GeoAI and Multimodal Geospatial Data Fusion: A New Path to Building Inclusive Urban Transportation

This article reviews the applications of GeoAI and multimodal geospatial data fusion technologies in inclusive urban transportation, analyzes 57 core literature pieces from 2019 to 2025, reveals the trajectory of technological development, proposes a fairness-oriented development roadmap for transportation systems, and provides practical guidance for urban planners and policymakers. It focuses on the dilemma of urban transportation inequality and explores how to solve the travel problems of marginalized communities through technological innovation.

While urbanization brings prosperity, it also creates spatial inequality. Historically marginalized communities face issues such as insufficient public transportation coverage, long commute times, and high costs. Traditional transportation planning struggles to address systemic inequality due to data limitations and a single perspective. The rise of GeoAI and multimodal geospatial data fusion technologies provides new possibilities for building inclusive transportation systems.

GeoAI is the deep integration of AI and geospatial science, capable of processing large-scale diverse datasets and automatically discovering spatiotemporal patterns. Multimodal data fusion is its core capability. Urban transportation data types include GPS trajectories, satellite imagery, social media, sensor networks, etc. The main fusion strategies are: feature-level fusion (integrating similar features in the preprocessing stage), decision-level fusion (synthesizing after independent analysis of each modality), and deep learning fusion (neural networks automatically learning cross-modal correlations).

The research emphasizes the core position of "fairness". Traditional optimization tends to exacerbate inequality. Cases include: accessibility analysis to identify service blind spots and reveal the exclusion of marginalized communities; demand forecasting to capture travel patterns of marginalized groups through multimodal fusion; resource allocation using fairness-aware algorithms to ensure investment is tilted toward underserved communities.

Key technologies include: Graph Neural Networks (GNN) for modeling road networks and travel patterns; CNN and Vision Transformer for analyzing satellite/street view images to identify land use; LSTM/Transformer time-series models for predicting traffic flow and demand; Federated Learning to solve data privacy issues and enable cross-institutional collaborative training.

Challenges: Data bias (solidifying existing inequalities), technical barriers (lack of infrastructure and talent in developing countries), privacy and security issues. Future directions: Develop few-shot learning to adapt to data scarcity; promote causal inference to understand mechanisms; strengthen the application of multimodal large language models; establish interdisciplinary collaboration mechanisms; formulate AI ethical guidelines and governance frameworks.

GeoAI and multimodal fusion provide powerful tools for inclusive transportation, but technology is neutral and must be centered on human dignity and social fairness. It is recommended that planners and policymakers focus on interpretability (XAI technology) and participatory methods in technology applications to ensure transparent and fair decision-making, so that smart cities truly serve everyone.