Earthquake prediction remains a key challenge in disaster mitigation, particularly in seismically active regions like Sumatra. For predictive purposes, the Epidemic-Type Aftershock Sequence (ETAS) model has recently been combined with a neural network. However, two main issues arise: (1) the difficulty to tune a number of hyperparameters within the neural network models and (2) an impractical separated procedure. In this study, we incorporate Evolutionary Neural Architecture Search (ENAS) to obtain optimal hyperparameters and consider multi-output deep learning architectures to perform simultaneous forecasts of maximum seismicity rate and highest-risk geographical zone in the Sumatra region of Indonesia. The results indicate improvements in predictive performance relative to conventional deep learning techniques, featuring improved accuracy and computational efficiency. These findings emphasize the promise of ENAS and multi-output architectures in enhancing seismic forecasting, offering practical insights for disaster preparedness and risk reduction in earthquake-prone areas.
Revealing spatial and temporal dynamics of foot-traffic-based business recovery after extreme weather events remains a critical but understudied aspect of post-disaster resilience. This study examines post-disaster foot-traffic recovery patterns and their associated drivers in the Houston Metropolitan Statistical Area following Winter Storm Uri in February 2021. Using cellphone location-based foot traffic data, we track daily visits to 10,971 customer-facing retail and service businesses across multiple sectors. A recovery index is developed to quantify the number of days required for businesses to return to at least 80% of pre-storm foot traffic levels. Recovery is evaluated across four post-disaster periods (2-, 5-, 7-, and 10-days) to capture temporal variation in recovery behaviour. Both cumulative and non-cumulative probit regression models are applied to identify factors influencing early and delayed recovery. Explanatory variables include business characteristics, neighbourhood socio-demographic and housing conditions, disaster-related stressors, and a lagged spatial association indicator capturing the recovery status of nearby businesses. Results show that early recovery (2–5 days) was primarily constrained by disaster-related stressors and socio-demographic vulnerability. Small and medium-sized businesses and essential sectors, particularly grocery and health services, recovered more quickly, while the lagged spatial association significantly increased recovery likelihood. In contrast, later recovery (7–10 days) was more strongly associated with institutional and internal business factors, as the influence of disaster stressors and the lagged spatial association declined. Overall, the findings highlight the time-dependent and spatially heterogeneous nature of post-disaster foot-traffic recovery and demonstrate the importance of temporally sensitive, spatially targeted policies to support equitable and efficient recovery following extreme weather events.
GIScience is crucial for spatial understanding, analysis, and decision-making across multiple sectors. As GIScience demonstrates its continual impact in addressing critical societal and environmental challenges, growing concerns regarding its responsible use rise accordingly. The integration of vast and intricate datasets and the application of translucent algorithms hold the potential to infringe on privacy, perpetuate biases, exacerbate societal inequities, or undermine reliability. This work integrates responsible thinking with GIScience by conceptualizing Responsible GIScience and proposes the STEP (Security, Trustworthiness, Equity, Philanthropy) framework, which organizes the core concerns of Responsible GIScience into four interacting pillars. A collection of subtopics is identified and categorized for each pillar through an integrative review of the literature and trending issues. The framework offers definitions and descriptions of the subtopics, summarizes current solutions, rekindles missing parts, demonstrates the cruxes, and translates abstract values into example measurable indicators and workflow checkpoints. The role of Responsible GIScience is elaborated by Protection, Instruction, Promotion, Evolution, and future research directions and challenges are discussed. This work stands at the intersection of GIScience, AI, and Big Data, as well as other interdisciplinary fields, and sheds light on a potential pathway to ensure GIScience proceeds responsibly. It calls for attention from GIScience professionals and advocates a transformation of more inclusive education, the establishment of new regulations, and joint efforts of a broader audience to secure this long-term agenda.
Efficient spatial planning of logistics infrastructure under deep uncertainty remains a critical challenge in geo-information science. This study advances spatial decision modelling by proposing an Iterative Tri-Level Optimization Framework that integrates Geographic Information Systems (GIS), the Best-Worst Method (BWM), and Z-number Data Envelopment Analysis (Z-number DEA) within a closed-loop, meta-optimized system. The framework operates through three interconnected levels: (1) base evaluation of geographic suitability ([Formula: see text]), operational efficiency ([Formula: see text]), and Spatial Risk Index ([Formula: see text]) for 59 candidate sites across Iran; (2) portfolio synthesis via a Composite Decision Index [Formula: see text] and selection of [Formula: see text] networks using a Network Performance Score [Formula: see text] that balances demand-weighted efficiency with geographic dispersion through [Formula: see text] regularization; and (3) meta-optimization employing a Genetic Algorithm to calibrate strategic weights [Formula: see text] and maximize [Formula: see text]. The optimized 15-node portfolio spans 15 provinces, achieving strong spatial equity [Formula: see text], [Formula: see text] population coverage, and [Formula: see text]. Z-number DEA robustly handles deep uncertainty by jointly modeling parameter imprecision [Formula: see text] and information reliability [Formula: see text]. Geopolitical scenario analysis quantifies systemic impacts ranging from [Formula: see text] (comprehensive sanctions) to [Formula: see text] (Caspian expansion). The framework illustrates conceptually cross-national transferability through a structured Egyptian adaptation and establishes quantifiable linkages to nine UN Sustainable Development Goals (SDGs) as emergent outcomes – not design objectives. This work contributes to geo-information science by embedding GIS-based spatial evaluation within an iterative network optimization framework, enabling spatial suitability, operational efficiency, and spatial risk to be jointly integrated into a network-aware Spatial Decision Support System for strategic infrastructure planning under deep uncertainty.
Analyzing urban morphology across different geographical contexts can reveal recurring patterns of urbanization and support the critical evaluation of urban design practices. Existing machine-learning approaches generally follow either supervised classification or unsupervised clustering. Supervised methods retain established morphological knowledge, but force all samples into predefined classes, whereas unsupervised methods can identify new patterns but often lack clear connections to existing urban-morphological concepts. To address this challenge, this study introduces generalized category discovery into urban morphology and establishes an open-world evaluation paradigm for jointly recognizing reference morphotypes and discovering categories withheld from training. Based on this paradigm, we propose AnyMor-VLM, a novel urban-morphotype-aligned, scalable framework that leverages vision-language models capable of identifying any geometry-based urban morphotypes. The AnyMor-VLM integrates techniques of language-guided representation learning, supervised learning, self-supervised contrastive learning, and cross-modal alignment learning. Comprehensive experiments show that the AnyMor-VLM achieves the strongest overall classification performance on the proposed benchmark and improves the recognition of established reference classes and novel categories withheld from training. Two proof-of-concept applications further illustrate the use of the urban-area delineation and the identification of informal-settlement-associated fabrics. These findings provide evidence that open-world techniques can bridge knowledge-driven classification and data-driven pattern discovery in computational urban morphology.
