RAMPVIS

Most complex models in the literature took decades and sometimes centuries to develop. Many were built on a substantial amount of post-hoc evidence and analysis, while many are still being improved today. The success can usually be attributed to a collective effort by generations of modelling scientists in data collection, observation, and analysis; hypothesis formulation; and model development, validation, deployment, monitoring, and improvement. In combating COVID-19, such a collective effort in decades has to be compressed into a period of weeks and months. Without adequate visualization (VIS) and visual analytics (VA) support, a modelling workflow typically compels modelling scientists and epidemiologists to spend more time acquiring information from data, to observe less data or make less frequent observations, to rely solely on data mining algorithms blindly without keeping track of potential limitations and errors, or to rely solely on intuition without using algorithms to explore different options. Hence, there is an urgent need to increase the VIS/VA capacity in such rapid model development workflows, and to maintain the provision of this capacity throughout the period of combating the COVID-19 pandemic. There are many different visualization tasks in different model-developmental workflows. They can be broadly categorised into four levels according to the size and complexity of the search space for an optimal decision.

Level 3. Analytical visualization, which is an investigative aid for examining and understanding complex relationships (e.g., correlation, association, causality, contradiction, etc.). Despite the advancement of data mining technology, our requirements analysis shows that most modelling scientists and epidemiologists do not have adequate access to data mining algorithms, such as time series analysis, classification and clustering, association analysis, and dimensionality reduction, nor the visualization techniques that accompany such data mining algorithms.

Level 4. Model-developmental visualization, which is a developmental aid for improving existing models, methods, algorithms, and systems, as well as for creating new ones. These visualization techniques are typically model-specific and developed through close collaboration between VIS/VA specialists and modelling scientists. Our requirements analysis shows that while almost all modelling scientists use visualization to depict simulation results and model predictions, most do not have access to VA specialists for developing techniques such as ensemble visualization and parameter optimization.