To understand the dynamics of land use changes in the landscape, it is necessary to process available data from the past and reconstruct the landscape and its development based on these data or model the development of the landscape in the future. Many different sources of historical data allow us to reconstruct land use to a certain time horizon; in general, these sources can be divided into spatial and statistical sources. As historical spatial data sources, aerial or satellite archival imagery (Badjana et al., 2015), old medium-scale maps (Eremiášová and Skokanová, 2009; Gimmi et al., 2011; Skaloš et al., 2012; Lathouwers et al., 2023), and, less frequently, large-scale maps (Bender et al., 2005; Bürgi et al., 2015) are used to analyse land use changes. As statistical data, written documentation of land cadastres can be used (Bičı́k et al., 2001), or various statistical lexicons and yearbooks can contain information on the structure of land use in each territorial unit. If this statistical information is available for different time horizons, trends and changes in the structure of land use can be observed, but the spatial distribution within a given territorial unit is missing. Many studies use a combination of multiple types of spatial data within a single study (Cousins, 2001; Petit and Lambin, 2002; Hamre et al., 2007; Skaloš and Engstová, 2010; Pindozzi et al., 2016; Chen et al., 2019), usually a combination of old maps for the earliest time horizons and satellite or aerial imagery for more recent periods or field surveys for the current state of the landscape to provide a complete understanding of land use change.
The type of spatial data to choose for a given study is influenced by many factors: the availability of data for a given area and the required time period, and the purpose of the study—to monitor the change in the area as a whole or only a selected phenomenon. Tracking a selected phenomenon is applied, for example, in a study (Istrate et al., 2023) in which the evolution of forests is monitored. The evolution of ponds is investigated (Pavelková et al., 2016), and the evolution of water bodies and wetlands is investigated (Gimmi et al., 2011; Šantrůčková M. et al., 2017). Other factors influencing the selection of input data include the possibility of linking with other data, e.g., demographic, geological, or socioeconomic data; the chosen methodology and tools used in processing and analysing the input data (whether raster or vector models will be analysed); and, last but not least, the desired outputs and interpretation of the analysis results. When studying the evolution of land use, the supporting information is obtained from the planimetric component of the old map, but in some cases, it may also be appropriate to process the altimetric component of the map and to look at the change in relief in the area, or a combination of both, that is, to look for a relationship between land use change and relief change (Tortora et al., 2015; Statuto et al., 2017; Istrate et al., 2023).
The subject of this study is old large-scale maps, which, with their detailed spatial information and fine resolution, are valuable sources of information for the study of land use change over time, providing detailed information on the landscape, land ownership and land use patterns. Old, large-scale maps allow researchers to conduct a temporal analysis of land use change over decades to centuries, which is an advantage over aerial and satellite imagery that does not go as far back in time. Large-scale maps allow land use development to be tracked at the individual parcel level, which allows for additional tracking data, such as landowner information, to be linked to individual parcels. The socioeconomic context of land use change is thus reflected in the results of the analysis, as changes in land ownership patterns can be related to changes in land use. Old large-scale maps are particularly useful when studying urban growth and expansion; they can show how cities have evolved, where new development has occurred and how land use has changed over time in urban areas (Scăunaș et al., 2019).
The vector models of the territory created on the basis of old large-scale maps contain geometric objects (points, lines, polygons), their interrelations and properties, and serve not only for the visualisation of the territory at the time corresponding to the creation of the original map but also for spatiotemporal analysis (it becomes one of the layers of the resulting multitemporal GIS). The vectorization process, in this study understood as the transition from scanned georeferenced map bases to vector models, can generally be performed manually, semiautomatically, or automatically. The appropriate method must be chosen considering the type, quantity and quality of the input map data and the available software options. Manual vectorization is completely under the control of the user and can be performed in standard GIS software without special knowledge of the software; the raster base may be of poorer quality, but this method is time-consuming for large volumes of data. Semiautomatic vectorization is also under user control, and the user must also control the vectorization process in this case; however, using special software tools, snapping to the raster or automatic tracking of the raster cells with subsequent vector generation is used. Automatic vectorization consists of automatic vector generation, which offers the possibility of using machine learning and neural networks and can be used to vectorize even larger volumes of data very efficiently. The potential use of automatic vectorization of the content of digitised old maps or part of it is discussed in (Iosifescu et al., 2016; Chiang et al., 2020; Kratochvílová and Cajthaml, 2020).
There are two approaches for vectorizing the time series of old maps (Skokanova, 2008). The first approach, called sequential vectorization, consists of creating historical vector models from different time horizons independently. The advantage of this method is the simplicity of the processing and the possibility of creating multiple models simultaneously. The disadvantage is that each vector model carries its positional inaccuracy, which causes inaccuracies in overlay operations. The second approach, called reverse vectorization, establishes one of the maps as a reference, usually the current map, which should represent the most accurate map source. The vector model created from this map is the basis for the vectorization of the next period map. The other model is only adjusted if there is a real change from the reference map and not a change due to the positional inaccuracy of the two map sources. Thus, the old maps are used here only to identify changes, e.g., those used in (Bender et al., 2005). The advantage is the elimination of mutual positional inconsistencies of vector models and the speeding up of processing, especially in the case of smaller and simpler areas. The disadvantage may be the uncertainty in assessing whether there is a real change or an error caused by positional inaccuracies.
In general, vector models used for land use analyses are burdened by positional and thematic errors. Thematic errors consist of incorrectly assigned land use categories or incorrectly stated land use categories in the map source. The positional accuracy of old maps of various scales concerning their use in historical landscape research is addressed in (Cajthaml and Krejčí, 2008; Frajer and Geletič, 2011). Positional errors are caused by the inaccuracy of the old map and by errors arising during individual processing processes (scanning, georeferencing, and vectorization). During overlay operations, these errors lead to the formation of sliver polygons. The presence of sliver polygons, small polygons, and narrow residual polygons resulting from positional inconsistencies in the mapping sources can cause a problem in accurately capturing land use changes. These polygons can lead to false positive interpretations of changes and compromise the reliability of the analysis (Mas, 2005; Skokanová, 2008; Clercq et al., 2009).
This study focuses on the verification of the positional accuracy of large-scale map sources on the example of model areas in the surrounding of the Vltava River and methods for solving the problem of sliver polygons that appear when overlaying vector models and examines the influence of these residual polygons on the results of the analyses and the possibility of their elimination.