Skip to main content

Meta-Model Weather Forecasting

Combining public and private weather forecasts into a weighted ensemble that beat the best single vendor by 20%.
Climate
Forecasting

Overview

No single weather forecast is best everywhere. At ClimateAi, I built a meta-model that combines several public and private forecasts, weighting each one by its recent skill at each location and lead time, to produce more accurate short-term forecasts of mean temperature and precipitation.

My Role

  • Designed a multi-model ensemble algorithm combining 6 data sources, with separate weights for 30 customer locations and 4 lead times
  • Improved short-term forecasts by 20% over the best vendor’s model
  • Built automated Google Cloud pipelines for geospatial data extraction, preprocessing, and forecast skill & uncertainty analysis, updating the model weights weekly
  • Presented Tableau map visualizations comparing our model’s performance with competitors’

Tools

Python · Xarray · pandas · Google Cloud · Tableau

Outputs

  • Kadi, J., Ahmadalipour, A., Evans, M. C. Meta-model Weather Forecasting: Integrating Public and Private Weather Forecasts. AGU Fall Meeting, 2020. Poster. Abstract