Doctoral Colloquium of the DGPF Working Group 'Geoinformatics – Methods'

Maike Gatzlaff (University of Auckland, New Zealand)

Movement data are often either not shared or shared in unregulated and unscrutinised ways. While geomasking methods have successfully been established for location data, no suitable anonymisation method for movement data has been established yet. In addition, the complex nature of movement data is challenging and results in high re-identification risks after anonymisation. To contribute to the advancement in trajectory anonymisation methods, this project aims to answer the need for context-dependent anonymisation algorithms that maintain the utility of trajectory data. Current anonymisation methods are applied to the same movement dataset at varying levels of granularity. Privacy preservation and utility of anonymisation results will be assessed based on well-established anonymisation metrics and newly developed metrics focusing on individual trajectories. Based on this evaluation, a new anonymisation algorithm will be developed – focusing on significant locations and critically assessing the role of synthetic data in preserving privacy around the most sensitive information. The resilience of the anonymised trajectories will be tested by attacking the data before data utility is evaluated by applying both raw and anonymised movement data to example applications. By reliably anonymising movement data and minimising the trade-off between privacy preservation and data utility, more human movement data will be releasable for the public good.

Constantin Meyer (ARL, Hannover, Germany; University of Würzburg, Germany)

Geographic Information Systems (GIS) offer a wide range of possibilities for processing, analysing and visualising spatial data. They thus represent valuable support options for complex planning processes, especially at the supra-local/regional level. When these applications are designed in such a way that they are tailored to concrete planning and decision-making processes, they are referred to in the literature as "Planning Support Systems" (PSS). PSS may also feature a normative dimension by proposing concrete planning specifications based on methods of multicriteria decision analysis (MCDA). This ongoing PhD-project deals with the enhancement and expansion of an existing supra-local planning instrument, the so-called Bavarian “Alpenplan”, a supra local zoning concept of the Bavarian Alps, which is legally binding and in force since 1972. It involves the development of a cross-sectoral GIS-based assessment and decision-making model that is intended to simulate the process of weighing relevant planning requirements and criteria from the perspective of supra-local spatial planning, forming a PSS for this specific planning purpose of producing future zoning proposals. The application orientation of the PSS shall be achieved by a modular structure, flexible weightings and the development as an interactive WebGIS.

The presentation will in particular bring up the following aspects for discussion:

• Availability of suitable geodata for spatial planning requirements
• Choice of specific raster resolution for geoprocessing
• Methods for weighting of criteria
• (Geo-)statistical challenges (model sensitivity, multicollinearity, ..)

Oliver Huber (TU Dortmund University, Germany)

In the context of climate change, urban flood hazards are increasing. Hence, the demand for flood hazard risk analysis is increasing as well. For the creation of flood hazard maps, high resolution digital terrain models (DTMs) are required, as well as detailed knowledge about simulation methods. Furthermore, DTMs must be revised by experts to remove hydrological errors. Due to these circumstances, flood hazard maps are not as widely available as they need to be to contribute to climate change adaption. In countries of the global south, the situation is even more serious. Data and expert knowledge are often unavailable. Furthermore, increasing wealth and urbanization lead to an aggregation of material value and human lives in confined spaces, which increases the potential for damage.

The aim of the work presented is to develop and evaluate new methodological approaches in the collection of geo data for DTMs and new methods to simplify the generation of flood hazard maps and thereby facilitate access to hazard information.

To achieve this goal, models based on neural networks are being developed and trained to find DTM-errors and to predict flood hazards. For the data collection in the field, especially in areas, where no high resolution data is available, new low cost devices (such as I-Pads) that can be handled by untrained staff are being evaluated.

To test the applicability of the new approaches, sample areas in North Rhine-Westphalia, Germany, and Manila, Philippines will be investigated. In the German sample areas, high resolution data in large quantities is available, which is used to train the neural networks. The Manila sample areas on the other hand provide a good example for the challenges given in global south countries and also offer the opportunity to test the new models in areas with different settlement patterns and inconsistent data quality.

Jiarui Qin (Nanjing Normal University, China)

This research explores the urban vitality of Nanjing's main urban area by examining its economic, social, cultural, and innovation aspects. With the development of smart cities, innovation has become a crucial component of urban vitality. The research first constructs an urban vitality index system and a built environment index system based on multisource data. It then uses the suitable entropy method to evaluate the urban vitality of Nanjing's main urban area comprehensively. A stepwise regression model is established to select significant variables from the built environment index system, and linear and multi-scale geographically weighted regression models are established to analyze the relationship between the built environment and the vitality of the main urban area of Nanjing.

The results show that Nanjing's central urban area has a multi-centered comprehensive vitality that decreases outward from the city center. The overall vitality of the southern region is higher than that of the northern region. The analysis identifies several factors that have significant effects on urban vitality, including road density, building density, the average number of buildings, and the number of business circles. The multi-scale geographically weighted regression model has the most spatial explanatory power and reveals that the positive effect of building density is higher north than south of the Yangtze River. The analysis also shows an increasing quantity of business districts from south to north and a decrease in road density from the center to the periphery of the city.

Overall, this research provides a multidimensional perspective on the spatial distribution pattern of urban vitality in the main urban area of Nanjing. The analysis highlights the importance of accurate planning and improved city image to enhance urban vitality and the relationship between urban vitality and the built environment. The findings can inform urban planning and policy-making to promote economic, social, cultural, and innovative development in urban areas.

Lina Budde (Technical University of Darmstadt, Germany)

In remote sensing the creation of land cover maps is a basic task. Using deep learning models for this application is today’s state-of-the art. The quality of such methods depends highly on the quantity and quality of the used data. While for everyday images large databases with reference data for different tasks and environment exist, the available reference data of remote sensing data is very rare due to the high variations of spatial or spectral variations. In addition, remote sensing data is often pre-processed in some way e.g. atmospheric correction, pan-sharpening etc. For better understanding of the deep learning process and also to enable quality improvements, it is important to extract quality information of the input data by the deep learning process. Therefore, it is important to separate the different uncertainty sources which can affect the quality of the results of the deep learning model, especially the data and model uncertainty.

A general approach for determining the model quality is the Monte-Carlo dropout. To investigate the data quality, the features maps are analysed regarding to anomalies. To identify anomalous regions normally specialized deep learning models are used, but so far only few datasets exist for training process, which mainly contains industrial applications. Due to this fact my current research question is, how can such a model be transferred to the remote sensing domain and included in the semantic segmentation process?

Chhavi Arya (TU Kaiserslautern, Germany)

Information and communications technologies are becoming increasingly diffused within the material spaces of the city, generating novel ways of representing complex, hitherto 'invisible' urban behaviours. The emerging new sources of data have the potential to capture dimensions of social and geographic systems that are difficult to detect in traditional urban data (eg. census data). The rapid pace of urbanisation, complex urban dynamics and the necessity for improved urban operations put thrust on the need to explore new data sources for their potential in urban decision-making processes.

The dissertation aims to bring out the possibilities of urban analytics using user-generated data. By integrating the traditional sources of data with the dynamic datasets can give an understanding of human behaviour patterns in urban spaces and help in identifying problems hidden in current urban form. The research focusses on the spatio-temporal dynamics of human activity in cities specifically with regard to baby boomer generation (born between 1955-69). The main objective is to provide cities with a new means of exploration and analysis through the incorporation of user-generated data into city analytics, and to explore the potential of emerging data sources on the understanding of cities and their dynamics with regard to the ageing population.

In association with the aim and objective of the thesis, the main research question is: How to facilitate urban analytics through the integration of human activity dynamics surfacing from user-generated data?

The dissertation is being carried out as a part of the Carl-Zeiss funded project: 'Ageing Smart – Designing Spaces Intelligently'

Lisa Beumer (Forschungszentrum Jülich, Germany)

Nuclear safeguards require the application of a support systems to collect retrieve and analyze large amounts of information retrieved from different sources. Satellite imagery represents a key source of information for the implementation and verification of nuclear nonproliferation agreements.

The quality and quantity of imagery data is increasing as rapidly as the methods for analyzing the data sets. The growing archives of satellite imagery are providing novel insights for detecting changes at nuclear facilities worldwide. However, this is accompanied by challenges regarding an appropriate and timely use of this imagery data. Currently, Deep Learning (DL) approaches, such as Convolutional Neural Networks (CNNs) in particular, achieve cutting-edge accuracies to address Big Data challenges in a more efficient way.

This paper demonstrates a workflow for the analysis of Sentinel-2 data collected by the European Commission's Copernicus Programme for the purpose of detecting changes at nuclear facilities. Such change detection analysis can serve as a reference to support on-site inspection planning, detect and monitor changes and activities at nuclear facilities and also to verify the completeness and accuracy of information provided by a treaty participant. The datasets are used to investigate the application of two Fully Convolutional Neural Networks, the U-net [1] and the Residual U-net (ResUNet) [2]. Since the datasets and the software are freely available this opens new possibilities. Applying this methodology shows that the accuracy of change detection can be improved and could therefore add a big value in the safeguards verification process.

[1] Ronneberger, O., Fischer, P., Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Navab, N., Hornegger, J., Wells, W., Frangi, A. (eds) Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. MICCAI 2015. Lecture Notes in Computer Science(), vol 9351. Springer, Cham. 10.1007/978-3-319-24574-4_28

[2] Diakogiannis, Foivos & Waldner, Francois & Caccetta, Peter & Wu, Chen. (2020). ResUNet-a: A deep learning framework for semantic segmentation of remotely sensed data. ISPRS Journal of Photogrammetry and Remote Sensing. 16. 94-114. 10.1016/j.isprsjprs.2020.01.013.

Egor Kotov (Max Planck Institute for Demographic Research, Rostock, Germany)

Tiger mosquitoes have been present in Spain 2004 (Aranda et al., 2006; Collantes et al., 2015). This mosquito species presents a significant threat to the population, as it is a vector of diseases such as Dengue, Chikungunya and Zika. Due to socioeconomic segregation, exposure to mosquito bites and the associated infection risks will likely affect various populations disproportionately. This preliminary work aims to evaluate these exposure inequalities.

In Spain, the last decade was characterized by an increased income inequality due to income concentration at the top of the income distribution (Martín-Legendre et al., 2021). The increasing segregation and income inequality may lead to an increased burden on already disadvantaged groups. For example, vulnerable populations are already disproportionately exposed to air pollution exceeding the maximum permitted levels (Moreno-Jiménez et al., 2016). Similarly, these groups are also likely to be disproportionately affected by mosquito presence.

This study aims to evaluate mosquito exposure inequalities by sex, income and activity space in densely populated Spanish functional urban areas at high spatial resolution. We will compare mosquito report probability (or report propensity score (Palmer et al., 2017)) with the income Gini coefficient and gender-age structure of the population at the census district level. Taking into account the activity space allows us to go beyond the simplified understanding of socio-spatial segregation only through residential location (Müürisepp et al., 2022) and evaluate the segregation based on the range and available amenities, as well as the overlap of activity space of groups with different income. Fig. 1 below provides an example based on open mobility data (MITMA, 2022) where higher and lower-income groups' activity space is significantly smaller (though probably for different reasons) in morning rush hours compared to a medium-income group. These differences in mobility patterns between income groups may contribute to different exposure to mosquito-related risks and the ability to avoid them.

References:

Aranda, C., Eritja, R., & Roiz, D. (2006). First record and establishment of the mosquito Aedes albopictus in Spain. Medical and Veterinary Entomology, 20(1), 150–152. 10.1111/j.1365-2915.2006.00605.x.

Collantes, F., Delacour, S., Alarcón-Elbal, P. M., Ruiz-Arrondo, I., Delgado, J. A., Torrell-Sorio, A., Bengoa, M., Eritja, R., Miranda, M. Á., Molina, R., & Lucientes, J. (2015). Review of ten-years presence of Aedes albopictus in Spain 2004–2014: Known distribution and public health concerns. Parasites & Vectors, 8(1), 655. 10.1186/s13071-015-1262-y.

Martín-Legendre, J. I., Castellanos-García, P., & Sánchez-Santos, J. M. (2021). Neighborhood inequality and spatial segregation: An analysis with tax data for 40 Spanish cities. Cities, 118, 103354. 10.1016/j.cities.2021.103354.

MITMA. (2022). Estudio de movilidad de viajeros de ámbito nacional aplicando la tecnología Big Data. Informe metodológico (National Level Traveler Mobility Study Using Big Data. Methodological Report).

Moreno-Jiménez, A., Cañada-Torrecilla, R., Vidal-Domínguez, M. J., Palacios-García, A., & Martínez-Suárez, P. (2016). Assessing environmental justice through potential exposure to air pollution: A socio-spatial analysis in Madrid and Barcelona, Spain. Geoforum, 69, 117–131. 10.1016/j.geoforum.2015.12.008.

Müürisepp, K., Järv, O., Tammaru, T., & Toivonen, T. (2022). Activity Spaces and Big Data Sources in Segregation Research: A Methodological Review. Frontiers in Sustainable Cities, 4, 861640. 10.3389/frsc.2022.861640.

Palmer, J. R. B., Oltra, A., Collantes, F., Delgado, J. A., Lucientes, J., Delacour, S., Bengoa, M., Eritja, R., & Bartumeus, F. (2017). Citizen science provides a reliable and scalable tool to track disease-carrying mosquitoes. Nature Communications, 8(1), Article 1. 10.1038/s41467-017-00914-9.