Spatial parameters for transportation: A multi-modal approach for modelling the urban spatial structure using deep learning and remote sensing

Abstract

A significant increase in global urban population affects the efficiency of urban transportation systems. Remarkable urban growth rates are observed in developing or newly industrialized countries where researchers, planners, and authorities face scarcity of relevant official data or geo-data. In this study, we explore remote sensing and open geo-data as alternative sources to generate missing data for transportation models in urban planning and research. We propose a multi-modal approach capable of assessing three essential parameters of the urban spatial structure: buildings, land use, and intra-urban population distribution. Therefore, we first create a very high-resolution (VHR) 3D city model for estimating the building floors. Second, we add detailed land-use information retrieved from OpenStreetMap (OSM). Third, we test and evaluate five experiments to estimate population at a single building level. In our experimental set-up for the mega-city of Santiago de Chile, we find that the multi-modal approach allows generating missing data for transportation independently from official data for any area across the globe. Beyond that, we find the high-level 3D city model is the most accurate for determining population on small scales, and thus evaluate that the integration of land use is an inevitable step to obtain fine-scale intra-urban population distribution.

Empiric recommendations for population disaggregation under different data scenarios

“…Another consideration is that there are significant differences in the errors based on the size, population density and land use of the spatial unit. This corresponds with previous studies that also found relationships between higher errors and lower percentages of residential land use and higher mixed or other land uses [11] and population density [21]. Therefore, the PLOS ONE physical configuration of an urban landscape affects the performance of the disaggregation methods.…”

“…Therefore, when VHR data are available, the dasymetric method is found to be the best option: the accuracies are better, it takes less computational effort, and it is more straightforward and easier to apply. When land use data are available, the categorical dasymetric approach is found to provide the best results: accuracies improve considerable if the number of stories is available, confirming results of Stiller et al [11].…”

“…Within this group, three main methods can be conceptualized, which are explained in more detail in methods section. The dasymetric method employs geospatial auxiliary variables to allocate the population, which can be binary if it relies on built-up areas or volume [11,16,17,20,21,37], categorical if it considers the land use [20,38,39] or it can be based on the built-up density [21,40]. A recent contribution is Disaggregator, a tool that automatizes and facilitates the dasymetric disaggregation of geodata [41].…”

“…On the other hand, Wan et al [46] examined the performance of alternative sources of ancillary data, including imperviousness, land cover, road networks and night-time lights, on the accuracy of the dasymetric method. Only two studies performed a multi-fold validation, evidencing a high influence in the measured accuracies [11,21]. However, none of these studies made a complete comparison of all the factors influencing the evaluation of disaggregation methods.…”

Deep Learning-Based Generation of Building Stock Data from Remote Sensing for Urban Heat Demand Modeling

“…Data on land-use for the identification of residential buildings can be substituted for other European countries with data from the European Urban Atlas (https://land.copernicus.eu/local/urban-atlas). For other countries, land-use data from the OpenStreetMap project have been successfully incorporated to assign land-use to buildings [63]. Another crucial data set for modeling of the energy demand, however, is the construction period of the building.…”