Visualization of uncertainties in 2D images

Two-dimensional (2D) images are often used to communicate the results of scientific investigations and predictions. Examples are weather maps, earthquake hazard maps and MRI slices. In contrast to statistical analyses of individual variables or time series, there are currently no established methods for visualizing the uncertainties in the 2D images. However, this would be necessary to make the information in the 2D images clear to scientists as well as to the non-expert public audiences in order to avoid misinterpretation and over-interpretation.

In this study, we demonstrate the challenges and approaches to uncertainty visualization using the case study of drought forecasting, which is relevant for climate adaptations and mitigations. A drought is a deviation (anomaly) from the parameter value expected from long-term data. In our case, the parameter under consideration is soil moisture, which is an important parameter for various environmental processes. The soil moisture can be used in combination with soil type to estimate the amount of water available to plants in the topsoil. If the amount of water available to plants according to the so-called percentile approach deviates significantly from the value expected from long-term data, this is referred to as an agricultural drought.

The drought forecast is based on ensemble modelling. This means that the results of various weather forecast models are used to predict the development of soil moisture for the period of the weather forecast. For each weather model used, a possible soil moisture development is predicted. Each of these is used for a drought forecast. The result of the ensemble modelling is therefore several forecasts, which can differ significantly. Due to the use of different weather models and the consideration of uncertainties in the models, the result of ensemble modelling is therefore a large number of drought forecast maps. When visualising the results, often only a map of the mean values resulting from the predictions is shown. If only the mean value is displayed, however, the information about a possible difference and thus the uncertainty of the predictions is lost. In other words: If individual cases from the ensemble predict the possibility of drought, this will not be clearly visible in the mean value map.

In this presentation, we will demonstrate and discuss different approaches to visualize the uncertainty in the prediction.