Durham, N., & Rizzo, J. (2021).
FogNet: A multiscale 3D CNN with double-branch dense block and attention mechanism for fog prediction.
Machine Learning with Applications, 5, 100038.
Link (ScienceDirect)

Kamangir, H., Krell, E., Collins, W., King, S. A., Tissot, P. (2022).
Importance of 3D convolution and physics on a deep learning coastal fog model
Environmental Modelling & Software, 105424.
Link (ScienceDirect)

Krell, E., Kamangir, H., Collins, W., King, S. A., & Tissot, P. (2023).
Aggregation Strategies to Improve XAI for Geoscience Models That Use Correlated, High-Dimensional Rasters
Environmental Data Science.
Manuscript accepted, available soon

Presentations

Kamangir, H., Tissot, P., Collins, W., King, S. A., Dinh, H. T. H., Durham, N., & Rizzo, J. (2021, January).
FogNet: A Multiscale 3D CNN with an Attention Mechanism and a Dense Block for Fog Predictions.
In 101st American Meteorological Society Annual Meeting. AMS.
Link (AMS 101)

Explaining Complex 3D Atmospheric CNNs Using SHAP-Based Channel-Wise XAI Techniques with Interactive 3D Visualization
Trustworthy Artificial Intelligence for Environmental Science Summer School 2021

Kamangir, H., Krell, E., Collins, W. G., King, S. A., & Tissot, P. E. (2022, January).
Importance of 3D Convolution-and Physics-Based Modeling of Atmospheric Predictions: Fog Forecasting Case Study.
In 102nd American Meteorological Society Annual Meeting. AMS.
Link (AMS 102)

Krell, E., Kamangir, H., Friesen, J., Judge, J., Collins, W. G., King, S. A., & Tissot, P. E. (2022, January).
Explaining Complex 3D Atmospheric CNNs Using SHAP-Based Channel-wise XAI Techniques with Interactive 3D Visualization.
In 102nd American Meteorological Society Annual Meeting. AMS.
Link (ResearchGate)

The Influence of Feature Aggregation for Explainable AI for High Dimensional Geoscience Applications
DoD Cloud Post-Processing and Verification Workshop

Code

• FogNet: architecture source code
• FogHat: utilities to download, prepare NAM, MUR data for FogNet
• fog_benchmark_models: 2D CNNs for comparison with FogNet
• xai-raster-tools: Scripts for visualizing 3D raster XAI outputs

DataShare

Link to DataShare

The following link contains a variety of datasets related to FogNet. Because of the large size of the input data, it is not included with the GitHub repo and must be downloaded separately. The steps to do so are documented in the source code README.

• figures/ Various graphics describing the FogNet architecture
• archive/ FogNet inputs (predictors) and archived outputs (classifications). These correspond to publications to allow reproducing results & enabling further research.
• datasets/ Each subfolder is a set of FogNet predictors & targets for a task (i.e. 24-hour lead time)
• pretrained/ Trained model weights for each dataset
• results/ Each subfolder has the output classification, analysis results for each dataset
• 2D_benchmarks/
• fog_benchmark_models_outputs-02132022.tar.gz Archived model weights, test results for FogNet ablation study (under review)
• xai/ Outputs generated during explainable AI (XAI) analysis
• FogNetXAI-02152022.tar.gz XAI outputs & scripts for AMS 2022

Acknowledgements

This material is based upon work supported by the National Science Foundation under award 2019758.

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

License

The data & source code are licensed under a CC0 1.0 Universal license.