ESIP Lab Update: March 2019 

In this email: 

1. Google Summer of Code Update – SO much activity
2. Lab Funded Project Highlight: Deep Learning for Hurricane Prediction
3. Challenge Highlight: Earthquake forecasting
4. Augmenting snowpack modeling with citizen science observations

Google Summer of Code Projects

• Geo-Selection For Hyrax Data Request Form (OPeNDAP)
• OrcaCNN: Detecting and classifying killer whales from acoustic data (Axiom Data Science)
• Developing methods and libraries to ingest in situ data into Google Earth Engine (Montana State University)
• A next-generation GUI for visualizing big gridded data in Python (USGS/Anaconda)
• Ag-Net: building a customized deep neural network for recognizing crop categories based on spectral characteristics (George Mason University/Axiom Data Science)
• Implement advanced features in the Community Ontology Repository (ESIP Semantic Technology Committee)

Lab-Funded Projects Highlight: Deep Learning for Hurricane Prediction 

We proposed a novel, artificial intelligence-driven approach using a deep learning algorithm, to predict hurricanes. We developed a hybrid three-step (direction, distance traveled, and intensity) deep-learning based hurricane forecasting model using the output of dynamical hurricane models, remote sensing data, and observation network. The preliminarily results of our model for 19 tropical storms in 2017 showed statistical advantages (~8% and ~30% improvement in track and intensity forecast biases respectively) over the National Hurricane Center official forecasts for 24-hours ahead forecasts. For predicting more than one-day ahead, we are currently implementing a post-processing analysis by ensemble Kalman filter to mitigate the bias of our deep learning ensemble model. Based on the new model, we are proposing the long-term hourly forecasting of hurricane’s intensity, path, precipitation for the entire 2017. This proposal will be featured in NASA Advanced Information Systems Technology (AIST) program. We also submitted an abstract to the 1st Workshop on Leveraging AI in the Exploitation of Satellite Earth Observations & Numerical Weather Prediction, hosted by NOAA, in late April. We are also working to create a framework for an operational hurricane forecasting system across the United States.

Challenge Highlight: Can you predict laboratory earthquakes?

Augmenting snowpack modeling with citizen science observations

Lab Fellow Ben Roberts-Pierel highlights one of his favorite data exploration tools

In the snow hydrology world, as with many fields, there are limited in situ stations to monitor snowpack in many of the world’s mountain regions. Water content in snow is measured by snow water equivalent (SWE), a product of snow density and depth. However, with a lack of in situ data and extreme heterogeneity in mountain regions, there are many times a paucity of resources for forcing or validating models to predict SWE and forecast runoff. One way that this deficit is being addressed is through citizen science.

The use of citizen science data has long occupied a hazy and at times controversial area in scientific research, but with projects like the Globe Observer Program and others, NASA has shown they are fully behind the idea. One group that is working on addressing this issue is Community Snow Observations, a group of snow hydrologists and researchers, funded by NASA’s Citizen Science for Earth Systems project. They have teamed up with outdoor professionals, winter recreators and a host of other organizations interested in understanding cryospheric change and water availability to gather snow depth data from across the Northern Hemisphere. Snow depth is a crucial yet easily measured data point, particularly given that many people traveling in the mountains are already carrying a simple depth gauge in the form of an avalanche probe. Researchers, including colleagues here at Oregon State, are then using these data in existing models. The inclusion of these additional depth data has resulted in marked improvements to their outputs.