Hindcasts simulate how accurately a forecasting model would have predicted past events, essentially a “forecast of the past”. All of the hindcasts are out of sample
Options for accessing Salient GemAI hindcast (from 2020-01-01 onward).
| Source | Description | Best For | Limitations |
|---|---|---|---|
| Bulk data archive | Salient provides all hindcasts | Fast validation "on rails" | Limited flexibility |
| 260 locations | |||
| cloudflare_access notebook | Pull hindcasts from a cloud bucket | Direct access to native hindcast zarr archive | No debiasing |
| No ECMWF ENS/GEFS | |||
| Requires xarray familiarity | |||
| Salient Software Development Kit | Use validate_gem notebook to get hindcasts and calculate skill | Streamlines parallelism, rate limiting, and caching | Large data pull |
| SDK is python-only | |||
| API | Get data directly from HTTPS | Maximum flexibility | Hand-code data access & manipulation |
A hindcast archive for 260 global airport locations is available in netCDF format for the following variables: Min, max, avg temp, precipitation, and wind speed 10m
This archive includes the following:
A hindcast archive for global gridded data and custom locations is available from Cloudflare, a cloud bucket in a zarr store. If you do not require debiasing and have some familiarity with xarray, then you can use this option by requesting an access key. Email [email protected] if you do not have credentials.
Accessing Salient’s forecasts via Cloudflare: cloudflare_access
You can also access GEM forecasts using the SDK, downloading each individually.
To get started, install the python Software Development Kit (SDK) and make sure that you have Salient access credentials (username+password or apikey). Instead of using the raw API, the SDK contains utilities that simplify data calls, parallelize for efficiency, and cache to prevent repeated calls.
In each of these notebooks, you will provide a set of named lat/lons to acquire. The notebooks all contain code that compares the (optionally-debiased) forecast to a "truth" value (either ERA5 analysis or met station observations) to calculate a skill score
If your organization is not a python shop or restricts software installations, you can also access and validate Salient forecasts using the fully-documented Salient API. This path will require the most effort to implement. Key steps: