Exploring essential variables in the settlement selection for small-scale maps using machine learning

Abstract. The decision about removing or maintaining an object while changing detail level requires taking into account many features of the object itself and its surrounding. Automatic generalization is the optimal way to obtain maps at various scales, based on a single spatial database, storing up-to-date information with a high level of spatial accuracy. Researchers agree on the need for fully automating the generalization process (Stoter et al., 2016). Numerous research centres, cartographic agencies as well as commercial companies have undertaken successful attempts of implementing certain generalization solutions (Stoter et al., 2009, 2014, 2016; Regnauld, 2015; Burghardt et al., 2008; Chaundhry and Mackaness, 2008). Nevertheless, an effective and consistent methodology for generalizing small-scale maps has not gained enough attention so far, as most of the conducted research has focused on the acquisition of large-scale maps (Stoter et al., 2016). The presented research aims to fulfil this gap by exploring new variables, which are of the key importance in the automatic settlement selection process at small scales. Addressing this issue is an essential step to propose new algorithms for effective and automatic settlement selection that will contribute to enriching, the sparsely filled small-scale generalization toolbox.

The main idea behind this research is using machine learning (ML) for the new variable exploration which can be important in the automatic settlement generalization in small-scales. For automation of the generalization process, cartographic knowledge has to be collected and formalized. So far, a few approaches based on the use of ML have already been proposed. One of the first attempts to determine generalization parameters with the use of ML was performed by Weibel et al. (1995). The learning material was the observation of cartographers manual work. Also, Mustière tried to identify the optimal sequence of the generalization operators for the roads using ML (1998). A different approach was presented by Sester (2000). The goal was to extract the cartographic knowledge from spatial data characteristics, especially from the attributes and geometric properties of objects, regularities and repetitive patterns that govern object selection with the use of decision trees. Lagrange et al. (2000), Balboa and López (2008) also used ML techniques, namely neural networks to generalize line objects. Recently, Sester et al. (2018) proposed the application of deep learning for the task of building generalization. As noticed by Sester et al. (2018), these ideas, although interesting, remained proofs of concepts only. Moreover, they concerned topographic databases and large-scale maps. Promising results of automatic settlement selection in small scales was reported by Karsznia and Weibel (2018). To improve the settlement selection process, they have used data enrichment and ML. Thanks to classification models based on the decision trees, they explored new variables that are decisive in the settlement selection process. However, they have also concluded that there is probably still more “deep knowledge” to be discovered, possibly linked to further variables that were not included in their research. Thus the motivation for this research is to fulfil this research gap and look for additional, essential variables governing settlement selection in small scales.