November 13, 2023

8:00 am - 10:30 am

8:00 am - 8:25 am

Identifying flash flood hotspots with explainable machine learning using urban features

Zhewei Liu, Tyler Felton, and Ali Mostafavi

Pluvial flash floods are fast-moving hazards and causes significant disruptions in urban areas. This study presents interpretable machine learning models for predicting urban flash flood hotspots based on intertwined land and built environment features. Various features related to land and built environment characteristics are constructed using diverse datasets, and the occurrences of flash floods are captured using crowdsource data from the events. Using these features and datasets, the flash flood hotspots of cities are predicted with two ensemble models based on decision trees. The results demonstrate that the models can achieve good accuracy in identifying flooded/non-flooded locations. The model interpretation results indicate that land features related to hydrological and topological features have greater impacts on flash flood risk, than built environment features. The data-driven machine learning models presented in this study provide a useful tool for predicting flash flood hotspots based on the intertwined features of land and the built environment in cities to enable nowcasting and proactive monitoring of flash flood hotspots for emergency response and also inform integrated urban design and development towards flash flood risk reduction.

8:25 am - 8:50 am

Multiscale Multifeature Vision Learning for Scalable and Efficient Wastewater Treatment Plant Detection using Hi-Res Satellite Imagery and OSM

Sukanya Randhawa, Guntaj Randhawa, Yuze Li, Olena Sivak, Johannes Zech, Maria Martin and Alexander Zipf

Filling data gaps in various global regions requires a robust approach that can accurately provide detection results from earth observation data. One of the challenges arises from significant heterogeneity in satellite images and variation in features and characteristics for specific ground objects like Wastewater Treatment Plants (WTPs).

To overcome these challenges, we propose a novel multiscale multifeature hybrid model. This model leverages the power of deep learning-based object detection models, namely Yolov6, RTMDET, EfficientDET, and Domain Adaptation, to accurately and efficiently identify WTP locations worldwide. Our approach focuses on performance enhancements, including reduced false positives (FPs) and broad coverage.

The strategies for achieving these improvements involve effective data processing approaches, model tuning, and adaptation. Moreover, we optimize training data features using Volunteered Geographic Information (VGI) data. We demonstrated the effectiveness of the suggested approach for three diverse global regions: Germany, France, and Malaysia. Our study gives new insights into WTP distribution when compared to existing databases like OpenStreetMap (OSM). The resulting pipeline delivers good results even in challenging rural and urban context. Moreover, it is well-suited for generating large scale WTP datasets, which is useful for many applications such as Critical Water Infrastructure mapping, Urban Planning, Climate Action and many more.

8:50 am - 9:15 am

Clustering Building Footprint Polygons Based on Graph Similarity Measures

Sophie Duong, Peter Rottmann, Jan-Henrik Haunert, and Petra Mutzel

Footprints of buildings can provide cues about architectural styles and functional types. Learning such latent thematic information from geometry is relevant for various applications, such as urban planning and map generalization. A common task in this context is to cluster a set of building footprints based on their shape characteristics. In this paper, we present a novel method for this task which is based on concepts of graph similarity. We use a graph similarity measure that combines ideas from the Weisfeiler-Lehman-method and optimal transport theory. For the final clustering, we use spectral clustering. To obtain a meaningful graph representation, we propose two algorithms that transform the medial axis of a building footprint into a skeleton graph. We tested our algorithm on a data set from Boston and performed a small user study, where we also compared the results to an existing feature-based clustering method. The study gives a first hint that the results of our algorithm are in line with human similarity perception. Future work is needed to improve the stability of the proposed similarity measure and to confirm our findings with more extensive experiments.

9:15 am - 9:40 am

Map Diffusion – Text Promptable Map Generation Diffusion Model

Marcin Przymus and Piotr Szymański

This paper introduces a novel text promptable map generation model, leveraging recent advancements in generative models. Promptable map generation has broad applications, democratizing access to geographic data, enhancing decision-making, improving communication, and enabling customization. Map Diffusion generates maps based on textual descriptions, allowing users to describe a region, and the model generates a corresponding map. We conduct a comprehensive review of related work, highlighting the unique contributions of our model. We also provide insights into dataset creation, model architecture, training procedures, and experimental results. This research marks a significant step in harnessing generative models for map generation, opening doors for future exploration in this field.

9:40 am - 10:05 am

CrashFormer: A Multimodal Architecture to Predict the Risk of Crash

Amin Karimi Monsefi, Pouya Shiri, Ahmad Mohammadshirazi, Nastaran Karimi Monsefi, Ron Davies, Sobhan Moosavi, and Rajiv Ramnath

Reducing traffic accidents is a crucial global public safety concern. Accident prediction is key to improving traffic safety, enabling proactive measures to be taken before a crash occurs, and informing safety policies, regulations, and targeted interventions. Despite numerous studies on accident prediction over the past decades, many have limitations in terms of generalizability, reproducibility, or feasibility for practical use due to input data or problem formulation. To address existing shortcomings, we propose Crash-Former, a multi-modal architecture that utilizes comprehensive (but relatively easy to obtain) inputs such as the history of accidents, weather information, map images, and demographic information. The model predicts the future risk of accidents on a reasonably acceptable cadence (i.e., every six hours) for a geographical location of 5.161 square kilometers. CrashFormer is composed of five components: a sequential encoder to utilize historical accidents and weather data, an image encoder to use map imagery data, a raw data encoder to utilize demographic information, a feature fusion module for aggregating the encoded features, and a classifier that accepts the aggregated data and makes predictions accordingly. Results from extensive real-world experiments in 10 major US cities show that CrashFormer outperforms state-of-the-art sequential and non-sequential models by 1.8% in F1-score on average when using “sparse” input data.

10:05 am - 11:00 am

Predicting urban tree cover from incomplete point labels and limited background information

Hui Zhang, Ankit Kariryaa, Venkanna Babu Guthula, Christian Igel, and Stefan Oehmcke

Trees inside cities are important for the urban microclimate, contributing positively to the physical and mental health of the urban dwellers. Despite their importance, often only limited information about city trees is available. Therefore in this paper, we propose a method for mapping urban trees in high-resolution aerial imagery using limited datasets and deep learning. Deep learning has become best-practice for this task, however, existing approaches rely on large and accurately labelled training datasets, which can be difficult and expensive to obtain. However, often noisy and incomplete data may be available that can be combined and utilized to solve more difficult tasks than those datasets were intended for.

This paper studies how to combine accurate point labels of urban trees along streets with crowd-sourced annotations from an open geographic database to delineate city trees in remote sensing images, a task which is challenging even for humans. To that end, we perform semantic segmentation of very high resolution aerial imagery using a fully convolutional neural network.

The main challenge is that our segmentation maps are sparsely annotated and incomplete. Small areas around the point labels of the street trees coming from official and crowd-sourced data are marked as foreground class. Crowd-sourced annotations of streets, buildings, etc. define the background class. Since the tree data is incomplete, we introduce a masking to avoid class confusion.

Our experiments in Hamburg, Germany, showed that the system is able to produce tree cover maps, not limited to trees along streets, without providing tree delineations. We evaluated the method on manually labelled trees and show that performance drastically deteriorates if the open geographic database is not used.

11:00 am - 12:00 pm

11:01 am - 11:20 am

Designing SafeMap Based on City Infrastructure and Empirical Approach: Modified A-Star Algorithm for Earthquake Navigation Application

Omid Veisi, Delong Du, Mohammad Amin Moradi, Fernanda Guasselli, Sotiris Athanasoulias, Hussain Abid Syed, Claudia Müller, and Gunnar Stevens

Designing routing systems for earthquakes requires frontend usability studies and backend algorithm modifications. Evaluations from subject-matter experts can enhance the design of both the front-end interface and the back-end algorithm of urban artificial intelligence (AI). Urban AI applications need to be trustworthy, responsible, and reliable against earthquakes, by assisting civilians to identify safe and fast routes to safe areas or health support stations. However, routes may become dangerous or obstructed as regular routing applications may fail to adapt responsively to city destruction caused by earthquakes. In this study, we modified the A-star algorithm and designed an interactive mobile app with the evaluation and insights of subject-matter experts including 15 UX designers, 7 urbanists, 8 quake survivors, and 4 first responders. Our findings reveal reducing application features and quickening application use time is necessary for stressful earthquake situations, as emerging features such as augmented reality and voice assistant may negatively backlash user experience in earthquake scenarios due to over-immersion, distracting users from real world condition. Additionally, we utilized expert insights to modify the A-star algorithm for earthquake scenarios using the following steps: 1) create a dataset based on the roads; 2) establish an empty dataset for weight; 3) enable the updating of weight based on infrastructure; and 4) allow the alteration of weight based on safety, related to human behavior. Our study provides empirical evidence on why urban AI applications for earthquakes need to adapt to the rapid speed to use and elucidate how and why the A-star algorithm is optimized for earthquake scenarios.

11:20 am - 11:40 am

Understanding Urban Economic Status through GNN-based Urban Representation Learning Using Mobility Data

Yuyol Shin, Gwanghwan Seong, Namwoo Kim, Seyun Kim, and Yoonjin Yoon

With growing urban population and urban concentration, various data-driven efforts are being made to achieve sustainable growth to promote equity, inclusion, and well-being. Among abundant urban data, mobility data is a source with rich semantic about urban environments in which social and economic activities are dissolved. In this paper, we employ graph attention network (GAT) to obtain urban representation learning embedding based on taxi trips and subway ridership data in Seoul, South Korea. Our GAT-based region embedding model outperformed all baseline models in predicting the number of employees and housing prices. For the number of employees prediction, our model achieved R-squared value of 0.649 using mobility data only. We also found that increasing the embedding dimensions to stack the elderly and disabled subway user types can further improve the model’s capability in the number of employees and housing prices predictions. Our study results suggest that a transportation network is a key contributor to shaping the economic landscapes of urban regions. Such findings also indicate that understanding people’s activities and movements is essential in achieving sustainable urban growth and promoting equity and inclusion for all. Our research contributes to the growing body of research on the urban region representation learning in understanding the economic impact of transportation systems on urban regions, especially for vulnerable populations.

11:40 am - 12:00 pm

Spatiotemporal Variability of Electric System Reliability Metrics

Nagendra Singh and Femi Omitaomu

As urbanization continues to grow, urban density also rises, placing significant stress on critical infrastructure. Cities often face challenges in investing in new infrastructure or upgrading existing ones. One area of infrastructure that has come under strain is the electric grid, leading to prolonged power outages. According to data from the U.S. Energy Information Administration, the average duration of power outages has steadily doubled from 2013 to 2021, primarily due to extreme weather events. The IEEE has developed a series of reliability metrics to assess the reliability of the electric system. However, these metrics are only applied at the utility level, making it challenging to comprehend the variations in these indices at finer spatial resolutions and, consequently, evaluate reliability at the non-utility level. In this study, we leveraged electric outage data collected at 15-minute intervals at the county level to estimate electric system reliability over a six-year period using a novel data analysis approach. Our findings reveal a gradual decline in reliability in the U.S. Southwest compared to national averages. Furthermore, the duration of outages per customer per event has significantly decreased in large portions of California and Florida. These discoveries can be valuable for utility companies as they seek to pinpoint regional anomalies and plan for prioritized remediation efforts.