Poster Gallery for Session 1 (February 10th): Earth Science

Moderator: Hui Su, JPL

đź“‘ Combined PDF of posters for this session

Poster 1 Thumbnail

Multispectral Analysis of Land Surface Reflectance Time-Series for Clustering Change Events

Authors
Srija Chakraborty
Theme
Earth Science
Poster ID
1

Abstract: (Not available.)

Poster 2 Thumbnail

Prediction of Global Geomagnetic Field Disturbances using Recurrent Neural Network

Authors
Hyunju Connor, Shishir Pryadarshi, Mathew Bladin, Amy Keesee
Theme
Earth Science
Poster ID
2

Abstract: (Not available.)

Poster 3 Thumbnail

The Annotation Game: Towards Collaborative Science with Humans, Robots, and AI

Authors
Zhiang Chen, Tyler R Scott, Ethan Duncan, Harish Anand, ALG Prasad, Sarah Bearman, Devin Keating, Chelsea Scott, Brenth Hayashi, Mark Wronkiezicz, Jnaneshwar Das, Ramon Arrowsmith
Theme
Earth Science
Poster ID
3

Abstract: (Not available.)

Poster 4 Thumbnail

Automatic Per-Pixel Classification of UAVSAR Imagery for Hurricane Flood Detection

Authors
Michael Denmina, Zaid J Towfic, Matthew Thill, Brian Bue, Yunling Lou
Theme
Earth Science
Poster ID
4

Abstract: The main objective of this project was to assess the viability of convolutional neural network-based image classifier architectures to automatically detect flooded areas in polarimetric radar imagery collected by the NASA/JPL Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) instrument. UAVSAR is a fully polarimetric L-band synthetic aperture radar, flown aboard a NASA Gulfstream III aircraft.

Poster 6 Thumbnail

Deep Earth Learning, Training, and Analysis (DELTA) - Automating Machine Learning for Earth

Authors
P Michael Furlong, Brian Coltin, Scott McMichael, Tevin Achong, Roberto Campbell, Deanna Flynn, Keanu Nichols, Elizabeth Carter, Kevin Dobbs, Jonathan Eggleston, Rachel Sleeter
Theme
Earth Science
Poster ID
6

Abstract: Machine learning has achieved “human-level” intelligence in tasks ranging from object recognition and speech recognition to mastering the game of Go. However, Earth scientists have yet to fully take advantage of deep learning’s potential. The biggest obstacles are lack of expertise, the high barrier to entry for existing deep learning toolkits, and the intensive computational and data requirements. We are addressing these challenges with DELTA (Deep Earth Learning, Training, and Analysis), a toolkit for Earth scientists and commercial analysts to easily apply deep learning to their own problems.

Poster 11 Thumbnail

Improve Hurricane Intensity Forecast by Machine Learning of NASA Satellite Data

Authors
Hui Su, Longtao Wu, Raksha Pai, Alex Liu, Peyman Tavallali, Albert Zhai, Jonathan Jiang, Mark DeMaria
Theme
Earth Science
Poster ID
11

Abstract: Employ machine learning (ML) techniques and apply NASA satellite observations to improve tropical cyclone (TC) intensity forecast, especially rapid intensification (RI) forecast.

Poster 12 Thumbnail

Automated Machine Learning as a Service for the Earth Sciences

Authors
Brian Wilson, Alice Yepremyan, Diego Martinez, Sami Sahnoune, Edwin Goh, Sujen Shah, Kai Pak, Santiago Lobeyda, Chris Mattmann, Wayne Burke
Theme
Earth Science
Poster ID
12

Abstract: As part of the DARPA D3M program, JPL is curating a library of ML/DL “primitives” (algorithms) with sufficient metadata and hyperparameter tuning hints to enable auto-assembly (in Python) of pipeline steps. These steps include preprocessing, feature extraction & selection, tuning an ensemble of models, ranking models using a metric, etc. The library contains 90+ classic ML algorithms from scikit-learn, pre-trained deep learning (DL) nets from Keras & PyTorch, and a set of advanced primitives from the D3M performer teams. JPL’s MARVIN tools provide an environment to annotate, discover, install, compose, and execute ML/DL primitives and pipelines. Pipelines and metadata are specified in a declarative manner using a community-defined JSON schema and taxonomy. MARVIN automates the creation of Docker containers containing the primitives and software dependencies, which are executed on a Kubernetes cluster either on premise or at any Cloud vendor supporting Kubernetes. D3M is designed to solve 15+ problem types.

Poster 13 Thumbnail

Development of Gap-Agnostic Machine Learning Techniques for Earth Science Applications

Authors
Soni Yatheendradas, Sujay Kumar, Christa Peters-lidard, David Mocko
Theme
Earth Science
Poster ID
13

Abstract: (Not available.)

Poster 14 Thumbnail

xBD: A Dataset for Assessing Building Damage from Satellite Imagery

Authors
Ritwik Gupta, Bryce Goodman, Nirav Patel, Richard Hosfelt, Sandra Sajeev, Eric Heim, Jigar Doshi, Keane Lucas, Howie Choset, Matthew Gaston
Theme
Earth Science
Poster ID
14

Abstract: Logistics, resource planning, and damage estimation are difficult tasks after a natural disaster, and putting first responders into post-disaster situations is dangerous and costly. Using passive methods, such as analysis on satellite imagery, to perform damage assessment saves manpower, lowers risk, and expedites an otherwise dangerous process.

Poster 15 Thumbnail

Exploring Sentinel-1 and Sentinel-2 Diversity for Flood Inundation Mapping Using Deep Learning

Authors
Goutam Konapala, Sujay V Kumar
Theme
Earth Science
Poster ID
15

Abstract: Identification of flood water extent from satellite images has historically relied on either synthetic aperture radar (SAR) or multi-spectral (MS) imagery. But MS sensors may not penetrate cloud cover, whereas SAR is plagued by operational errors such as noise-like speckle challenging their viability to global flood mapping applications. An attractive alternative is to effectively combine MS data and SAR, i.e., two aspects that can be considered complementary with respect to flood mapping tasks. Therefore, in this study, we explore the diverse bands of Sentinel 2 (S2) derived water indices and Sentinel 1 (S1) derived SAR imagery along with their combinations to access their capability in generating accurate flood inundation maps using a fully connected deep convolutional neural network known as U-Net.