ABSTRACT Hospital visitation reflects a complex destination‐choice process shaped by travel cost, perceived hospital attractiveness, and neighborhood socioeconomic conditions. However, less is known about how these factors jointly structure realized hospital destination choice and how their relationships vary across travel distances and community contexts. This study integrates 4 years (2020–2023) of anonymized mobility data in Houston, TX, USA, with hospital attributes, Google Maps ratings and review counts, and 2023 American Community Survey socioeconomic indicators, including health‐insurance coverage. Five predictive models were evaluated across multiple test‐set proportions using 10 repeated train‐test splits. The heterogeneous graph neural network (HGNN) consistently achieved the strongest predictive performance, while interpretation of the HGNN identified drive time as the dominant predictor and showed that hospital capacity, occupancy, spatial context, online visibility, ratings, and community characteristics jointly modify visitation patterns. Controlled scenario analyses indicated that rating‐related distance‐decay crossovers vary across socioeconomic contexts rather than following a uniform threshold. These findings highlight hospital visitation as a heterogeneous travel behavior shaped by interacting spatial, institutional, and socioeconomic factors.
ABSTRACT The rapid advancement of Generative AI (GenAI) creates new opportunities for spatial planning. However, most existing applications focus on macro‐scale overhead perspectives, leaving street‐level visualization constrained by strict physical and spatial fidelity challenges. We introduce a controllable GenAI‐human interactive framework for producing spatially faithful and realistic street‐level imagery that reflects envisioned planning scenarios. Using Large‐scale solar photovoltaic (LSPV) site planning as a test case, it leverages vision‐language models to integrate human inputs (e.g., textual intent, scribbles) and existing status photos, synthesizing photorealistic outcomes. To support systematic evaluation, we constructed a large‐scale street‐view imagery dataset of before‐and‐after solar panel installation across the United States. We comprehensively evaluated 11 model settings covering five open‐sourced diffusion models and one commercial vision foundation model to test fidelity and realism. Our study illustrates how a GenAI‐human interactive pipeline can support visualizing proposed environmental interventions, and offers a benchmark for conditioned GenAI visualization applications.
ABSTRACT Time underpins analysis across the natural and social sciences, yet representing and computing it across extreme temporal scales and multiform expressions remains challenging, particularly in cross‐disciplinary data integration and discovery in the geosciences. Textual forms are difficult to compute, numerical forms often suffer from unit inconsistencies and symbolic complexity, and existing temporal models remain limited in cross‐scale continuity and flexibility. To address these issues, this study conducts a systematic analysis of the distinct characteristics of time and proposes a novel time‐coding framework—Variable‐length Ultra‐scale Time Segmented Coding (VUTSC). This method encodes time as three components: a Base reference, a Level indicator, and a variable‐length sequence of Codes, which together enable consistent and scalable representation across ultra‐wide temporal scales within a unified structural framework. Using the geosciences as a case study, we build a Deep‐Time Digital Earth Timeline System based on VUTSC. Experiments on multi‐scale and interdisciplinary time expressions validate the feasibility, breadth, simplicity, flexibility, and efficiency of VUTSC. Results indicate that this model can effectively express geoscientific time across a wide range of scales, while its coding efficiency is significantly higher than that of timestamp‐based methods. The encoding and decoding strategies are stable and support hierarchical traversal, neighborhood search, and time‐relationship calculations. For time‐centric scientific disciplines, including the Earth sciences, this approach provides a promising pathway for multi‐source, multi‐scale data mining and knowledge discovery.
ABSTRACT Detecting spatial heterogeneity boundaries from irregular point observations is important for understanding abrupt changes in spatial processes and producing interpretable boundary maps. However, irregularly sampled spatial data lack a natural ordering, making conventional one‐dimensional change point detection difficult to apply directly. We develop a modular two‐stage spatial analysis framework that connects spatial interpolation with established change point detection procedures. First, interpolation reconstructs a gridded representation of the underlying spatial field. Second, the interpolated surface is decomposed into geographically ordered one‐dimensional spatial profiles, where change point detection is used to identify boundaries between heterogeneous regions. Simulation experiments under different boundary configurations, spatial error structures, interpolation settings, and detection methods show that the framework can recover major boundary patterns, especially when appropriate interpolation smoothing is combined with stable detection methods. An application to January 2025 PM 2.5 monitoring data in China demonstrates its usefulness for extracting interpretable boundary structures from irregular environmental monitoring networks.
ABSTRACT Rapid urbanization in semi‐arid regions with complex geomorphological settings poses significant environmental and territorial challenges, particularly in developing countries where planning frameworks struggle to keep pace with spatial dynamics. This study investigates the spatiotemporal dynamics, geo‐environmental drivers, and future trajectories of urban expansion in the Wilaya of Constantine, northeastern Algeria, from 1985 to 2050. We used multi‐temporal land‐cover data from the GLC_FCS30D dataset (30 m resolution; 1985–2022) to quantify land‐use/land‐cover (LULC) change and urban growth. We extracted built‐up areas from the impervious surface class and analyzed areal expansion, growth rate, and spatial trajectory. To identify the geo‐environmental determinants of urbanization, 49,878 stratified sample points (24,939 urban and 24,939 non‐urban pixels) were generated, and eight predictor variables slope, elevation, topographic wetness index, drainage density, distance to streams, seismic hazard, geology, and soil type were incorporated into four machine‐learning algorithms. Model performance was assessed through five‐fold cross‐validation, and predictor contributions were interpreted via SHAP (SHapley Additive exPlanations) analysis. Future urban expansion for 2030, 2040, and 2050 was projected under a business‐as‐usual scenario using a Cellular Automata–Markov Chain (CA–Markov) model calibrated using LULC maps for 2000 and 2010, validated by simulating the 2020 land‐cover map and comparing it with the observed GLC_FCS30D 2020 map (Overall Accuracy = 95.7%, Kappa = 0.926, K_no = 0.994). Future projections used a transition matrix derived from the 2010 to 2020 period. Results reveal a near‐fourfold increase in urban area between 1985 and 2022 (+303%; Compound Annual Growth Rate (CAGR) ≈3.84%), with the most intense expansion phase occurring between 1995 and 2000. Random Forest (RF) achieved the highest predictive accuracy (AUC‐ROC = 0.906), with soil type, elevation, and seismic hazard identified as the dominant geo‐environmental predictors of urbanization. Under a business‐as‐usual scenario, the urban area is projected to expand from 113.8 km 2 in 2020 to 205.9 km 2 by 2050, with a plausible range of 124–243 km 2 across low, BAU, and high growth scenarios. The integrated ML/CA‐Markov framework proposed here constitutes a robust, transferable, and policy‐relevant approach for evidence‐based urban planning in rapidly urbanizing Mediterranean cities.
