Evaluating map specifications for automated generalization of settlements and road networks in small-scale maps

Automatic generalization is an essential technological step in the development of modern cartography. It can be a response to the legally imposed obligation to automate the process. An additional motivation is the growing demand of the information society for up-to-date and available data and maps at various detail levels. Currently, despite the strong computerisation and automation in many fields, effective solutions for the automation of selection have not been presented in the scientific community. Until now, scientists have devoted little attention to small-scale maps, which are updated less frequently.

Even partial replacement of the cartographer’s work with an automated generalization algorithm would significantly reduce the time required to design a map, the costs and the number of people involved. These challenges have been known to both the scientific community and public administration for a long time. Benefits from automation have been noticed, and the small-scale map specifications have been defined by Polish law, in the form of Geodetic and Cartographic Law (1989.) and Regulation on the Topographic Objects Database and General Geographic Objects Database, as well as standard cartographic products (MSWiA 2011). However, these map specifications have not been verified so far. The motivation of this paper is thus to fulfil this gap by evaluating map specifications concerning small-scale maps contained in the Regulation. The article presents the formalization and implementation of the rules from the Regulation. This is the first stage in developing a comprehensive solution for automatic small-scale map generalization in Poland.

The issue of generalization has been undertaken by research centres, national mapping agencies and commercial companies for years. The subject of this research is very often settlements and road networks, because of the key role they play in the maps.

Geodetic and Cartographic Law imposes the obligation to prepare and update maps on the Surveyor General of Poland, and thus he is also responsible for automating this process. Regulation, on the other hand, contains a set of map specifications defining the mode and technical standards of map design in a basic scale series. In particular, it includes the principles of selecting objects, resolving graphic conflicts and a collection of cartographic symbols together with their priorities. The order of information provided in annexes to the Regulation

suggests the following generalization stages:

Small-scale maps are designed based on the General Geographic Object Database (GGOD), whose detail level corresponds to the 1:250 000 scale. Settlement point layer, in small-scale maps, contains objects selected from GGOD’s class of settlement point features (OT_ADMS_P). The Regulation contains detailed guidelines and object selection principles. At the selection stage at 1:500 000 scale, it is necessary to take into account the population density of the district in which the settlement is located. At the detail levels corresponding to 1:500 000 and 1:1 000 000 scales, six types of settlements are distinguished based on the type of settlement, the administrative significance and the number of inhabitants (population).

Roads contained in the General Geographic Objects Database are obtained from the road class (OT_SKDR_L), from the thematic group roads and related objects. Selection of the most important roads, which should be presented on the small-scale maps, is given in the principles contained in the Regulation. It is based on the following attributes: road class, number of lines, the category of existence and type of geometric representation.

At 1:500 000 scale, nine types of roads are distinguished, while at 1:1 000 000 scale, six types of roads are considered. Each road type is represented with a different symbol.

This research aimed to verify and evaluate the correctness of generalization process carried out on the basis of map specifications from the Regulation

Regulation of the Ministry of Interior on 17 November 2011 on the Topographic Objects Database and General Geographic Objects Database, as well as standard cartographic products, Journal of Laws of 2011, No 279 item 1642.

Based on the principles contained in the Regulation, generalization models have been designed for each thematic layer and detail level considered. It allows generalization to be carried out nearly fully automatically. The processes include such steps as: data enrichment, selection and symbolization. Enriching the data includes adding additional attributes to the source data, the value of which is important in the selection. The selection stage consists of selecting objects using the given algorithm with specific parameters. Symbolization consists of assigning objects with an appropriate symbol, as defined in the Regulation.

The process was carried out to generalize GGOD for the area of Poland, while the test area was constrained to the border area of the Mazovian, Lublin and Świętokrzyskie Voivodeships (Figures 1 and 2). The Vistula River crosses it from the South to the North. The size of the area is 47 km (West-East) by 55 km (North-South).

Figure 1

Figure 2

The research area includes 2585 km², and within its borders there are 550 settlements and 604,5 km roads in GGOD.

The generalization process was carried out in accordance with the principles contained in the Regulation. In order to assess the quality and correctness of the generalization results, they were compared with reliable, manually designed small-scale maps and databases elaborated at the corresponding scales. The verification process consisted of visual inspection of objects presented in the results and also shown in the reference materials, namely atlas maps. Based on the visual results inspection, quantitative analyses have been carried out and summed up in tables 3-8.

The results obtained were verified by comparing them with maps created by experienced cartographers; these maps in the study are called reference materials. In the case of the 1:500 000 scale, the following reference maps were considered:

The digitized settlements and road network from the reference maps are shown in Figure 6.

Figure 6

The result obtained was assessed visually by the authors, and then compared with well-designed paper maps. Uneven distribution of the settlements on the resulting map is visible. During the generalization, the uniformity of the arrangement of the objects and the density of the settlements is not taken into account, while it is very important in the case of the manual map design done by the cartographer.

In the resulting map, the boundaries of districts are clearly visible because of the different settlement densities. For example, in Lipski District, there is a higher concentration of settlements (Figure 7). The reason is the low population density in this district (47 people / km2), which is taken into account as a key factor in the settlement selection process according to the map specifications contained in the Regulation. While in neighbouring districts, with higher population density, the selection process is stricter. This situation results in exceedingly significant diferences among neighbouring districts.

Figure 7

In order to evaluate the generalization results, the authors have also conducted a quantitative analysis (Figure 8). In Tables 7-9, the result of the comparison of the number of settlements presented in the reference maps and databases is presented. Table 5 shows all the combinations of the presentation of the settlements. The first row shows the settlements shown on all maps, the second row is the settlements omitted in EuroGlobalMap and presented on the remaining ones, etc. The highest value was assigned to the settlements that were shown on both reference maps and omitted in the generalization result. For these settlements, regularities were sought according to which of them, despite the small number of inhabitants, were selected and displayed on the map. Their primary functions, as well as the distance to the nearest neighbour, were also calculated. As a result, no regularities were found by examining the attributes of objects. However, their presence on the target map seems to be necessary for correctness and aesthetic reasons. It is difficult to indicate the objective factors determining the choice of these settlements. It was likely decided by the subjective assessment of the cartographer and supported by his experience.

Figure 8

In the case of roads, the network in the resulting map is unevenly dense, and there are cul-de-sacs. While comparing it to the manually designed reference maps, it is visible that some connections are missing and there are places where the road density is too high according to this detail level. In the case of a map from the Atlas of the Republic of Poland, it can be seen that the terrain situation has changed and it is out of date, as there are no roads that were built after 1997.

In summary, GGOD generalized to the detail level corresponding to the scale 1:500 000 has 112 settlements, while according to the well known Topfer and Pillewizer (1966) Radical Law, this layer should contain 391 objects

with 553 objects in 1:250 000 scale, in generalization to 1:500 000 scale

In the case of the 1:1 000 000 scale, the following reference map and database were considered:

Digitized settlements and road network from the maps are shown in Figures 9 and 10.

Figure 9

Figure 10

The small number of settlements and the strict selection criteria made the result of the generalization seem more correct than in the 1:500 000 scale. Particularly interesting is the comparison with EuroGlobalMap. The database corresponding to the small-scale detail level was derived from the databases, whose detail level corresponds to the 1:250 000 scale. This process was conducted in a fully automated mode. There are very few settlements presented in this database. According to the map from the geographical atlas, not enough settlements were presented. Therefore, this database should also be improved regarding settlements selection.

The number of objects as a result of the selection is comparable to the number of settlements on the Demart map developed by the cartographer (Table 10 and Figure 10). This indicates a relatively correct density of settlements. In Table 11 it can be seen that only a few settlements have been presented on one of the maps. Only one settlement was presented on the EuroGlobalMap and omitted in the atlas map and generalization result (marked with an asterisk in Table 12). What is more, six settlements were shown only in the Demart atlas (marked with an asterisk in table 12). An interesting case is Bochotnica, presented on both the atlas map and EuroGlobalMap, while it was omitted in generalization result. This town is essential from the communication point of view because it is located at the intersection of provincial roads.

In conclusion, in GGOD generalized to the detail level corresponding to 1:1 000 000 scale, 18 settlements were presented, while according to Topfer and Pillewizer (1966) Radical Law, this layer should contain 276 objects

with 553 objects in 1:250 000 scale, in generalization to 1:1 000 000 scale

The result of automatic generalization difers to some extent from the content developed by cartographers and shown on the manually designed maps. However, these differences are not very significant. At the 1:500 000 scale, settlements selected on the basis of map specifications contained in the Regulation are not uniformly distributed. This is due to the conditions of district density set during the selection process. It suggests that the generalization principles contained in the Regulation should be modified and supplemented.

In the case of roads, the biggest problem is the selection of local roads, which should complement the network and maintain a connection with the settlements shown on the map.

Automatic road selection at 1:1 000 000 scale, due to simple and formalised selection criteria, can be considered as relatively correct. There are few differences between the generalization result and the reference maps designed manually by a cartographer.

The values collected in Table 13 are the confirmation that regional and national roads in all analysed sources were chosen fairly. Among municipal and district roads, many more roads were selected and presented at EuroGlobalMap.

The verification of map specifications leads to the conclusion that in the future additional selection criterions should be added. After comparing the results of automatic generalization with the reference materials, it is possible to conclude that considering more geometrical and semantic settlements and road characteristics may contribute to improving the quality of the objects selection.

In the case of settlements, an important additional criterion for selection can be the relation to neighbouring objects; for instance, the proximity of the river, road network nodes or a railway line (in particular the stations). Features such as the function of the settlement and its distance to the nearest neighbour, as well as the settlements density, may play an equally important role.

In the case of road network generalization, the essential aspect that should be considered is road network connectivity, so it is necessary to take into account the neighbourhood of the settlements, bridges, harbours and railway stations. It is also important whether other roads are performing the same connectivity function, as well as if the network density is correctly represented on the generalized map.

The research conducted has proven that it is impossible to determine generalization parameters that would allow for obtaining correct results throughout Poland, therefore smaller units should be considered for this purpose. However, they should not necessarily be based on administrative boundaries. What is more, the algorithms for automatic generalization of the road network and settlements should take into account the relationships between objects from diferent layers. At the same time, manually designed maps are and should be material for checking the quality and correctness of automatic generalization results.

While conducting the research, the incompleteness and need to enrich the GGOD structure was found. Based on the analysis of the content of the EuroGlobalMap settlement layer, it can also be concluded that the generalization algorithms used to obtain this layer should also be improved. The lessons learned from this and future research could be used to enhance the quality of both GGOD and EuroGlobalMap databases. Last but not the least, it can be stated that the map specifications contained in the Regulation need to be complemented and improved to be able to generalize maps automatically.