Spatio-Qualitative Data Visualization: SoftGIS and Weighted Average Visualization

Veiw figure View Figure

Recent advances in spatial sciences and computer technologies have granted qualitative GIS with the possibility to be incorporated in the latest versions of computer-aided data analysis ^[12]. This has resulted in a considerable breakthrough in acquisition of geocoded qualitative data. The innovation of ‘SoftGIS’, developed by Marketta Kyttä and her team at Aalto University, refers to a collection of internet-based surveys (Figure 1) that allow the locality-based study of human experiences ^[6]. ‘SoftGIS’ provides a combination of ‘soft’ subjective data (qualitative) with ‘hard’ objective spatial data. SoftGIS has been capable of collecting large datasets for the use of urban planners interested in the development of more user-friendly physical settings ^[6].

With the considerable increase in professionals’ access to soft geocoded data, the need for a solid analysis methodology has become more significant; one that suits the specific characteristics of the spatio-qualitative data and has the potential to extract the desired information from the huge mass of data. There typically two approach in studying and analysis of spatio-qualitative data; one that uses narration as a means of conveying the desired information ^[14] and one that uses a quantitative basis to analyze the data ^[13].

This research is based on the later approach and attempts to reach knowledge discovery through visual analysis of quantified SoftGIS dataset. Accordingly, this paper proposes a new visualization approach which has the potential to reveal useful information from SoftGIS or any other similar spatio-qualitative datasets.

2. Spatial Data Visualization

Visualization is a major component of information perception. One can take advantage of his or her visual perception abilities in order to signify his or her cognition of the abstract data ^[10]. Moreover, visualization facilitates the understanding process of the users ^[7]. Therefore, it is important to benefit from the visualization techniques which best suit the data in order to amplify its most important underlying information.

The most common visualization methods used in geoinformatics are of maps and most people are familiar with them ^[9]. Moreover, in geoinformatics many types of thematic maps are also used. A thematic map can be used to emphasize the spatial pattern of one or more geographic attributes ^[11].

The potential users of SoftGIS data are urban authorities that are not necessarily experts in spatial sciences. Thus, the proposed methods not only must be informative, but also should be as simple as possible.

3. SoftGIS Data

The SoftGIS data was provided by Mapita Ltd. and it includes the recorded people’s experiences in Espoo and Helisnki region in Finland. The dataset originally consisted of six categories of records, three categories of positive and three categories of negative records (atmosphere, appeal, social). For the simplicity of the study, the categories were generalized into two major classes of positive and negative records.

4. Qualitative Data Quantification

Studying qualitative data is normally more challenging than quantitative datasets. Accordingly, in disciplines involved with qualitative values (e.g. psychological sciences), it is often a good practice to quantify the values in order to improve the computational and comparison capacity of the qualitative datasets ^[1]. Since qualitative datasets contain nominal values, the computational techniques widely used for quantitative dataset are typically inapplicable to them. Computationally-enabling the qualitative data allows the application of a diversity of visual and analytical techniques that can help to make the analysis more feasible. Therefore, in this study, in order to overcome the aforesaid computational limitations, a simple quantification is used. Accordingly, the values +1 and -1 are assigned to the nominal Positive and Negative quantities respectively.

Figure 2. Point representation of Positive and Negative experiences. From [8]

Download as

Veiw figure View Figure

5. Current Visualization Approaches

Kyttä et al. have proposed several methods for visual representation of SoftGIS data ^[8]. The simplest method used in the mentioned literature is a point map (Figure 2) in which the recorded experiences are divided into two classes of Positive and Negative and are demonstrated accordingly.

As it can be speculated in Figure 2, using a point representation of the recorded experiences has considerable limitations in conveying the desired information to the user. The first and most significant limitation is that in such a dense dataset it is not feasible to use point representation as there are too many overlapping features in the area. For instance one might assume that the marked region on the map dominantly contains positive records. However, as it can be observed in Figure 3, there are quite many negative records in that very same region as well, which are not visible in the first representation.

Figure 3. Negative records in Helsinki metropolitan area

Download as

Veiw figure View Figure

In addition, observing large masses of point clouds does not provide any clear image on the overall recorded experiences of the neighborhoods and its variation throughout the region.

Another visualization used in the same literature is a ratio map which represents the proportion of positive markings to all markings within a cell of a certain size (Figure 4). Although in this visualization the data is managed in a more proper way, it is yet not highly practical to use as it requires continual reference to the legend. Moreover, it does not provide any information about the actual amount of negative and positive markings in each cell and this can result in confusion.

Figure 4. The share of positive and negative place markings. From [8]

Download as

Veiw figure View Figure

Furthermore, another visualization method has been used in ^[8] that utilizes a Natural Neighborhood interpolation technique in order to create a continuous surface of the discrete markings (Figure 5).

Figure 5. The natural neighborhood visualization of residents’ positive (yellow) and negative (blue) experiences in a part of Espoo [8]

Download as

Veiw figure View Figure

Although this visualization looks more attractive, it has several limitations. In order to explore its limitations, the visualization was recreated in this study (Figure 6) ^[13].

Figure 6. The NN visualization of residents’ positive (green) and negative (red) impressions in Helsinki metropolitan area (Recreated)

Download as

Veiw figure View Figure

The first and most significant limitation of this visualization is that it can be misleading. That is because in some areas the projections may be based on only a few observations (Figure 7). In other words, as a result of interpolation there might be values for locations for which there are not enough observations indeed. Moreover, as a result of natural neighborhood interpolation a contradicting record (in comparison to the dominant experience in its surrounding) can significantly influence the overall result in an area.

Figure 7. Limitations of using NN interpolation in SoftGIS data visualization

Download as

Veiw figure View Figure

6. Weighted Average Visualization

The methods discussed to this point, were incapable of depicting a clear transition between overall experiences within areas. Moreover, for some purposes, such as urban planning, one may require a more precise understanding of the recorded locations. Thus we need to use a more location-based approach that provides a fair representation of all existing ideas.

The approach proposed and implemented in this study attempts to tackle the above discussed limitations by calculating a weighted average of markings within a predefined cell. Mathematically speaking, we are interested in calculating E (Experience) for every given cell as below:

Where n_n and n_p are the number of respectively negative and positive markings within the cell and negative and positive values are defined as the -1 and +1 respectively. Obviously, E is a continuous value in the range of [-1, 1] and the higher its value the more positive the overall experience would be.

The size and shape of the cells can considerably affect the visualization result. Obviously the larger the size of the cell, the less locationally precise it would be, though the visualization would become smoother. The optimum cell size in this case was considered as 300 meters. Finally the calculated E’s were visualized on a gradual scale as in Figure 8.

Figure 8. Weighted average impression map

Download as

Veiw figure View Figure

In Figure 8 we can see patterns which were not visible in the formerly discussed visualizations. For example in Helsinki city center, south of Kamppi (marked with a black oval in Figure 9), a highly negative experience can be observed. According to participants’ views this is mostly associated with the city center traffic congestions and dirtiness.

Figure 9. Negative impression pattern in Helsinki city center

Download as

Veiw figure View Figure