Climatology refers to climatological normal conditions, or the average value for a particular weather variable for a given time of year. Climatology is defined across several parameters - variable, location, and timescale. An additional parameter is reference period, which is the number of years to use to calculate the climatology. The scientific convention has been to use a 30-year reference period that is updated every 10 years. However, Salient provides alternate reference periods, or climatologies, for different use cases.
Climatologies are used to:
The Salient blend model intelligently combines the most skillful forecasts across location and time of year; occasionally that is a climatological forecast. Salient’s probabilistic climatological forecasts provide full probabilistic distributions that characterize the uncertainty around the calculated climatological mean. The available options: Salient Climatology, 30-year Climatology, 10-year Norm or 5 year Norm are outlined in the table below.
| Model Name | Timescale | API model |
Update frequency | Description |
|---|---|---|---|---|
| Salient Climatology | Weeks 1-5 Months 1-3 Quarters 1-4 | salient_clim | 5 years | Salient’s proprietary 40-year trailing average based on the global ERA5 dataset. It includes a linear estimate of trend and is smoothed/tuned for optimal performance. We apply a series of low-pass filtering operations to remove day to day noise, but maintain the sharpness of seasonal transitions to improve the usability and accuracy of our proprietary climatology. It is probabilistic and properly cross-validated. Reliability diagrams and the tercile (above-below-near normal) view are based on the Salient climatology. |
| 30 year Climatology | Weeks 1-5 Months 1-3 Quarters 1-4 | clim | 10 years | 30-year stationary baseline with smoothing and no trend. It is probabilistic and properly cross-validated. This is equivalent to NOAA’s normals, e.g. 1991-2020, 1981-2010, 1971-2000. |
| 10 year Normal | Weeks 1-5 Months 1-3 Quarters 1-4 | norm_10yr | Annual | Simple 10 year trailing average |
| 5 year Normal | Weeks 1-5 Months 1-3 Quarters 1-4 | norm_5y | Annual | Simple 5 year trailing average |
The most common use of a climatology is to convert forecast values into anomalies, through the equation **anomaly = forecast value - reference climatology**. The values within a specific forecast are constant, but the anomaly can vary based on the selected reference climatology. These reference climatologies are not probabilistic - they are simply the mean value of a given variable at a given location (with optional temporal smoothing).
| Reference Period | reference_clim | Time Period | Update frequency | Description |
|---|---|---|---|---|
| 30 year Climatology | 30_yr | 1991-2020 | 10 years | NOAA’s 30-year historical average with smoothing |
| 10 year Normal | 10_yr | 2019-2023 | Annual | 10 year straight average |
| 5 year Normal | 5_yr | 2014-2023 | Annual | 5 year straight average |
<aside> ❗ Note: The median anomaly of a climatological forecast model will only be 0 if using the appropriately matched reference period - i.e. the 30-year climatological model and the 30-year reference period.
</aside>
/v2/forecast_timeseries - reference_clim: 30_yr, 10_yr, 5_yr
The forecast climatologies can be used as baselines when calculating skill scores. This allows users to see the relative effect of using different historical periods to calculate climatologies on the skill of the forecasts. For example, you can compare the skill improvement of Salient’s forecast compared to a 10 year or 5 year normal climatology.