Hello
I'm Ellie White (she/her) currently a senior data scientist with the Data Science Branch at U.S. Geological Survey. I spend about 3/4 of my time working on data science projects and the other 1/4 on improving our data science best practices.
Examples of my USGS work available on GitHub include:
- 🔮 from workflows to streamflows: how we build shared understanding to predict streamflow droughts using deep learning. Poster I made for AGU Fall meeting 2023 on how we work. https://github.com/elliewhite-usgs/dissemination/blob/main/agu_2023_drought_final.pdf
- 🌊 beyond-a-reasonable drought: Data visualization I made for #chartchallenge23, and the chart is meant to communicate the difference in the length of droughts in the Common Era. The hardest part was setting up the pipeline that processes the data. https://github.com/whiteellie/beyond-a-reasonable-drought
- 🐟 fish-in-hot-water: Data visualization I made for #chartchallenge23, and the chart is meant to communicate the effects of climate change on fish spawning. https://github.com/whiteellie/fish-in-hot-water
Examples of my USGS work available on GitLab include++:
-
🌍 national-geospatial-attributes-template: Created an R-based rescaling attributes pipeline (using targets), which transfers data from one spatial polygon to another. The 10 forks spawned off this template pipeline support the nationwide water availability assessments and allows one to produce harmonized datasets with a consistent method. https://code.usgs.gov/wma/dsp/pipeline-templates/rescaling-attributes-template
-
🏁 Created an R pipeline and Python workflow that processes land cover information at the catchment scale for CONUS while leveraging HPC resources to handle large-scale geospatial data. These data are used as inputs into machine learning models improving their predictive capability.
-
✅ model-eval: Developed pipelines for drought model evaluation and test/train splits using data on AWS S3 buckets. These pipelines integrate geospatial basin groupings to address spatial and temporal autocorrelation improving model performance assessment. The model evaluation pipeline implements drought pooling and matching for evaluation of observations against predictions. It how well a given ML model predicts streamflow droughts across several metrics important to decision makers including drought occurrence, onset, recovery, severity, and duration. https://code.usgs.gov/water/RegionalDroughtEarlyWarning/modeling/model-eval
Examples of my work under other affiliations:
- 🎓 geostatistics-tutorial: Took concepts and computations done manually in a geostatistics course in the Department of Hydrology at UC Davis and translated it to R. Other students have used this to guide their studies and graduate research. https://github.com/whiteellie/geostatistics-tutorial
- 👑 us-presidents: Made a classroom style poster of all U.S. presidents calling out some notable events in history. https://github.com/whiteellie/us-presidents
Writing samples:
- 👌 quality-assurance: Quality assurance plans for predictive modeling and data visualizations that list best practices and the Data Science Branch's cultural norms. https://github.com/whiteellie/quality-assurance
- 📔 quirky-keras: blog post about asymmetric loss functions. https://towardsdatascience.com/quirky-keras-custom-and-asymmetric-loss-functions-for-keras-in-r-a8b5271171fe
++ You will only have access to these repos if you are a USGS employee on VPN