Perception accuracy and user acceptance of legend designs for opacity data mapping in GIS

Abstract: In a GIS system, the need to encode geospatial data without standardized cartographic representations, such as population data, weather information, etc., on top of a map is common. This can be done with a layer on top of a base map, with the effect that the base map is partially or fully hidden. Reducing the opacity (or its equivalent - increasing the transparency) of this overlay data layer is a frequently seen solution to show the geographic context. With the colours of the base map being combined with the overlay layer’s colours, the resulting visualization can become difficult to interpret for the end-user. To help the user decode the values encoded in maps, legends are a common tool for non- interactive maps and data visualizations. This user study investigates the decoding accuracy using a map without a legend, as well as 4 different legend designs for opacity-mapped data overlayed on a static base map. A secondary objective is to measure how helpful the users considered the different legend designs were to decode values. Baseline categories for comparison were (i) no legend - only having the range of the data values being expressed in text and (ii) a legend design imitating the ArcGIS legend for opacity data mapping. Three different legend designs were produced to introduce more contextualisation from the map background to the background of the legend and reducing the distance from the legend to the data. This was done using (iii) a sample of the map as background for the legend (iv) having the most common colours of the map base layer as legend background (v) attaching a legend directly to the edge of the overlay data area. Using a web interface, the users were requested to visually estimate the value at the location of a marker within the overlay data area. In statistical analysis of the results, there was clear statistical effect in reduced errors when having a legend compared to when no legend was included. There was, however, no statistically significant difference in estimation/perception errors between the legend designs tested. The acceptance of respondents - defined as how useful they considered the legend types were to help estimate the value - did have statistically higher estimates when sampling the map background (marked as iii when introduced in text above) and when attaching the legend to the data area (v) compared to the default ArcGIS design (i).