Global mean sea level is rising at a rate unprecedented in the historical era as indicated by satellite altimetry retrievals since 1993. Recent studies have shown that the forced responses of greenhouse gas and aerosols have begun to emerge in the pattern of rise during the altimeter data and many features can be tied directly to these influences. Our work investigates future patterns of sea level change. To do this, we use machine learning (ML) applied to climate model outputs to understand the extent of causal contributions from different factors including greenhouse gases, industrial aerosols, and biomass burning aerosols. We apply deep learning-based spatio-temporal forecasting techniques to climate model simulations to learn a model of the underlying complex climate patterns responsible for the observed sea level rise. The ultimate goal is to exploit the learned patterns and the observed altimeter data to generate projections of sea level rise 30 years into the future. Specifically, we train a UNet model which is a fully convolutional neural network with an encoder and decoder architecture defining a U-shape. We add dilations in the convolution filters to increase the model’s receptive field and capture long-term spatial dependencies. This model takes in sea level maps at 2 degrees of spatial resolution and monthly temporal resolution and outputs a projection 30 years ahead at the same spatial resolution. These maps are converted from sea surface height to linear trend estimates (in mm/year) calculated over 30 years, to reduce monthly variability. Input trend maps are for years from 1930 to 2040; we keep 20 years for validation and 10 for the test set. A weighted mean squared error (MSE) is used as a loss function, where the weights are derived from the cosine of the latitude of all the spatial points. By training our ML model with the CESM1 and CESM2 large ensemble experiments, we show an RMSE of 0.11mm/year for the predicted sea level trend on the test data. We also demonstrate improvements against a persistence model as a baseline. The next steps are to include more climate models and to adapt the ML model that learned complex spatiotemporal features in climate model data to the observed altimetry data, in order to produce future projections.
CI
Week-ahead Solar Irradiance Forecasting with Deep Sequence Learning
Saumya Sinha, Bri-Mathias Hodge, and Claire Monteleoni
In Proceedings of the 11th International Conference on Climate Informatics (Environmental Data Science journal) 2022
@inproceedings{sinha2022weekahead,title={Week-ahead Solar Irradiance Forecasting with Deep Sequence Learning},author={Sinha, Saumya and Hodge, Bri-Mathias and Monteleoni, Claire},booktitle={Proceedings of the 11th International Conference on Climate Informatics (Environmental
Data Science journal)},year={2022},}
2021
Neurips CCAI
Subseasonal Solar Power Forecasting via Deep Sequence Learning
Saumya Sinha, Bri-Mathias S Hodge, and Claire Monteleoni
In NeurIPS 2021 Workshop on Tackling Climate Change with Machine Learning 2021
@inproceedings{sinha2021subseasonal,title={Subseasonal Solar Power Forecasting via Deep Sequence Learning},author={Sinha, Saumya and Hodge, Bri-Mathias S and Monteleoni, Claire},booktitle={NeurIPS 2021 Workshop on Tackling Climate Change with Machine Learning},url={https://www.climatechange.ai/papers/neurips2021/23},year={2021},}
CI
Variational Autoencoder Anomaly-Detection of Avalanche Deposits in Satellite SAR Imagery
Saumya Sinha, Sophie Giffard-Roisin, Fatima Karbou, and 5 more authors
In Proceedings of the 10th International Conference on Climate Informatics 2021
@inproceedings{10.1145/3429309.3429326,author={Sinha, Saumya and Giffard-Roisin, Sophie and Karbou, Fatima and Deschatres, Michael and Karas, Anna and Eckert, Nicolas and Col\'{e}ou, C\'{e}cile and Monteleoni, Claire},year={2021},isbn={9781450388481},publisher={Association for Computing Machinery},address={New York, NY, USA},url={https://doi.org/10.1145/3429309.3429326},doi={10.1145/3429309.3429326},booktitle={Proceedings of the 10th International Conference on Climate Informatics},pages={113–119},numpages={7},location={virtual, United Kingdom},series={CI2020},}
2020
EGU
Detecting avalanche debris from SAR imaging: a comparison of convolutional neural networks and variational autoencoders
Sophie Giffard-Roisin, Saumya Sinha, Fatima Karbou, and 5 more authors
In EGU General Assembly Conference Abstracts May 2020
@inproceedings{2020EGUGA..22.9487G,author={{Giffard-Roisin}, Sophie and {Sinha}, Saumya and {Karbou}, Fatima and {Deschatres}, Michael and {Karas}, Anna and {Eckert}, Nicolas and {Col{\'e}ou}, C{\'e}cile and {Monteleoni}, Claire},booktitle={EGU General Assembly Conference Abstracts},year={2020},series={EGU General Assembly Conference Abstracts},month=may,eid={9487},pages={9487},doi={10.5194/egusphere-egu2020-9487},adsurl={https://ui.adsabs.harvard.edu/abs/2020EGUGA..22.9487G},adsnote={Provided by the SAO/NASA Astrophysics Data System},}
2019
Neurips CCAI
Detecting Avalanche Deposits using Variational Autoencoder on Sentinel-1 Satellite Imagery
Saumya Sinha, Sophie Giffard-Roisin, Fatima Karbou, and 4 more authors
In NeurIPS 2019 Workshop : Tackling Climate Change with Machine Learning NeurIPS workshop Dec 2019
@inproceedings{sinha:hal-02318407,author={Sinha, Saumya and Giffard-Roisin, Sophie and Karbou, Fatima and Desch{\^a}tres, Micha{\"e}l and Karas, Anna and Eckert, Nicolas and Monteleoni, Claire},url={https://hal.archives-ouvertes.fr/hal-02318407},booktitle={{NeurIPS 2019 Workshop : Tackling Climate Change with Machine Learning NeurIPS workshop}},address={Vancouver, Canada},year={2019},month=dec,hal_id={hal-02318407},hal_version={v1},}
CI
Can Avalanche Deposits be Effectively Detected by Deep Learning on Sentinel-1 Satellite SAR Images?
Saumya Sinha, Sophie Giffard-Roisin, Fatima Karbou, and 5 more authors
@inproceedings{sinha:hal-02278230,title={{Can Avalanche Deposits be Effectively Detected by Deep Learning on Sentinel-1 Satellite SAR Images?}},author={Sinha, Saumya and Giffard-Roisin, Sophie and Karbou, Fatima and Desch{\^a}tres, Micha{\"e}l and Karas, Anna and Eckert, Nicolas and Col{\'e}ou, C{\'e}cile and Monteleoni, Claire},url={https://hal.archives-ouvertes.fr/hal-02278230},booktitle={{Climate Informatics}},address={Paris, France},year={2019},month=oct,hal_id={hal-02278230},hal_version={v1},}