Colin Ware, Peter Mitchell John G.W. Kelly, and Roland Arsenault.

To display patterns of winds or water currents researchers and forecasters mostly resort to graphical representations such as streamlines or grids of little arrows representing the vector field. Arrow grids are still the most common method despite an empirical evaluation (Laidlaw et al.) showing them to be ineffective. We developed an optimization process for the production of a high quality visualization. Our starting point was the theory of human contour perception which suggests that head-to-tail graphical elements which we call streaklets. We used Jobard and Lefer's algorithm to displaying streaklets along evenly spaced streamlines. An interactive interface was developed to give users control of 28 parameters that controlled the mapping between the data and the shape color and spacing of the streaklets. Through a process of "human in the loop hill climbing" 176 optimized representations of a flow pattern were generated with random starting points in the space of possible representation. Study participants who included designers from the Rhode Island School of Design and meteorologists and visualization experts. The results were then ranked into four categories from good to bad by independent designers and the findings have been incorporated into a package called FlowVis2D that can reliably and automatically produce superior representations of flow patterns. This package can be run on a server to automatically produce images for web mapping sites (e.g. NOAA's nowCOAST) or it can be used as an interactive design tool. Figure 1 shows an image of the atmosphere at 350 millibars.

Figure 1. Wind patterns at 350 millibars