Delphi Forecast Reports

Overview

Check the weekly health checks first: they list missing locations, horizons, and ensemble components, and the pipeline’s warnings and errors.
The weekly fanplots are used by the team to visually inspect the forecasts.
The season reports provide a general analysis of the 2024-2025 season’s data and forecaster performance.
The backtesting reports were pre-season tests of a variety of forecasters on the 2023-2024 season’s data.
The current forecaster consists of
- Main component: An AR model using the most recent week and the week before that, with training data taken from a 7 week window centered on the date of the forecast in all previous year’s data. This is the seasonal training window described here
- Minor secondary component: The same model without nssp, but potentially normalized and using alternative historical sources. This is described here.
- Backup: climate linear, which combines the historical quantiles with a linear extrapolation of the most recent data.
A more detailed description of the forecaster families explored is on the forecasters page.