Sept. 8, 2025

• Debiasing is available for NOAA GEFS

debias=true for model=noaa_gefs&frequency=daily in forecast_timeseries

• Date event type: Enhancement to custom quantity definition to predict the date when certain conditions are met. This is particularly useful in agricultural use cases such as a suitable planting window.
• salientsdk has a new notebook (validate_gem) showing how to validate using debias & obs

https://github.com/Salient-Predictions/salientsdk-public/blob/main/salientsdk/examples/validate_gem.ipynb

• Change to forecast date defaulting logic if data parameter is omitted

The forecast date defaulting logic has been updated. The forecast availability time is 0800z. Now, if a date parameter is omitted from the API request, the forecast date is determined relative to 0800z. For example, if a user calls forecast_timeseries without a date parameter between 0000z and 0800z, the forecast_date will be set to the prior UTC date. Previously, calls made during this time window would generate an error if the processing was not complete or you could call the API at 0700z to get the most recent GEFS or GEM forecast if processing completed early.

This change does not affect API requests that include a date parameter.

This change impacts the following endpoints:

• /forecast_timeseries
• /risk
• /downscale
• /forecast_distribution

August 25, 2025

• “Table” format has been deprecated from the API. This change is not backwards-compatible.
• Percentile option is available for NOAA GEFS model so you can view other quantiles in addition to the 50th percentile.
• Salient SDK supports validation of GEM with debiasing enabled to compare against weather observation data

August 13, 2025

Debiasing Enhancements

New Debiasing model for GEM and Blend and ERA 5 Daily Downscale

A new postprocessing model, applied to 0.25deg native forecasts, provides the best estimate for the weather at a specific location  or list of locations (down to precision of 100m). On average, this model reduces the forecast error by about 1.5C when validated against a specific station. The performance gain varies by location with greater improvements near oceans and mountains.

This initial release is for temperature models (includes tmin, tmax, temp, cdd, hdd) and accessible via the API with the debias=True option for endpoints will be /forecast_timeseries?timescale=daily, /data_timeseries, /downscale.

August 11, 2025