ABSTRACT Global Ocean Surface Currents (GOSC) play a critical role in understanding ocean dynamics and tracking pollutant dispersion, such as marine microplastics. However, accurately reconstructing GOSC remains a challenge. To address this, we propose PhyGOSCR-Net, a physics-driven deep learning model that employs convolutional layers to extract multiscale features and attention mechanisms to maintain global coherence. We train the model using geostrophic currents, Ekman currents, and Stokes drift as inputs, with Argo float velocities at 0 m and drifter-derived velocities at 15 m as labels, to reconstruct physics-driven GOSC at both depths. Experimental results show that our model outperforms GLOBCURRENT, particularly in equatorial regions where geostrophic assumptions fail, reducing Root Mean Square Error by 10–30 cm/s across both zonal and meridional components. The high correlation with observational data (correlation coefficient > 0.8) confirms its ability to model coupled physical processes. Based on reconstructed ocean currents, microplastic transport simulations successfully identify the five major global accumulation zones, including the Great Pacific Garbage Patch, demonstrating their strong potential for environmental research. This work advances high-precision ocean current reconstruction and supports pollution tracking and mitigation efforts.

ABSTRACT Accurate estimation of snow depth on Antarctic sea ice is critical for understanding ice mass balance, surface thermodynamics, and satellite-altimetry-based sea ice thickness retrievals. This study introduces a dual-mode retrieval framework for deriving a snow depth product on Antarctic sea ice using the Microwave Radiation Imager (MWRI). The MWRI snow depth product outperforms existing passive-microwave products in accuracy, seasonal adaptability, and temporal consistency. Validation against multiple in situ datasets shows that MWRI snow depth achieves superior performance across most Antarctic regions, notably with an RMSE below 9.0 cm and a correlation coefficient above 0.70 in West Antarctica. During the melt season, validation with AWI snow buoys yielded an RMSE of 8.5 cm, demonstrating robustness under complex surface conditions. Time-series comparisons with ICESat-2 snow depth demonstrate that the MWRI snow depth effectively captures seasonal variability (r = 0.79), accurately reproducing the winter-to-spring snow-accumulation trend. Seasonally, snow depth rises in winter and early spring and diminishes during summer melt and compaction. Interannually, long-term snow-covered zones—particularly the Weddell and Amundsen Seas—remain relatively stable and thick, while marginal ice areas exhibit a clear thinning trend. The dataset is available at https://doi.org/10.57760/sciencedb.25039.

ABSTRACT Geospatial data and workflows are rapidly becoming central to environmental and sustainability compliance reporting, but the relationship between evolving EU regulation and available Earth Observation (EO) and GIS technologies remains poorly systematized. Focusing on the European Union and including the private sector, we review binding and soft-law instruments such as the CSRD and ESRS, the EU Taxonomy, the SFDR and the EUDR, and analyse which provisions require which type of spatial information for which business activities. Building on a structured scoping review of scientific and grey literature, we synthesise state-of-the-art EO and geospatial approaches that can operationalise these legal requirements, with a particular emphasis on deforestation-related supply-chain regulations. We propose a conceptual framework that distinguishes three families of geospatial workflows—risk screening, attribution and verification—and maps them to typical corporate reporting processes. Across the reviewed legislation and applications, we identify recurring gaps between legal definitions and EO-derived classifications, significant uncertainties in global land use/land cover products, and challenges in integrating geospatial data quality metadata into auditable reporting processes. At the same time, we highlight emerging opportunities from open EO archives, global benchmark products and AI-enabled processing platforms that can support scalable, transparent and repeatable geospatial workflows for sustainability reporting. The paper concludes by outlining a research and innovation agenda for standardised reference datasets, benchmarking protocols and interoperable platforms that jointly involve regulators, data providers and corporate users to ensure that geospatial workflows are fit for purpose in environmental and sustainability compliance reporting.

ABSTRACT Addressing the global environmental crisis necessitates coordinated efforts, supported by open and reproducible research practices. Such practices aim to enhance the reliability, efficiency, and credibility of scientific outputs. Innovative tools are necessary for systematic conservation planning. This technical note presents a reproducible and automated approach for supporting land management and planning by identifying and updating ecological infrastructure (EI). Grounded in open science Findable, Accessible, Interoperable, and Reusable principles data management, and digital twin (DT) concepts, the method focuses on the Canton of Geneva, Switzerland. It integrates species distribution modelling, ecosystem service assessments, and spatial prioritization within a shared JupyterLab environment. The infrastructure centralizes data, automates indicator calculations, and ensures transparency, traceability, and reproducibility through version control and metadata generation. Ecological tools like Zonation enable the identification of high-priority conservation areas aligned with international targets. The system facilitates collaborative workflows and indicator updates. Its architecture allows scalability to broader regions and scenario modelling, laying the foundation for a DT of Geneva’s environment. Despite challenges in harmonizing workflows across institutional partners, this solution enhances efficiency and replicability in EI planning. The methodology is transferable to other regions and adaptable to various environmental modelling domains, offering a robust base for sustainable territorial management.

ABSTRACT Large Language Models (LLMs) have demonstrated substantial progress in task automation and natural language understanding. However, without domain expertise in geographic information science (GIS), they continue to encounter limitations including reduced accuracy and unstable performance when processing complex spatial tasks. To address these challenges, we propose GeoJSON agents—a novel multi-agent LLM architecture specifically designed for geospatial analysis. This framework transforms natural language instructions into structured GeoJSON operations through two widely adopted LLM enhancement techniques: function calling and code generation. The architecture integrates three core components: task parsing, agent collaboration, and result integration. The planner agent systematically decomposes user-defined tasks into executable subtasks, while specialized worker agents perform spatial data processing and analysis either by invoking predefined function APIs or by dynamically generating and executing Python-based analytical code. The system produces reusable, standards-compliant GeoJSON outputs through iterative refinement. To systematically evaluate both approaches, we constructed a hierarchical benchmark comprising 70 tasks spanning basic, intermediate, and advanced complexity levels, conducting experiments with OpenAI’s GPT-4o as the core model. Results indicate that the code generation–based agent achieved 97.14% accuracy, while the function calling–based agent attained 85.71%—both significantly outperforming the best-performing general-purpose model (48.57%). Comparative analysis reveals that code generation offers superior flexibility for complex, open-ended tasks, whereas function calling provides enhanced execution stability for structured operations. This study represents the first systematic integration of GeoJSON data with a multi-agent LLM framework and provides empirical evidence comparing two mainstream enhancement methodologies in geospatial contexts, offering new perspectives for improving GeoAI system performance and reducing barriers to GIS application.

ABSTRACT Accurate 3D reconstruction of dynamic street scenes is crucial for autonomous driving, yet existing methods either require costly 3D annotation boxes or fail to capture fine object motion. To overcome these limitations, we propose SSTD-GS, a self-supervised Gaussian Splatting framework for annotation-free dynamic scene reconstruction and novel view synthesis. Specifically, we design a spatiotemporal deformation field to model the detailed motion of dynamic objects, and develop an uncertainty dynamic mask guided self-supervised strategy to enable joint optimization of dynamic and static scene components. To further improve the quality of novel view synthesis, with the help of the powerful priors of the depth completion model and diffusion model, we design a confidence dense depth prior module and a diffusion model virtual view prior module to provide additional geometric and appearance constraints. Moreover, a geometry aware Gaussian adaptive control mechanism is employed to suppress inaccurate densification in 3DGS caused by rendering errors. Experimental results on the Waymo and KITTI datasets show that SSTD-GS outperforms existing NeRF and 3DGS-based methods in 4D scene reconstruction and novel view synthesis. In the novel view synthesis task, the PSNR reaches 29.83 and 28.59 dB, respectively, which are 1.72 and 1.36 dB higher than the suboptimal PVG.

ABSTRACT Nature-based solutions are increasingly being adopted worldwide to enhance urban sustainability and mitigate climate change impacts. Among them, green roofs (GRs) are widely promoted through policies offering technical support, tax benefits, and financial incentives, with some cities mandating GRs in new buildings. Since vegetation drives most GR-related benefits, such as urban heat island (UHI) mitigation, stormwater regulation, biodiversity, and carbon capture, monitoring its well-being is essential. This study introduces two novel satellite-based indices to assess GR vegetation health. The Vegetative Stress Index (VSI) is a single-parameter tool that tracks annual vegetation growth trends. The Multiparameter Vegetative Stress Index (MVSI) integrates climatic and urban variables to evaluate their combined influence on the Normalized Difference Vegetation Index (NDVI). Both indices were applied to Lisbon, Portugal, named European Green Capital in 2020 for its efforts in expanding green spaces. Results demonstrate that VSI and MVSI effectively capture long-term improvements in vegetative well-being across GRs, reflected in significant NDVI increases. Compared to other urban green spaces, GRs show enhanced vegetation resilience, underscoring their added value. These indices offer a practical, scalable method for urban planners and policymakers to monitor and support green infrastructure development and performance over time.

ABSTRACT Wildfires are becoming more frequent and intense, which highlights the need for precise and effective forest burned area (FBA) detection. Current burn mapping approaches are hindered by challenges including the integration of multimodal datasets, high computational complexity of traditional attention mechanisms in segmentation models, and cloud contamination in optical satellite imagery. To address these issues, we proposed the Cloud-Aware Mixture-of-Experts Linear Transformer U-Net (CA-MTransU-Net). Our model integrates Sentinel-1 SAR and Sentinel-2 optical satellite data using a novel dynamic weighting approach, employs a computationally efficient Mixture-of-Experts (MoE) linear attention mechanism to effectively capture global feature dependencies, and incorporates a cloud-weighting method specifically designed to reduce the adverse impacts of cloud cover in optical satellite data. The developed architecture significantly outperformed several well-known segmentation algorithms, including U-Net, ResNet, SegFormer, TransU-Net, PSPNet, and DeepLabv3+, achieving the highest mean Intersection-over-Union (mIoU) of 87.00%, surpassing baseline models by an average of +6.29%. It also demonstrated superior computational efficiency with faster inference speeds (6.26 ms) compared to conventional transformer-based models like SegFormer (7.81 ms) and TransU-Net (13.17 ms). Despite its achievements, the model exhibits higher peak memory usage, which may limit deployment in resource-constrained environments. Additionally, like other tested models, it occasionally misclassified water bodies as burned areas.

ABSTRACT This study assesses the accuracy of ten satellite-based and reanalysis precipitation datasets available in Google Earth Engine (GEE) using in-situ rain gauge measurements across Czechia, Central Europe, from 2001 to 2021. The gauge-adjusted GSMaP dataset (GSMaPGA) was the most accurate dataset overall (Pearson’s correlation coefficient r = 0.79), followed by ERA5-Land (r = 0.75), with both showing superior performance for rainy days above 1 mm of precipitation. In contrast, CHIRPS, GLDAS, and PERSIANN-CDR showed the weakest performance (r ≈ 0.41–0.42). All datasets overestimated precipitation on days with no or with very light rain (≤1 mm/day) and underestimated it during heavy rainfall events ( >5 mm/day). ERA5-Land systematically overestimated annual precipitation by 15–35%, while GSMaPGA showed slight underestimation by 0.5–9%. Although absolute errors generally increased with elevation, GSMaPGA showed the smallest elevation-related biases, highlighting the importance for gauge-adjustment. Part of the observed spatial and seasonal biases may be explained by the combination of coarse spatial resolution and the challenges of capturing short-lived summer convective storms over complex terrain. Overall, GSMaPGA is recommended for most applications due to its superior accuracy, while ERA5-Land is suitable for long-term studies because of its long historical record extending back to the 1950s.

ABSTRACT Points of Interest (POIs) data are vital for location-based services, yet their production remains challenging due to labor-intensive collection and verification processes. Generating POIs from street-view imagery (SVI) has recently emerged as a promising solution. However, the lack of open benchmark hinders its development. Existing methods typically treat SVI as isolated images without fully leverage their multi-view representations of geographical entities. We present SVI2POI, a novel end-to-end framework for POI extraction from SVI. It brings two key innovations. In the signboard recognition stage, the proposed YOLOv11s-DLKA detector enhanced performance degraded by geometric distortions commonly occurred in SVI. In the POI generation stage, we propose a clustering strategy combined with large language model-based naming and photogrammetric positioning to consolidate multi-view information for accurate POI identification. Furthermore, we introduce the first open dataset for end-to-end POI generation from SVI. It contains a training dataset including 3,097 SVIs with 13,182 manually annotated regions of interest (ROIs), and a benchmark dataset with 927 SVIs and 1,004 manually labeled POIs, with 190 verified against OpenStreetMap-POIs and therefore contains coordinates. Our framework achieves 61.69% precision, 50.70% recall, and 55.66% F1-score, outperforms state-of-art method with 5.59%, 1.16%, and 3.04%, respectively, via cross-method and cross-dataset comparison.