Understanding the spatiotemporal patterns of human movement is important across diverse applications including urban design, disease control, social and cognitive science, and emergency response planning. Recently, multilayer mobility networks were used to study how movements between spatial units (e.g., census tracts) can significantly vary when they are stratified according to different industry sectors-e.g., movements to grocery stores, to schools, or to hospitals. Here, we study the predictability of movements across different industry sectors using statistical and machine learning models trained on demographic, socioeconomic, and infrastructure information. We compare ten predictive models and identify advantages for nonlinear models (with random forest regression being a consistent top performer). We identify the most important features enabling prediction (population size for outward movements from regions and industry-related infrastructure for movements into regions). Of the two, prediction for inward movements (i.e., in-degrees) is generally more difficult; however, the difference is small for movements associated with food services. We also compare the prediction of weekly and time-averaged movements, finding that with the addition of time-encoding input features, weekly movements are easier to predict than time-averaged values (at least for the nonlinear predictive models). These findings provide a practical step toward using machine learning for human movement modeling and the many downstream applications.

Speaker anonymization effectively conceals speaker identity in speech signals to protect privacy. To address issues in existing anonymization systems, including reduced voice distinguishability, limited anonymized voices, reliance on an external speaker pool, and vulnerability to privacy leakage against strong attackers, a novel distinguishability-driven voice generation for speaker anonymization via random projection and the Gaussian Mixture Model (GMM) is proposed. This method first applies the random projection to lower the dimensionality of the X-vectors from an external speaker pool, and then constructs a GMM in the reduced dimensional space to fit the generative model. By sampling from this generative model, anonymous speaker identity representations are generated, ultimately synthesizing anonymized speech that maintains both intelligibility and distinguishability. To ensure the anonymized speech remains sufficiently distinguishable from the original and prevents excessive similarity, a cosine similarity check is implemented between the original X-vector and pseudo-X-vector. Experimental results on the VoicePrivacy Challenge datasets demonstrate that the proposed method not only effectively protects speaker privacy across different attack scenarios but also preserves speech content integrity while significantly enhancing speaker distinguishability between original speakers and their corresponding pseudo-speakers, as well as among different pseudo-speakers.

Abstract Understanding human mobility during disastrous events is crucial for emergency planning and disaster management. We develop a methodology to construct time-varying, multilayer networks where edges encode observed movements between spatial regions (census tracts) and network layers encode movement categories by industry sectors (e.g., schools, hospitals). Using the 2021 Texas winter storm as a case study, we find that people markedly reduced movements to ambulatory health care services, restaurants, and schools, but prioritized movements to grocery stores and gas stations. Additionally, we study the predictability of nodes’ in- and out-degrees in the multilayer networks, which encode movements into and out of census tracts. Inward movements prove harder to predict than outward movements, especially during the storm. This case study highlights our methodology’s effectiveness for detecting/characterizing mobility shifts, and our specific findings on the reduction, prioritization, and predictability of sector-specific movements aim to support mobility-related decisions during future extreme weather events.

Expansion of the wildland–urban interface (WUI), fire suppression, and climate change have substantially increased human exposure to wildfire hazards. Although data-driven machine learning approaches have been widely applied to wildfire spread prediction, they often lack physical consistency and do not adequately represent the landscape and environmental conditions controlling fire propagation, including vegetation and fuel conditions, surface moisture, topography, and dynamic weather. Remote-sensing-based Geospatial Foundation Models (GeoFMs) pretrained on large-scale remote sensing datasets provide rich environmental representations (i.e., embeddings) for Earth observation, yet their potential for wildfire spread simulation remains underexplored. In this research, we propose a physics-informed GeoAI framework that couples GeoFM-derived environmental embeddings with downstream decoders tailored for wildfire spread prediction, with three complementary deep learning strategies, including a physics-informed parameter learning, a physics-informed loss function enforcing plausible fire-state transitions, and a physics-guided forward bias representing the slope and wind-driven propagation effects. In addition, the proposed GeoAI framework integrates multiple physical coupling mechanisms, including state transition, environmental propagation factors, and heat transfer. Specifically, an LSTM-based decoder is designed for mapping the sequential hidden states with physics constraints for the forward process and physics-informed loss optimization, while a CNN–based partial differential equation (PDE) decoder explicitly models wildfire propagation through a learnable PDE formulation and integrates it with physics-based parameter learning within a unified framework. Using 45 large wildfire events in California, the experiment results demonstrate that the proposed physics-informed GeoAI framework outperforms both purely data-driven deep learning and physical models. Specifically, the GeoFM-LSTM and GeoFM–PDE decoders achieve up to 15\% and 6\% improvements in average F1-score (classification performance), respectively. Furthermore, the proposed framework provides insights on process-level interpretability by characterizing diffusion, advection, and source-dominated wildfire spread patterns.

Understanding how AI will represent and reason about geography should be a key concern for all of us, as the broader public increasingly interacts with spaces and places through these systems. Similarly, in line with the nature of foundation models, our own research often relies on pre-trained models. Hence, understanding what world AI systems construct is as important as evaluating their accuracy, including factual recall. To motivate the need for such studies, we provide three illustrative vignettes, i.e., exploratory probes, in the hope that they will spark lively discussions and follow-up work: (1) Do models form strong defaults, and how brittle are model outputs to minute syntactic variations? (2) Can distributional shifts resurface from the composition of individually benign tasks, e.g., when using AI systems to create personas? (3) Do we overlook deeper questions of understanding when solely focusing on the ability of systems to recall facts such as geographic principles?

People have been increasingly using social media to post messages during a natural disaster, and describe the locations of victims, damages, difficult situations, and relief resources. Many of these location descriptions are in the forms of detailed and multi-entity descriptions, such as door number addresses, road intersections, and highway exits. Currently, there is limited availability of datasets that contain these detailed location descriptions labeled in disaster-related messages. A lack of these datasets hinders the understanding of how people describe locations during disasters and the automatic extraction of these location descriptions. This paper fills this gap by providing a dataset that covers ten disasters in the United States and in five disaster types: hurricanes, floods, wildfires, tornados, and winter storms. The messages containing location descriptions are collected from the social media platform Twitter/X, and we describe the collection, labeling, and validation of this dataset. This dataset can be used for studying the ways people describe locations under disaster contexts and for training AI models to extract these important locations.

During a natural disaster, people post text messages on various platforms, such as social media and short message service (SMS) platforms, to share urgent information and seek help. Many text messages contain location descriptions about victims and accidents. Accurately extracting these location descriptions can help disaster responders reach victims more quickly and even save lives. These location descriptions, however, are often more complex than simple place names (e.g. city names), and cannot be extracted using typical named entity recognition approaches. While new machine learning models could be trained, they require labeled training data that are time-consuming to create without an effective data annotation tool. To fill this gap, we develop GALLOC, a GeoAnnotator for Labeling LOCation descriptions from disaster-related text messages. GALLOC is an open-source and Web-based tool that provides a variety of functions for supporting location description annotation, such as artificial intelligence powered pre-annotation and automatic spatial footprint identification. It also supports multilingual data annotation, and can be used by a group of users to collaboratively create a dataset. We present the design considerations and functions of GALLOC and evaluate it via a comparison with previous tools and an experiment to annotate a small set of disaster-related messages.

Predicting the spread of wildfires is essential for effective fire management and risk assessment. With the fast advancements of artificial intelligence (AI), various deep learning models have been developed and utilized for wildfire spread prediction. However, there is limited understanding of the advantages and limitations of these models, and it is also unclear how deep learning-based fire spread models can be compared with existing non-AI fire models. In this work, we assess the ability of five typical deep learning models integrated with weather and environmental variables for wildfire spread prediction based on over ten years of wildfire data in the state of Hawaii. We further use the 2023 Maui fires as a case study to compare the best deep learning models with a widely-used fire spread model, FARSITE. The results show that two deep learning models, i.e., ConvLSTM and ConvLSTM with attention, perform the best among the five tested AI models. FARSITE shows higher precision, lower recall, and higher F1-score than the best AI models, while the AI models offer higher flexibility for the input data. By integrating AI models with an explainable AI method, we further identify important weather and environmental factors associated with the 2023 Maui wildfires.

A deep hashing method using a time sequence speech visualization feature was proposed to improve speech retrieval accuracy, the efficiency of deep learning, and the noise robustness of speech deep hashing. The speech data is transformed into spectrogram time sequences. Two deep learning models, the 3D CNN (Convolutional Neural Networks)-BiLSTM (Bidirectional Long Short-Term Memory) deep model, and the bidirectional ConvLSTM deep model, are constructed to learn from the time sequence speech visualization feature to generate deep hashing for speech full-text retrieval. Experimental results demonstrate that the proposed models only require fewer iterations to achieve satisfied training accuracy. Moreover, the time sequence speech visualization feature can effectively represent speech content to achieve high retrieval accuracy and recall. Additionally, the proposed deep hashing exhibits high robustness to noise. The proposed method has higher accuracy and is robust to speaker identity compared to existing methods.

The 2022 Buffalo blizzard was a catastrophic winter storm that struck Buffalo, New York in the week of Christmas in 2022. It claimed 47 lives and left much of the region stranded for the holiday week. In this disaster, the 311 call service was used by many residents to request help for issues due to the blizzard. This study examines these 311 help requests and their potential disparities across communities. Specifically, we aim to: (1) understand the spatial and temporal distributions of different types of 311 help requests; (2) identify the physical and social vulnerability factors, as well as human behavior factors, that are associated with the use of 311 calls. Methodologically, we leverage both explainable geospatial artificial intelligence (GeoAI) methods and statistical analysis to analyze 311 help requests and their associated factors. Our analysis shows significant spatial disparities in 311 help requests across communities. Results from explainable GeoAI and statistical analysis also reveal complementary insights on key factors associated with 311 help requests, such as historical 311 request behavior and percentage of minority population. These results could inform future disaster management decisions and help mitigate the negative impacts of winter storm disasters.

Multimodal biometric databases are critical for enhancing the security and accuracy of identity authentication systems and advancing research in multimodal fusion. However, developing these databases is challenged by high acquisition costs, privacy concerns, and ownership issues. This paper introduces a multimodal biometric database named LUTBIO, which includes nine types of biometric data: voice, face, fingerprint, contact-based palmprint, electrocardiogram (ECG), opisthenar, ear, contactless palmprint, and periocular. The database specifically focuses on collecting biometric data from middle-aged and elderly individuals to address the challenges of identity authentication, which are associated with an ageing population. Moreover, there is a significant gap in comparative research on multimodal fusion methods involving more than four biometric modalities. To address this, the paper systematically investigates the effectiveness of various fusion strategies in enhancing recognition performance using the LUTBIO database. Experimental results show that LUTBIO is highly effective for multimodal fusion, with both score-level and decision-level fusion markedly improving overall recognition performance and robustness. For instance, a model uses a score-level fusion strategy with the maximum scores from eight low-accuracy unimodal models. This model achieved an accuracy of 0.991, while a model employing a majority voting decision-level fusion strategy attained an accuracy of 0.998. Further analysis explores how the quality and type of sensors influence multimodal fusion results. The experimental results indicate that within a reasonable range of sensor quality and collection conditions, multimodal biometric authentication systems can achieve reliable recognition performance while maintaining low costs. The database is available here.

Established spatial and life course methods have helped epidemiologists and health and medical geographers study the impact of individual and area-level determinants on health disparities. While these methods are effective, the emergence of Geospatial Artificial Intelligence (GeoAI) offers new opportunities to leverage complex and multi-scalar data in spatial aging and life course research. The objective of this perspective is three-fold: (1) to review established methods in aging, life course, and spatial epidemiology research; (2) to highlight some of the opportunities offered by GeoAI for enhancing research on health disparities across life course and aging research; (3) to discuss the shortcomings of using GeoAI methods in aging and life course studies.

Large pre-trained models, also known as foundation models (FMs), are trained in a task-agnostic manner on large-scale data and can be adapted to a wide range of downstream tasks by fine-tuning, few-shot, or even zero-shot learning. Despite their successes in language and vision tasks, we have not yet seen an attempt to develop foundation models for geospatial artificial intelligence (GeoAI). In this work, we explore the promises and challenges of developing multimodal foundation models for GeoAI. We first investigate the potential of many existing FMs by testing their performances on seven tasks across multiple geospatial domains, including Geospatial Semantics, Health Geography, Urban Geography, and Remote Sensing. Our results indicate that on several geospatial tasks that only involve text modality, such as toponym recognition, location description recognition, and US state-level/county-level dementia time series forecasting, the task-agnostic large learning models (LLMs) can outperform task-specific fully supervised models in a zero-shot or few-shot learning setting. However, on other geospatial tasks, especially tasks that involve multiple data modalities (e.g., POI-based urban function classification, street view image–based urban noise intensity classification, and remote sensing image scene classification), existing FMs still underperform task-specific models. Based on these observations, we propose that one of the major challenges of developing an FM for GeoAI is to address the multimodal nature of geospatial tasks. After discussing the distinct challenges of each geospatial data modality, we suggest the possibility of a multimodal FM that can reason over various types of geospatial data through geospatial alignments. We conclude this article by discussing the unique risks and challenges to developing such a model for GeoAI.

Geographical random forest (GRF) is a recently developed and spatially explicit machine learning model. With the ability to provide more accurate predictions and local interpretations, GRF has already been used in many studies. The current GRF model, however, has limitations in its determination of the local model weight and bandwidth hyperparameters, potentially insufficient numbers of local training samples, and sometimes high local prediction errors. Also, implemented as an R package, GRF currently does not have a Python version which limits its adoption among machine learning practitioners who prefer Python. This work addresses these limitations by introducing theory-informed hyperparameter determination, local training sample expansion, and spatially weighted local prediction. We also develop a Python-based GRF model and package, PyGRF, to facilitate the use of the model. We evaluate the performance of PyGRF on an example dataset and further demonstrate its use in two case studies in public health and natural disasters.

Winter Storm Uri slammed Texas between February 13–17, 2021 and caused widespread power outages. Understanding the impacts of this catastrophic event on local communities has important meaning. In this study, we examine the impacts of this winter storm and its impact disparities on different population groups over three stages of this disaster: the initial-hit stage, power-outage stage, and recovery stage. The study focuses on Harris County, Texas which was severely affected by the winter storm. We leverage home-dwelling time information from anonymized mobile phone location data to study the constrained mobility of people due to the winter storm as a way to quantify its impacts on local communities. Considering that mobile phone location data may be affected by the power outages, we further integrate nighttime light (NTL) images into our analyses to assess disaster impacts during the power-outage stage, and use home-dwelling time to assess the impacts during the other two stages (i.e., the initial-hit stage and recovery stage). The results reveal disparate impacts of this winter storm on local communities in the three stages of this disaster. We also find impact disparities on population groups with different socioeconomic and demographic backgrounds, especially during the initial-hit stage. These results help us better understand the impacts of this catastrophic event, and could inform future response and mitigation efforts in identifying vulnerable communities, allocating resources, and curtailing negative impacts of similar disasters.

Background:The COVID-19 pandemic highlighted the importance of urban parks to provide safe places to visit and recreate. Recent research has suggested that park visitation over this time may not have occurred equitably, which may exacerbate existing health disparities. However, usual methods of estimating park visitation are labor intensive, requiring better solutions.. The objective of this study was to assess how park visitation changed in Buffalo, New York after the start of the COVID-19 pandemic utilizing human mobility data.Methods:Monthly mobile phone location data from January 2018 through October of 2021 from residents of Buffalo were analyzed to estimate total park visits for each census tract. A generalized linear mixed effect model was utilized to examine if selected factors affected park visitation. Factors examined included demographic, health, park, and crime data at the census tract level.Results:Across 587,487 park visits that were captured in the 79 census tracts, park visitation increased by 25% since March 2020. In our regression model, having cancer and currently smoking had negative effects on park visitation. The start of the COVID-19 pandemic positively affected park visitation. Season (of the year), was the other statistically significant variable that affected park visitation.Conclusions:Anonymous mobile phone data demonstrated that park visitation has increased by 25% since the beginning of the pandemic when we looked at census tract level data. While some data limitations must be addressed, mobile phone data is a novel method that can be used to understand behavioral and public health trends.

The Annual Meeting of the American Association of Geographers (AAG) in 2023 marked a five-year milestone since the first Geospatial Artificial Intelligence (GeoAI) Symposium was held at AAG in 2018. In the past five years, progress has been made while open questions remain. In this context, we organized an AAG panel and invited five panellists to discuss the advances and limitations in GeoAI research. The panellists commended the successes, such as the development of spatially explicit models, the production of large-scale geographic datasets, and the use of GeoAI to address real-world problems. The panellists also shared their thoughts on limitations in current GeoAI research, which were considered as opportunities to engage theories in geography, enhance model explainability, quantify uncertainty, and improve model generalizability. This article summarizes the presentations from the panellists and also provides after-panel thoughts from the organizers. We hope that this article can make these thoughts more accessible to interested readers and help stimulate new ideas for future breakthroughs.

This paper reviews trends in GeoAI research and discusses cutting-edge advances in GeoAI and its roles in accelerating environmental and social sciences. It addresses ongoing attempts to improve the predictability of GeoAI models and recent research aimed at increasing model explainability and reproducibility to ensure trustworthy geospatial findings. The paper also provides reflections on the importance of defining the "science" of GeoAI in terms of its fundamental principles, theories, and methods to ensure scientific rigor, social responsibility, and lasting impacts.

The content-based speech retrieval scheme using the hash technique boasts high accuracy; however, its index structures, such as the hash table, suffer from high space cost, low search efficiency, and security concerns. Moreover, implementing keyword search or ranking retrieval is a challenging task. We suggest a secure content-based retrieval scheme that utilizes a novel hashing-inverted index to address these issues. This method converts hash sequences into keyword sequences by employing a keyword hash template to establish a hashing-inverted index. The index table and all keywords are encrypted using a lightweight homomorphic encryption scheme to ensure security. When the data user submits a trapdoor of a spoken query, the cloud server performs homomorphic Hamming distance calculation and homomorphic location calculation, which makes the search process much more secure. The experimental results show that our index is lighter, easy to maintain, and expand. The retrieval efficiency is also higher. Security analysis proves that the privacy of the index and speeches is preserved. Generally, our scheme is accurate, efficient, and secure for the massive spoken audio data.

A vast amount of location information exists in unstructured texts, such as social media posts, news stories, scientific articles, web pages, travel blogs, and historical archives. Geoparsing refers to the process of recognizing location references from texts and identifying their geospatial representations. While geoparsing can benefit many domains, a summary of the specific applications is still missing. Further, there lacks a comprehensive review and comparison of existing approaches for location reference recognition, which is the first and a core step of geoparsing. To fill these research gaps, this review first summarizes seven typical application domains of geoparsing: geographic information retrieval, disaster management, disease surveillance, traffic management, spatial humanities, tourism management, and crime management. We then review existing approaches for location reference recognition by categorizing these approaches into four groups based on their underlying functional principle: rule-based, gazetteer matching-based, statistical learning-based, and hybrid approaches. Next, we thoroughly evaluate the correctness and computational efficiency of the 27 most widely used approaches for location reference recognition based on 26 public datasets with different types of texts (e.g., social media posts and news stories) containing 39,736 location references across the world. Results from this thorough evaluation can help inform future methodological developments for location reference recognition, and can help guide the selection of proper approaches based on application needs.