Most of this work shares a pattern: take large, messy environmental data and turn it into something validated, reproducible, and usable by the people who make decisions with it.
Co-develop California's cloud-based climate-data platform, delivering downscaled CMIP6 projections and analytical tools to energy utilities, regulators, and the public. I build the open-source Python packages and Jupyter notebooks that guide users through geospatial climate analysis — work that underpins California's Fifth Climate Change Assessment.
An open-source platform synthesizing ~15,900 surface weather stations from 27 networks across the Western U.S. (1980–2022). I helped design custom QA/QC protocols that reliably flag energy-relevant extremes — Santa Ana winds, heat, precipitation, solar radiation — so the dataset is both defensible and ready for downscaling work.
A Python toolkit and notebook suite that ingests decades of NOAA / NCEI Storm Events records and detects compound hazards — co-occurring and back-to-back events — at the county level. The hard parts were the data and the definitions: schema drift across years, county-vs-NWS-zone geography, reporting bias over time, and making "compound event" an explicit, configurable definition. It runs operationally, refreshing classifications and stakeholder-ready HTML summaries as new data is published.
High-resolution hourly profiles of solar-PV and wind generation potential through end-of-century for the California Energy Commission's SB100 supply analysis. The focus is on extreme low-generation events — solar and wind "droughts," and their co-occurrence (Dunkelflaute) — that stress a zero-carbon grid.
A precipitation "forecast handbook" for the Santa Clara Valley Water District, using pattern recognition to link large-scale weather patterns to rainfall and runoff — so the district can act on a storm's forecast rather than react to its impacts.
Climate projections for the Sacramento Municipal Utility District (1.5M+ customers), translating temperature, precipitation, wildfire, wind, and hydropower signals into infrastructure and operations guidance — framed for planning and executive audiences.
Ported and ran the Regionally Refined E3SM model on the BigRed200 supercomputer, generating 200+ TB of historical and future output. I designed a 14 km regionally-refined grid and engineered Dask + SLURM pipelines across ~3,840 cores — cutting analysis of 200+ TB from months to about a week — and contributed to recreating California's 1997 New Year's flood in a regionally refined Earth system model.
Want the full record?