Learn how to consider uncertainty when using seasonal forecasts. If you are new to this topic, you may want to read What are Seasonal Forecasts? to understand the basics of seasonal forecasts.
Seasonal to decadal forecasts
Article
10 minutes
Seasonal forecasts provide information on climate conditions beyond the time frame of weather forecasts, which affords more time for the information to be included in planning and decision-making processes. However, the predictability of seasonal climate conditions declines as the forecast window extends into the future. Hence, understanding and considering uncertainties in seasonal forecasts enables users to extract the maximum value from these forecasts.
Uncertainty is inherent to any prediction. In the context of prediction, “uncertain” does not mean “speculative”. Rather, it indicates that our knowledge of future climate conditions is incomplete. Uncertainty in seasonal forecasting is primarily caused by a combination of the following factors1:
See the article “What causes uncertainty in seasonal forecasts?” for a detailed description of these factors.
The uncertainty caused by the three factors listed in the section above means that probabilistic forecasts do not provide a single prediction. Instead, probabilistic forecasts provide the probability of different outcomes occurring, most commonly whether conditions will be above, near, or below normal. Normal is defined using a historical climatology for the 1991 to 2020 reference period. Forecasts on ClimateData.ca also provide the probability of conditions being unusually high or unusually low (e.g., the probability of an unusually high mean temperature or an unusually low total amount of precipitation for the season).
To extract the most value from the information in a seasonal forecast, the probabilities for all outcomes should be considered; however, in most cases, even decisions based on the most probable outcome from each forecast will be more beneficial than decisions based solely on past events. Learn more about the benefits of using seasonal forecasts repeatedly over time in the National Oceanic and Atmospheric Administration’s blog articles titled “Betting on Climate Predictions” and “When it comes to probabilities, don’t trust your intuition. Use a decision support system instead!”
Interpreting forecast probabilities
If above-, near-, and below-normal outcomes were equally likely, each would have a 33% chance of occurring; this represents the climatological probability of each outcome. For the forecasts on ClimateData.ca to be considered statistically different from climatological chance, the probability of any one outcome (above-normal, near-normal, or below-normal) must exceed 40%.
An outcome with a probability above 40% indicates that it is more likely to occur than the other outcomes. However, the observed outcome (i.e., the seasonal conditions that actually occur) isn’t always the one with the highest forecasted probability1.
For example, consider a forecast that predicts:
This forecast can be interpreted as a 70 in 100 chance of experiencing above-normal temperatures, a 20 in 100 chance of near-normal temperatures, and a 10 in 100 chance of below-normal temperatures. In this case, the seasonal prediction system would not be considered “wrong” if a near-normal or below-normal outcome occurred. While these outcomes are less probable, they are not impossible, and are in fact expected to occur some of the time.
This example forecast can be visualized using a wheel spinner, as illustrated in Figure 2 of the What are Seasonal Forecasts? article.
The “success” or “failure” of the seasonal prediction system is never assessed based on the performance of a single probabilistic forecast, because, as noted above, lower probability outcomes are still expected to occur some of the time. Instead, the skill of the seasonal prediction system is calculated using forecasts produced for past seasons between 1991 and 2020.
It is important to consider the skill of the seasonal prediction system when using seasonal forecasts. A performance metric for skill is provided with each forecast on ClimateData.ca. Skill typically improves as the forecast release date gets closer to the season being forecasted. Forecasts are updated each month. Skill can be calculated using many different performance metrics.
On ClimateData.ca, skill for probabilistic forecasts is provided using the Continuous Ranked Probability Skill Score (CRPSS), with a higher CRPSS value indicating higher skill.
The skill levels on ClimateData.ca are as follows:
Figure 1: Skill levels on ClimateData.ca are represented using a star rating system.
Consider using flexible planning processes and checking updated forecasts as the season of interest approaches, regardless of skill level.
To better understand what forecasts could mean for a location2, planners and decision-makers should consider past climate conditions for that location. Past climate conditions are provided as part of the ClimateData.ca forecasts using the historical climatology. Some locations may have unique terrain features, such as mountains or lakes, which affect the local climate but are smaller than the resolution of the historical climatology (i.e., smaller than a single grid box). It is therefore important to consider the effect of local terrain, as the historical climatology may not accurately represent past conditions at that specific location within the grid box. This is especially true in regions with large elevation changes, such as mountain ranges and coastal areas. In these cases, the forecast probabilities still hold, but local knowledge could be used to adjust the historical values.
Technical Note
The historical climatology on ClimateData.ca covers the period from 1991 to 2020. The historical median for each variable and season over this 30-year period is available. The cutoffs used to define the forecast outcomes are also available.
The historical climatology is provided on a 1/12 x 1/12 degree (approximately 10 km x 10 km) spatial grid across Canada.
Having information in advance on the most probable seasonal conditions can be beneficial for planning and decision-making3. Using seasonal forecasts enables planning to begin early and allows more time for preparation.
The decision of when and how to act based on probabilistic forecasts depends on various factors, including the intended use of the forecast and the risk tolerance of the decision-maker. Regardless, it is recommended that users check forecasts monthly, as forecasting skill generally improves as the season of interest approaches. Although seasonal forecasts are provided up to a year in advance, the skill level tends to be much lower for seasons more than a few months away. This is especially important for climate variables that typically have lower skill, such as precipitation. Ideally, plans should be re-evaluated and adjusted as updated forecasts become available.
For example, a decision-maker, who is planning for the July to September season, may need to make a decision by June 15th. In April, they consult the seasonal forecast on ClimateData.ca in order to understand the probabilities of different potential outcomes for this period. Based on these probabilities and the skill level of the forecast released in April, they can start planning flexible options, knowing that the forecast will be updated in May and again in June. They re-evaluate the forecast as it is updated, considering how the probabilities and skill level have changed, and refine their plans accordingly.
For more information, see the Learning Zone articles in the Seasonal to decadal forecasts section.
This project was undertaken with the support of:
