Learn about what causes uncertainty in seasonal forecasts. If you are new to this topic, you may want to first read What are Seasonal Forecasts? to understand the basics of seasonal forecasts and Considering Uncertainty in Seasonal Forecasts to learn more about uncertainty.
Seasonal to decadal forecasts
Article
10 minutes
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:
The sections below provide more information on these factors.
Seasonal forecasts take advantage of longer-term predictability in the climate system to reduce uncertainty deriving from inherent limitations in the predictability of the climate.
Even the most skilled seasonal prediction system has limitations when it comes to predicting the climate at timescales longer than about 14 days. The atmosphere is generally unpredictable at these timescales, meaning the seasonal prediction system cannot predict daily or weekly weather conditions within a season beyond about 14 days1. Consequently, seasonal prediction systems forecast the overall climate conditions for the entire season, such as mean temperature and total precipitation. Forecasting conditions over a season or longer capitalizes on the predictability of longer-term climate system patterns and uses this predictability to reduce uncertainty and improve the accuracy of the seasonal forecast.
The predictability of meteorological conditions longer than about 14 days in the future is primarily limited by the chaotic nature of the atmosphere, whereby small differences in initial conditions can lead to large differences in the final state2, as discussed further below. Other parts of the climate system such as the land and oceans change more slowly, and their influence on the atmosphere allows for longer-term predictability in climate conditions, enabling seasonal forecasts. For example, large-scale ocean/atmosphere patterns like the El Niño-Southern Oscillation (ENSO) influence the climate across the globe, and strong El Niño or La Niña events are linked to more skillful seasonal forecasts. More information on the sources of predictability on a seasonal timescale can be found in the What Are Seasonal Forecasts? article.
Probabilistic seasonal forecasts, like the forecasts on ClimateData.ca, are designed to measure and communicate the uncertainty inherent in longer-term predictions and provide a more complete understanding of forecast outcomes.
Seasonal forecasts are generated using an ensemble of predictions. An ensemble is a set of many individual model simulations from the seasonal prediction system, each with a slightly different starting point. These starting points, known as the “initial conditions“, are based on recent climate observations and are always subject to some degree of uncertainty due to measurement errors and observational gaps in time and space. Although these differences in the initial conditions are slight, the differences amplify over time due to the chaotic nature of the atmosphere. Figure 1 shows how small differences in the initial conditions can result in significant differences over time (sometimes referred to as the “butterfly effect”2).
All model simulations in the ensemble are considered together to estimate the probability of different outcomes (e.g., the probability that the mean temperature in summer, from June to August, will be above, near, or below normal). This type of probabilistic forecast helps to address uncertainty by providing a comprehensive view of the range of climate possibilities for a future season1.
Figure 1: An ensemble of model simulations evolves over time to produce a probabilistic forecast. The figure shows the temperature difference from the historical median (represented by the dotted line) for a given location. The historical median comes from the historical climatology for 1991 to 2020. Each of the 40 simulations begins with a slightly different starting point or “initial condition”, represented by the curve on the left showing the distribution of these conditions. The model is then run forward in time, with each simulation evolving differently as shown by the thin grey lines. Due to the chaotic nature of the atmosphere, small differences in initial conditions amplify over time. The range of predicted future temperatures is shown by the red (above normal), grey (near normal), and blue (below normal) shading. The median of this ensemble of simulations is represented by the solid black line. The vertical dashed lines span the season being forecast (i.e., the target season), and the curve on the right shows the distribution of temperatures over the target season that is used to generate the probabilistic seasonal forecast for the location.
Seasonal forecasts on ClimateData.ca are produced using the Canadian Seasonal to Interannual Prediction System version 3 (CanSIPSv3). CanSIPSv3 combines forecasts from a weather and a climate model. These weather and climate models are numerical representations of the major processes that govern the state of the atmosphere, land, ocean and sea ice. While these models simulate the most important physical processes that determine weather and climate, they approximate or omit some physical processes that can affect weather and climate. As such, these models provide a very good, but not perfect representation of the weather and climate system.
The strengths and limitations of CanSIPSv3 are assessed by evaluating its performance over the period from 1991 to 2020. One type of performance metric is the skill of the prediction system.
Skill indicates how much trust can be placed in the seasonal prediction system, based on its past performance3. Skill is calculated by comparing forecasts produced using historical initial conditions, also known as hindcasts, with relevant observations.
Generally, the CanSIPSv3 skill depends on:
All seasonal forecasts on ClimateData.ca include information about skill, which should be considered when using the forecasts.
Forecasts on seasonal timescales include uncertainty, however, this uncertainty can be addressed by:
For more tips on how to consider uncertainty when planning and making decisions based on seasonal forecasts, read the Considering Uncertainty in Seasonal Forecasts article. For more information about seasonal forecasts, see the Learning Zone articles in the Seasonal to decadal forecasts section.
This project was undertaken with the support of:
