Verisk Overview
I worked primarily on 7 Projects at my time at Verisk. Despite not having much prior experience with mapping or meteorology, I learned fast and became a sort of specialist.
A glance at the 7 projects I did in my ~30 month time at Verisk NOAA Global Severe Weather Prediction
A 16 month machine learning project to detect damaging weather events, similar to that of NOAA's SPC (below), but globally
Components:
- Python scripts for gathering atmospheric weather data(NOAA) and saving to AWS s3 buckets
- Imputation of data that is available globally (NOAA is just USA)
- Inputting into machine learning pipeline for converting our 25+ variables into risk probabilities
- Several iterations of models, loss functions, feature selection
- Full stack pipeline that could automatically fetch, transform, predict, display, and show metrics and about the weather events.
Skills:
- Working with AWS (s3 buckets, EC2 instances, etc)
- Learning how to operate in a new domain (no prior experience in meteorology)
- Working with a variety of Machine Learning models
- XGBoost, Gradient Boosted Regressor, and Randomized Search CV
- Delivering and demonstrating results to a team as the sole person on a project
- Dealing with data lakes, using compression algorithms, and coding efficiently to meet the strict demands of "real time" output.
Building Footprints ML Project
The team was expanding their risk detection globally, and needed maps showing where buildings are located, along with stats on them. In some regions, this data is not available, and we needed to find ways to make our own.
Components:
- Filling in missing data to create a map of all buildings across the globe
- Extensively learned about data usage licenses and rights, such as creative commons, fair use, etc.
- Determined there were no commercially usable datasets of this kind (without limitations), so we built our own
- Took multiple approaches, , satellite imagery + Machine Learning, accessing and gathering government datasets, combined to make good maps
- Used ArcGIS Pro for validation and cleaning
Skills:
- Working with several different map data formats
- Contacting government agencies in different countries to apply for access to their data
- Following strict data use guidelines and rules
- Retrieving, validating, transforming, and storing data from various sources, languages, and formats
- Performing all these tasks within ArcGIS Pro
Mapping Dashboard
I also designed a Dashboard in my Internship time using Mapbox, HTML, and JavaScript
- API Key for Mapbox is required
- Contact me or create one for free at mapbox.com for access