Drive-by sensing, using vehicles as mobile sensors to collect environmental data, offers high spatiotemporal resolution monitoring. In drive-by sensing tasks, urban areas with intense human activity or pollution – termed high-priority hotspots – demand frequent sensing for adequate data collection. However, prior studies rarely address optimizing vehicle selection to enhance hotspot coverage frequency while maintaining non-hotspot coverage. This study formalized this problem as the Maximal Hotspot Regular Coverage Problem and proposed an Adaptive Level-Aware Vehicle Selection algorithm. Using air pollution hotspot sensing in Beijing as an empirical case, results show the advantage of the proposed algorithm over other baselines: Random selection (RS), Non-hotspot Greedy Adding and Multi-type hotspot Greedy MaxMin, covering 29.99% of hotspots and 77.08% non-hotspot zones with 1000 sensors on average. The spatial distribution of covered area and statistical distribution of average visits per hour reveals a large spatial range and high revisit times of the method, and its advantage over baselines in a hybrid sensing scenario is also proven (with 54.18% daily average coverage among all zones and 31.22% on hotspots). This study provides a method to advance the coverage of hotspots in single-type and hybrid sensing scenarios, inspiring a future extension of optimization based on the proposed algorithm.

Human activity intensity prediction is crucial to many location-based services. Despite tremendous progress in modeling dynamics of human activity, most existing methods overlook physical constraints of spatial interaction, leading to uninterpretable spatial correlations and over-smoothing phenomenon. To address these limitations, this work proposes a physics-informed deep learning framework, namely Gravity-informed Spatiotemporal Transformer (Gravityformer) by integrating the universal law of gravitation to refine transformer attention. Specifically, it (1) estimates two spatially explicit mass parameters based on spatiotemporal embedding feature, (2) models the spatial interaction in end-to-end neural network using proposed adaptive gravity model to learn the physical constraint, and (3) utilizes the learned spatial interaction to guide and mitigate the over-smoothing phenomenon in transformer attention. Moreover, a parallel spatiotemporal graph convolution transformer is proposed for achieving a balance between coupled spatial and temporal learning. Systematic experiments on six real-world large-scale activity datasets demonstrate the quantitative and qualitative superiority of our model over state-of-the-art benchmarks. Additionally, the learned gravity attention matrix can be not only disentangled and interpreted based on geographical laws, but also improved the generalization in zero-shot cross-region inference. This work provides a novel insight into integrating physical laws with deep learning for spatiotemporal prediction.

Spatial clustering has been widely used for spatial data mining and knowledge discovery. An ideal multivariate spatial clustering should consider both spatial contiguity and aspatial attributes. Existing spatial clustering approaches may face challenges for discovering repeated geographic patterns with spatial contiguity maintained. In this paper, we propose a Spatial Toeplitz Inverse Covariance-Based Clustering (STICC) method that considers both attributes and spatial relationships of geographic objects for multivariate spatial clustering. A subregion is created for each geographic object serving as the basic unit when performing clustering. A Markov random field is then constructed to characterize the attribute dependencies of subregions. Using a spatial consistency strategy, nearby objects are encouraged to belong to the same cluster. To test the performance of the proposed STICC algorithm, we apply it in two use cases. The comparison results with several baseline methods show that the STICC outperforms others significantly in terms of adjusted rand index and macro-F1 score. Join count statistics is also calculated and shows that the spatial contiguity is well preserved by STICC. Such a spatial clustering method may benefit various applications in the fields of geography, remote sensing, transportation, and urban planning, etc.

Existing transit-oriented-development (TOD) classification studies primarily focus on the static characteristics around transit stations to measure the built environment’s density, diversity, and design. As a community development model, time-variant variables, dynamic human activities throughout different times of the day and week matter in further unpacking the characteristics of TODs. Given that this aspect has been under-discussed in most previous TOD literature, this research provides an activity-based framework to classify commuter transit station areas by considering the degree of local vibrancy - the temporal visiting pattern of all points of interest (POIs) that fall within the station areas. We apply a two-step semi-unsupervised clustering algorithm to classify 4,290 station areas from 54 metropolitan areas across the U.S. This method produces 13 distinct station area types. Next, we further examine the connection between station area types and neighborhood travel behavior. A cross-sectional comparison reveals that stations with consistent active morning activities are associated with a higher ratio of commuting by walking and biking and lower automobile usage measured in vehicle miles traveled (VMT). Using stations opened after 2009, we show that active weekend activity patterns are associated with a more significant increase in commuting by public transit.

Human emotion is an intrinsic psychological state that is influenced by human thoughts and behaviours. Human emotion distribution has been regarded as an important part of emotional geography research. However, it is difficult to form a global scaled map reflecting human emotions at the same sampling density because various emotional sampling data are usually positive occurrences without absence data. In this study, a methodological framework for mapping the global geographic distribution of human emotion is proposed and applied, combining a species distribution model with physical environment factors. State-of-the-art affective computing technology is used to extract human emotions from facial expressions in Flickr photos. Various human emotions are considered as different species to form their ‘habitats’ and predict the suitability, termed as ‘Emotional Habitat’. To our knowledge, this framework is the first method to predict emotional distribution from an ecological perspective. Different geographic distributions of seven dimensional emotions are explored and depicted, and emotional diversity and abnormality are detected at the global scale. These results confirm the effectiveness of our framework and offer new insights to understand the relationship between human emotions and the physical environment. Moreover, our method facilitates further rigorous exploration in emotional geography and enriches its content.

Dashboard visualizations are widely used in data-intensive applications such as business intelligence, operation monitoring, and urban planning. However, existing visualization authoring tools are inefficient in the rapid prototyping of dashboards because visualization expertise and user intention need to be integrated. We propose a novel approach to rapid conceptualization that can construct dashboard templates from exemplars to mitigate the burden of designing, implementing, and evaluating dashboard visualizations. The kernel of our approach is a novel deep learning-based model that can identify and locate charts of various categories and extract colors from an input image or sketch. We design and implement a web-based authoring tool for learning, composing, and customizing dashboard visualizations in a cloud computing environment. Examples, user studies, and user feedback from real scenarios in Alibaba Cloud verify the usability and efficiency of the proposed approach.

Spatial interpolation is a traditional geostatistical operation that aims at predicting the attribute values of unobserved locations given a sample of data defined on point supports. However, the continuity and heterogeneity underlying spatial data are too complex to be approximated by classic statistical models. Deep learning models, especially the idea of conditional generative adversarial networks (CGANs), provide us with a perspective for formalizing spatial interpolation as a conditional generative task. In this article, we design a novel deep learning architecture named conditional encoder-decoder generative adversarial neural networks (CEDGANs) for spatial interpolation, therein combining the encoder-decoder structure with adversarial learning to capture deep representations of sampled spatial data and their interactions with local structural patterns. A case study on elevations in China demonstrates the ability of our model to achieve outstanding interpolation results compared to benchmark methods. Further experiments uncover the learned spatial knowledge in the model’s hidden layers and test the potential to generalize our adversarial interpolation idea across domains. This work is an endeavor to investigate deep spatial knowledge using artificial intelligence. The proposed model can benefit practical scenarios and enlighten future research in various geographical applications related to spatial prediction.

Regionalization attempts to group units into a few subsets to partition the entire area. The results represent the underlying spatial structure and facilitate decision-making. Massive amounts of trajectories produced in the urban space provide a new opportunity for regionalization from human mobility. This paper proposes and applies a novel regionalization method to cluster similar areal units and visualize the spatial structure by considering all trajectories in an area into a word embedding model. In this model, nodes in a trajectory are regarded as words in a sentence, and nodes can be clustered in the feature space. The result depicts the underlying socio-economic structure at multiple spatial scales. To our knowledge, this is the first regionalization method from trajectories with natural language processing technology. A case study of mobile phone trajectory data in Beijing is used to validate our method, and then we evaluate its performance by predicting the next location of an individual’s trajectory. The case study indicates that the method is fast, flexible and scalable to large trajectory datasets, and moreover, represents the structure of trajectory more effectively.