A five-year milestone: reflections on advances and limitations in GeoAI research

Annals of GIS, Journal Year: 2024, Volume and Issue: 30(1), P. 1 - 14

Published: Jan. 2, 2024

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

Language: Английский

---

Explainable spatially explicit geospatial artificial intelligence in urban analytics

Environment and Planning B Urban Analytics and City Science, Journal Year: 2023, Volume and Issue: 51(5), P. 1104 - 1123

Published: Sept. 29, 2023

Geospatial artificial intelligence (GeoAI) is proliferating in urban analytics, where graph neural networks (GNNs) have become one of the most popular methods recent years. However, along with success GNNs, black box nature AI models has led to various concerns (e.g. algorithmic bias and model misuse) regarding their adoption particularly when studying socio-economics high transparency a crucial component social justice. Therefore, desire for increased explainability interpretability attracted increasing research interest. This article proposes an explainable spatially explicit GeoAI-based analytical method that combines convolutional network (GCN) graph-based (XAI) method, called GNNExplainer. Here, we showcase ability our proposed two studies within analytics: traffic volume prediction population estimation tasks node classification classification, respectively. For these tasks, used Street View Imagery (SVI), trending data source analytics. We extracted semantic information from images assigned them as features roads. The GCN first provided reasonable predictions related by encoding roads nodes connectivities graphs. GNNExplainer then offered insights into how certain are made. Through such process, practical conclusions can be derived phenomena studied here. In this paper also set out path developing XAI future studies.

Language: Английский

---

Assessment of a new GeoAI foundation model for flood inundation mapping

Published: Nov. 13, 2023

Vision foundation models are a new frontier in Geospatial Artificial Intelligence (GeoAI), an interdisciplinary research area that applies and extends AI for geospatial problem solving geographic knowledge discovery, because of their potential to enable powerful image analysis by learning extracting important features from vast amounts data. This paper evaluates the performance first-of-its-kind model, IBM-NASA's Prithvi, support crucial task: flood inundation mapping. model is compared with convolutional neural network vision transformer-based architectures terms mapping accuracy flooded areas. A benchmark dataset, Sen1Floods11, used experiments, models' predictability, generalizability, transferability evaluated based on both test dataset completely unseen model. Results show good Prithvi highlighting its advantages segmenting areas previously regions. The findings also indicate improvement adopting multi-scale representation learning, developing more end-to-end pipelines high-level tasks, offering flexibility input data bands.

Language: Английский

---

Multimodal Colearning Meets Remote Sensing: Taxonomy, State of the Art, and Future Works

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Journal Year: 2024, Volume and Issue: 17, P. 7386 - 7409

Published: Jan. 1, 2024

In remote sensing (RS), multiple modalities of data are usually available, e.g., RGB, Multispectral, Hyperspectral, LiDAR, and SAR. Multimodal machine learning systems, which fuse these rich multimodal modalities, have shown better performance compared to unimodal systems. Most research assumes that all present, aligned, noiseless during training testing time. However, in real-world scenarios, it is common observe one or more missing, noisy, non-aligned, either both. addition, acquiring large-scale, noise-free annotations expensive, as a result, lacking sufficient annotated datasets having deal with inconsistent labels open challenges. These challenges can be addressed under paradigm called co-learning. This paper focuses on co-learning techniques for data. We first review what available the domain key benefits combining context. then tasks would benefit from processing including classification, segmentation, target detection, anomaly temporal change detection. dive deeper into technical details by reviewing than 200 recent efforts this area provide comprehensive taxonomy systematically state-of-the-art approaches 4 missing noisy limited modality annotations, weakly-paired modalities. Based insights, we propose emerging directions inform potential future sensing.

Language: Английский

---

GeoAI Reproducibility and Replicability: A Computational and Spatial Perspective

Annals of the American Association of Geographers, Journal Year: 2024, Volume and Issue: 114(9), P. 2085 - 2103

Published: July 23, 2024

Language: Английский

---

Urban Fabric Decoded: High-Precision Building Material Identification via Deep Learning and Remote Sensing

Environmental Science and Ecotechnology, Journal Year: 2025, Volume and Issue: 24, P. 100538 - 100538

Published: Feb. 3, 2025

Precise identification and categorization of building materials are essential for informing strategies related to embodied carbon reduction, retrofitting, circularity in urban environments. However, existing material databases typically limited individual projects or specific geographic areas, offering only approximate assessments. Acquiring large-scale precise data is hindered by inadequate records financial constraints. Here, we introduce a novel automated framework that harnesses recent advances sensing technology deep learning identify roof facade using remote Google Street View imagery. The model was initially trained validated on Odense's comprehensive dataset then extended characterize across Danish landscapes, including Copenhagen, Aarhus, Aalborg. Our approach demonstrates the model's scalability adaptability different contexts architectural styles, providing high-resolution insights into distribution diverse types cities. These findings pivotal sustainable planning, revising codes lower emissions, optimizing retrofitting efforts meet contemporary standards energy efficiency emission reductions.

Language: Английский

---

Data fusion of complementary data sources using Machine Learning enables higher accuracy Solar Resource Maps

Solar Energy, Journal Year: 2025, Volume and Issue: 290, P. 113337 - 113337

Published: Feb. 18, 2025

Language: Английский

---

Grouting Geological Model (GGM): Definition, Characterization, Modeling, and Application in Determining Grouting Material and Pressure

Rock Mechanics and Rock Engineering, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 25, 2025

Language: Английский

---

Advancing Arctic sea ice remote sensing with AI and deep learning: now and future

Published: Jan. 22, 2024

Abstract. The revolutionary advances of Artificial Intelligence (AI) in the past decade have brought transformative innovation across science and engineering disciplines. Also field Arctic science, we witnessed an increasing trend adoption AI, especially deep learning, to support analysis big data facilitate new discoveries. In this paper, provide a comprehensive review applications learning sea ice remote sensing domains, focusing on problems such as lead detection, thickness estimation, concentration, extent forecasting motion detection well type classification. addition discussing these applications, also summarize technological that customized solutions, including loss functions strategies better understand dynamics. To promote growth exciting interdisciplinary field, further explore several research areas where community can benefit from cutting-edge AI technology. These include improving multi-modal capabilities, enhancing model accuracy measuring prediction uncertainty, leveraging foundation models, deepening integration with physics-based models. We hope paper serve cornerstone progress using inspire field.

Language: Английский

---

Reasoning cartographic knowledge in deep learning-based map generalization with explainable AI

International Journal of Geographical Information Science, Journal Year: 2024, Volume and Issue: 38(10), P. 2061 - 2082

Published: June 20, 2024

Cartographic map generalization involves complex rules, and a full automation has still not been achieved, despite many efforts over the past few decades. Pioneering studies show that some tasks can be partially automated by deep neural networks (DNNs). However, DNNs are used as black-box models in previous studies. We argue integrating explainable AI (XAI) into DL-based process give more insights to develop refine understanding what cartographic knowledge exactly is learned. Following an XAI framework for empirical case study, visual analytics quantitative experiments were applied explain importance of input features regarding prediction pre-trained ResU-Net model. This experimental study finds XAI-based visualization results easily interpreted human experts. With proposed workflow, we further find DNN pays attention building boundaries than interior parts buildings. thus suggest boundary intersection union better evaluation metric commonly qualifying raster-based results. Overall, this shows necessity feasibility part future development frameworks.

Language: Английский

---