Understanding Probabilistic Climate Forecasts and Uncertainties in Long Range Forecasts
By Eunice Koech
With contributions from the ICPAC Climate Change Technical Working Group
Weather is defined as the atmospheric condition of a particular place over a short period while climate refers to an average weather condition over a long period of at least 30 years. Many social and economic activities rely on weather conditions and this makes weather forecasting an essential undertaking in society. Forecasting began many centuries ago, and different communities used different signs to predict weather. This traditional knowledge commonly referred to as Indigenous Knowledge (IK) has been passed along generations and some communities (e.g. The Nganyi community of Kenya) still use to date. Forecasting of weather and climate has evolved over the last years and new techniques are progressively being adopted to increase the accuracy of the forecast.
Nowadays, climate scientists use complex mathematical models that take into account a wide range of parameters that are referred to as ‘indicators’ of the elements (e.g. rainfall) being forecasted. To forecast rainfall for instance, models take into account indicators such as sea surface temperatures, winds, soil moisture and humidity. Forecasts that are longer than a week are difficult to predict hence the need to adopt methodologies that are able to indicate the degree of error that is associated with the forecast. Since a wide range of atmospheric activities change from time to time, rainfall forecasts for example, for one month or three months period, given one month in advance have not always been accurate. Climate scientists have, however, accounted for the unpredictability of atmospheric patterns by giving the forecast in terms of probabilities.
ICPAC as a Regional Climate Centre gives reliable forecasts on both sub-seasonal (weekly and monthly) and seasonal timescales. Both monthly and seasonal forecasts are probabilistic in nature.
Why probabilistic climate forecasts
Probabilistic forecasts give the likelihood of an event occurring with reference to long term average (model climatology). Probabilistic climate forecasts are the best method of giving a forecast that is associated with uncertainties. Due to the uncertainties associated with long-range forecasts, ICPAC issues objective rainfall and temperature forecasts in probabilistic format.
Probabilistic forecasts communicate the degree of occurrence of an event hence providing an opportunity to the user to make a decision on what investments to make. Users are able to interpret the forecast in a way that is relevant for the sectors, and take action where necessary. For instance, a forecast of 50% chance of drier than usual conditions over a region can be used by humanitarian groups to mobilize food and non-food items for the population at risk.
Sources of uncertainties in climate forecasts
The uncertainties that are associated with long range forecasts come from different sources such as model initial conditions, external forcing, and model structure.
Model initial conditions refers to the state of the climate system as at the time the model is run. To get these initial conditions, parameters such as temperature, humidity, wind among others are measured around the globe. However, it is very difficult to get the exact conditions of the climate system globally. This is because first, observation networks are limited and secondly, measuring these parameters will always carry an element of some error even with quality control measures being employed. Due to the complexity of the climate system, small uncertainties in measuring initial conditions become significantly magnified in long range forecasts.
External forcing refers to the effects of external factors that were not taken into consideration when the model was run. Climate models incorporate various parameters that are proven to affect climate, but it is impossible to include all the parameters; For example some parameters have short life-span hence making it impossible to take them into consideration at the time of issuing forecasts. These factors eventually affect the season’s performance to various degrees.
Model structure is another source of uncertainty in climate forecasting. Climate models that scientists use in forecasting are realistic, but never perfect in representing climate processes. Models, for instance, divide the atmosphere into square boxes called grids (Figure 1), which are several kilometers wide. The models’ mathematical equations are applied for every grid point and forecasts made. However, these boxes break the actual continuity of the atmosphere, hence fine details are omitted in the model calculations. This limitation in model structure, therefore, causes uncertainties in the forecast.
Interpreting rainfall probabilistic forecast
A sample probabilistic forecast that was issued by ICPAC for MAM 2021 is shown in Figure 2. A forecast is generated by many models for each grid point and the result given in three categories that sum up to 100 as shown below. These categories are above (A) average (wetter than usual), average (N), and below (B) average (drier than usual).
The forecast above, for example, was divided into three zones: Zone 1: This zone is represented by the green areas. The map only shows the category with the highest probability (45), but the other probability values are indicated in the text (30 and 25). For this zone, the most likely category was above average (A) with a probability of 45%. This means that there is a 45% chance that above average (wetter than usual) rainfall will be recorded. Although this is the most likely outcome, there is also a 30% chance that average (N) rainfall will be received and 25% chance that below average (B) rainfall will be recorded. In zone III, the below average (B) category has the highest chance (45%) of occurring while average (N) and above average (A) categories have chances of 30% and 25% respectively. All the three categories in Zone II have equal chances (33%) of occurring, this shows that the model did not have confidence over that area.
Food for thought: Now that we have explained how to interpret a probability forecast, let us see if you got it right. In Zone 1, the forecast showed that there is a high likelihood that wetter than usual (45% chance) conditions will occur. However, at the end of the season, it was evident that average conditions were recorded. Was the forecast wrong?
