Carlos Eduardo Mota; TBD (oral presentation or poster)
The Mineral Research and Production Support Platform (P3M) was conceived and is being implemented by a team of specialists from the Geological Survey of Brazil (SGB). with the support and participation of public and private entities directly or indirectly related to the Brazilian mineral industry, such as: Secretariat of Geology, Mining and Mineral Transformation (SGM) of the Ministry of Mines and Energy (MME); National Mining Agency (ANM); Agency for the Development and Innovation of the Brazilian Mineral Sector (ADIMB); Brazilian Association of Mineral Research and Mining Companies (ABPM); National Association of Aggregate Producers for Construction (ANEPAC); Brazilian Institute of Geography and Statistics (IBGE), and Brazilian Mining Institute (IBRAM).
The core development of P3M is being carried out by the UFLA (Federal University of Lavras) Agency for Innovation in Geotechnologies and Intelligent Systems (Zetta). Launched in late 2022, the Platform's main objective is to increase the attractiveness of investments and promote the sustainable development of the mineral industry, through the integration of data on mineral potential, infrastructure, costs, legislation, protected areas, and socioeconomic indicators of Brazil.
It has a crucial role in planning mineral research and production, generating qualified knowledge that is essential to support strategic decisions in both the public and private sectors. For industries and investors in the mining sector, the data available on the Platform contribute to supporting development plans and the exploitation of mineral deposits. Government entities also benefit primarily from information for monitoring the competitiveness of mineral research and production, thereby promoting greater attractiveness of investments for national development.
In short, P3M provides information free of charge through two main environments: (i) a map and data viewer that integrates geoscientific, economic, environmental and legal information, mining rights, among others, from various national and governmental entities; and (ii) a dashboard system with statistical data presentation, with filters by territorial unit (e.g., region, state, or municipality) and commodity.
One of the main differentiators of the P3M platform, among the various software products of SGB, is its complete conception and development using FOSS4G. The main objective of this work is to present the solution architecture used in the development, focusing primarily on the use of FOSS4G projects. In early February of this year, version 2.4.4 was released, with several improvements in terms of user experience and performance.
The platform deployment is entirely container-based, provisioned in a Kubernetes environment using Helm Chart, and data storage relies on PostgreSQL 14 databases with PostGIS 3.4. In general terms, the P3M Platform can be subdivided into three main components: (i) map and dashboard visualization system, (ii) data pipeline orchestration, and (iii) spatial data infrastructure.
The map and dashboard applications are deployed in two container images: the first, a backend project with Django + Rest framework, brings a system of REST APIs to define the content controls for the maps and dashboards. The application also has an administrative interface so that content managers can modify the application's behavior and the list of available layers and groups. The front-end component is a TypeScript application that relies on libraries such as OpenLayers to render the layer tree, delivered by the backend, and the dashboard data.
The data pipeline orchestration environment was deployed from a custom Apache Airflow container image, supplemented with libraries such as GDAL, GeoPandas, and Apache Arrow/Parquet.
For P3M, some data pipelines were developed to collect data of mining processes and financial compensation – available from the National Mining Agency (ANM). This data, originally delivered in spreadsheets and File Geodatabases, is sanitized, enriched, and transformed using GDAL, and finally saved in proprietary PostgreSQL/PostGIS tables of active mines, commodity associations, and mining process statuses. Currently, these pipelines run on Mondays, Wednesdays, and Fridays and are crucial for maintaining the dashboards and authorial layers.
The provision of maps and metadata on P3M strictly follows the use of OGC Open Web Services standards, such as WMS, WMTS, and WFS, and the platform consumes maps and metadata from a GeoNode installation owned by the SGB, in order to fulfill the role of Spatial Data Infrastructure.
The SGB’s GeoNode was deployed to a production environment in late 2025, initially to support this platform. Prior to this GeoNode, P3M had a dedicated GeoServer instance in its own workspace to serve the maps. After going into production, all layers and configurations were moved to this GeoNode. Currently, about 320 vector and raster layers from various Brazilian government data sources are cataloged and registered. Initially, this data was organized and cleaned to create a database that was placed as a datastore in GeoNode's GeoServer.
The option of hosting duplicate data instead of consuming OGC services from other institutions was chosen due to the need to have this data available on the platform, even if the source provider is unavailable, in addition to the geological service having a robust IT infrastructure. The P3M authorial layers are also hosted on GeoNode, but directly from the exposure of the spatial tables to the GeoNode Geoserver. Some of these tables, such as mineral resource data and systematic mapping, are also fed from pipelines in Airflow.
Finally, the P3M Platform, since its conception, has always been guided by the use of free and open-source libraries and frameworks to ensure transparency and interoperability. Alongside major FOSS4G projects, P3M stands out among the select group of data and service providers applied to Geology and Mining sectors by adopting frameworks that are free from restrictive licensing and bundled sales practices. The application is publicly available at https://p3mgeo.sgb.gov.br/.
Junyoung CHOI; TBD (oral presentation or poster)
Proximity-based community planning has emerged as an important approach for improving urban livability by ensuring that essential services are accessible within short travel distances. Rather than relying solely on administrative boundaries, this planning approach focuses on the spatial organization of everyday urban activities and the accessibility of key services such as healthcare, education, retail, and public facilities.
However, operationalizing proximity-based planning requires analytical frameworks capable of integrating large-scale mobility data, accessibility modelling, and spatial optimization. In many cities, these analytical components remain fragmented across different tools and datasets, making it difficult to develop reproducible and scalable workflows for urban analysis. At the same time, recent advances in open geospatial technologies provide new opportunities to build transparent, interoperable, and reproducible analytical systems that support data-driven planning practices.
This research proposes an open geospatial analytical framework for proximity-based community planning that integrates mobility data, open-source geospatial software, and spatial optimization techniques into a unified and reproducible analytical workflow. The framework is designed to support two key planning tasks:
the identification of functional urban communities based on observed mobility patterns, and the evaluation of spatial strategies for improving equitable access to essential urban services.
By combining multiple open geospatial technologies, the proposed framework aims to provide a flexible analytical approach that can be applied across different urban contexts while maintaining transparency and reproducibility.
The analytical framework consists of three core components that correspond to different stages of proximity-based planning analysis.
The first component focuses on the delineation of functional urban communities using mobility-based community detection. Large-scale telecom mobility data are used to construct origin–destination interaction networks between spatial units, represented as a grid-based spatial system. These mobility networks capture aggregated daily travel patterns between locations and provide a behavioral representation of urban spatial structure. Community detection algorithms from network science are applied to these networks in order to identify clusters of spatial interaction that represent functional communities emerging from mobility patterns. Unlike traditional planning units that rely on administrative boundaries, these mobility-derived communities reflect actual patterns of urban activity and interaction. As a result, the framework allows planners to identify spatial units that function as integrated communities in terms of daily mobility and service access.
The second component performs multimodal accessibility analysis at a high spatial resolution using open geospatial routing tools. Accessibility is calculated on a 250 m grid by combining telecom-derived origin–destination mobility flows with network-based travel times. Road-based accessibility is estimated using open routing engines such as OSRM, which compute travel distances and travel times along road networks derived from open geospatial data sources. In addition, public transport accessibility is computed using the R5 routing engine through the r5py Python interface. The R5 engine enables multimodal routing that integrates pedestrian networks, road networks, and public transport schedules derived from GTFS data. This approach allows the framework to calculate multimodal travel times across different transport modes, including walking, road-based transport, and public transit. The use of R5 and r5py also supports efficient computation of accessibility metrics for large-scale urban datasets, enabling the evaluation of accessibility patterns at a fine spatial resolution. By integrating multiple travel modes, the framework provides a more realistic representation of accessibility conditions in dense urban environments where public transport plays a significant role in daily mobility.
The third component incorporates facility allocation models based on linear programming in order to optimize the spatial distribution of urban services. The optimization model uses accessibility indicators derived from the multimodal analysis to evaluate potential service locations and allocation strategies. The objective function simultaneously considers equity and economic efficiency in service provision, allowing the model to identify spatial configurations that improve service accessibility while minimizing spatial inequality and redundant infrastructure investment. The allocation model can therefore support planning decisions related to the placement of public facilities, community services, and other urban amenities that are critical for proximity-based community planning.
A key contribution of this study lies in the integration of multiple open geospatial tools into a unified analytical workflow. Rather than introducing a single standalone software package, the framework combines several existing open-source geospatial technologies, including Python-based network analysis libraries, open routing engines, and geospatial data processing tools. This integration demonstrates how different open geospatial components can be combined to support complex urban analytics tasks. Furthermore, the computational workflow is designed to support reproducible research practices by documenting analytical steps and enabling the release of scripts, models, and computational procedures under open-source principles. Such reproducibility is essential for ensuring transparency and enabling other researchers and practitioners to replicate and extend the analytical framework.
The framework is demonstrated through an empirical case study in Busan, South Korea, where telecom mobility data are used to analyze community structures and evaluate alternative service distribution scenarios. The case study illustrates how mobility-derived communities differ significantly from conventional administrative planning units and provide a more realistic representation of urban spatial interaction. The integration of multimodal accessibility analysis and spatial optimization further allows planners to examine how alternative facility configurations influence accessibility outcomes across the urban population. These results highlight the potential of combining open geospatial technologies with mobility data to support evidence-based urban planning.
This study contributes to the open geospatial research community in several ways.
First, it demonstrates how open geospatial technologies can support integrated urban analytics workflows for community-level planning.
Second, it connects methods from network science, multimodal accessibility modelling, and spatial optimization within a reproducible open-source analytical framework.
Third, it highlights how open geospatial ecosystems enable transparent and collaborative approaches to urban analysis and planning.
By presenting a reproducible analytical framework for proximity-based community planning built on open geospatial technologies, this research contributes to ongoing efforts within the FOSS4G community to advance open, scalable, and collaborative geospatial solutions for sustainable urban development.
Matthew Wilson, Luke Parkinson; TBD (oral presentation or poster)
Digital Twins are software systems that provide dynamic virtual representations of physical systems(1), enabling modelling and visualisation, with automated data exchange and analytics being key attributes. These systems are enabling the development of smart cities(2) and may also represent the natural environment(3–5). Common use cases for Digital Twins are to monitor and control manufacturing lines or smart cities, but in environmental applications they are less common. Digital Twins can be used to automate and connect computer models of the environment, enabling on-demand simulations or ingestion of model outputs in planning.
A key example of an environmental Digital Twin is the the EU's “Destination Earth” system, which is being developed as a Digital Twin for climate services, to facilitate access to weather and climate models which can be used for impact studies(6). Physics-based Digital Twins such as this will revolutionise access to and use of numerical model predictions. By connecting systems together through open-data and standards, a “Digital Twin web” will be created, powered by rapidly growing data and distributed cloud computing(7). Yet the development of each component remains challenging.
In this work, we describe the development of the Environmental Digital Data Intelligence Engine (EDDIE), an open-source framework for creating environmental Digital Twins. The concept of EDDIE is that it acts as a core engine which manages the ingestion and processing of spatial and other data, provides a modularised framework for running environmental models from these data, orchestrates them and ingests their results, and provides an (optional) web-based user interface and visualisation system. EDDIE is based on APIs, meaning that is it possible to connect two or more instances of EDDIE (or other Digital Twins) to share data and environmental models. For example, these Digital Twins can represent multiple different domains, such as hazard assessment, environmental monitoring, and community and urban planning. Here we describe the EDDIE system and provide some application examples.
EDDIE and its open-source module implementations help developers of novel Digital Twins by providing a structure to follow, and providing library functionality for key spatial data handling processes. A dashboard of existing spatial data becomes trivial to setup and fetching and combining open data for analysis becomes simpler by following existing workflows and patterns.
An application using EDDIE is comprised of multiple containers working together to form a web application. Key containers include PostGIS, GeoServer, TerriaJS and the EDDIE backend and processing containers. EDDIE’s Python library is used in the backend to prepare data and keep them up to date if required. When a model scenario is requested, the Python library is used within domain-specific modules to gather and process data to generate predictive outputs. TerriaJS is the typical frontend for an EDDIE application, allowing 3D visualisations as well as the ability to request model scenarios to be run. These requests use the OGC Web Processing Service standard, and return JSON results that are valid TerriaJS catalog items. This allows requests to use existing tooling with standardised inputs, with results that can be used in further processing scripts or can be automatically displayed on the web. The standard front-end for EDDIE applications is TerriaJS, with the backend containers able to expose detailed dynamic catalogs. These catalogs can also be used by other independent Digital Twins, enabling them to use all functionality available to create more powerful ecosystems of Digital Twins.
EDDIE is used in active research projects for multiple distinct Digital Twins developed by the Geospatial Research Institute Toi Hangarau. EDDIE was born from the Flood Resilience Digital Twin (FReDT), focused on automated prediction of flood risk and collation of data for impact analysis. Currently, FReDT allows users to select parameters relating to climate change to assess how sea-level rise and increased storm intensity may change flood inundation risk. Ongoing developments are focused on working with communities to develop nature-based solutions to reduce flood impact, while allowing them to trial many different scenarios using the web interface. The core modules were extracted from FReDT to be able to be reused to construct novel environmental Digital Twins, and this core has formed EDDIE.
From there, EDDIE was used as the basis for the Ōtākaro Digital Twin, a prototype environmental platform for monitoring the health of the Ōtākaro/Avon River in Christchurch, New Zealand. This Digital Twin was created in collaboration with Ngāi Tūāhuriri and Christchurch City Council. Modelling available within the platform currently focuses on the MEDUSA 2.0 stormwater pollutant runoff model using user-inputted rainfall event parameters, and potential future modelling may include linking this to rainfall gauge telemetry.
Most recently, EDDIE was the core framework used to create Te Awarua Kai Ora, a platform for Te Awarua / Porirua Harbour. This platform collates data from open data sources relating to the harbour, presents spatial data on environmental sampling, and allows the Porirua community to understand a flow model of the harbour created at the Geospatial Research Institute Toi Hangarau in collaboration with PHF Science. People can create story maps to describe the environmental data, as well as interact with overviews and detailed plots of flows within the harbour to understand how the catchment, streams, tide and rain contribute to sedimentation, flushing, or contaminant buildup.
Current and near-future developments of EDDIE include optimisations and templates for cloud deployments. Focusing on facilitating cloud deployments allows for dynamic scaling to occur, allowing for large amounts of processing power to be accessed for only the short amount of time needed. This will be invaluable for FReDT allowing us to run many proposed scenarios at once for communities. EDDIE was built on a containerised architecture, and these additional developments will remove barriers to deploying new EDDIE projects.
EDDIE provides a framework for building environmental Digital Twins with interoperable standards. This framework will help adoption of new Digital Twins and strengthen the community ecosystem of environmental Digital Twins. This will enhance access to data and insights for communities, for planning, for decision making and for research.
- Madni A, Madni C, Lucero S. Leveraging Digital Twin Technology in Model-Based Systems Engineering. Systems. 2019 Jan 30;7(1):7.
- Deren L, Wenbo Y, Zhenfeng S. Smart city based on digital twins. Comput Urban Sci. 2021 Dec;1(1):4.
- Blair GS. Digital twins of the natural environment. Patterns (N Y). 2021 Oct 8;2(10):100359.
- Bauer P, Stevens B, Hazeleger W. A digital twin of Earth for the green transition. Nat Clim Chang. 2021 Feb;11(2):80–3.
- European Commission. Destination Earth (DestinE) [Internet]. 2022 [cited 2022 Apr 18]. Available from: https://digital-strategy.ec.europa.eu/en/policies/destination-earth
- Hoffmann J, Bauer P, Sandu I, Wedi N, Geenen T, Thiemert D. Destination Earth – A digital twin in support of climate services. Clim Serv. 2023 Apr;30:100394.
- Autiosalo J, Siegel J, Tammi K. Twinbase: Open-Source Server Software for the Digital Twin Web. IEEE Access. 2021;9:140779–98.
TonyLiu; TBD (oral presentation or poster)
1-Introduction and Background
The geography of ancient Egypt and its mythology are closely connected, with the annual flooding of the Nile shaping both the physical landscape and the cultural worldview. In ancient Egyptian cosmology, the world emerged from a primordial watery void (Nun) as the first land, or "primeval mound". This study focuses on Esna, Upper Egypt, home to the Temple of Khnum and several other unexcavated or demolished temple sites. Inscriptions from the temple, such as the Festival of Seizing the Staff, metaphorically describe the local terrain as marshlands and document processional routes. While traditional Egyptological methods have correlated some text-based place names with physical locations, they often lack quantitative spatial analysis, leaving the precise geomorphological context of these narratives largely undefined. Macro-topography remains stable for a long time, barring massive modern anthropogenic intervention. This research adopts a landscape-first approach. By leveraging open-source geographic information systems (GIS) and remote sensing, we mathematically translate qualitative ancient texts into a DEM-driven topographic model to reconstruct the historical landscape that inspired the legends.
2-Study Area and Open Data Sources
The study area is Esna, with five documented temple sites surrounding it: Temple of Khnum, House of God, the Temple of Isis, and Kom Mir. The construction of the Aswan High Dam has masked historical paleochannels and floodplains; insufficient data are available for hydrological reconstruction. Consequently, this study heavily prioritizes Digital Elevation Models (DEMs). We lean on declassified 1960s CORONA panoramic stereo pairs to extract a pristine bare-earth microtopography, capturing the landscape long before recent agricultural expansion and infrastructure projects leveled it. We then complement this historical topographic baseline with decades of multispectral data from Sentinel-2 and the Landsat program (Landsat 1–9) to monitor remaining vegetation patterns and water indices. Aligning with the core philosophy of FOSS4G, our primary computational platform (QGIS) and all incorporated remote sensing datasets are entirely open-access.
3-Methodology
To bridge the gap between mythological narratives and spatial reality, we developed a reproducible, DEM-centric workflow using QGIS and Python spatial libraries. The best method we can use is layer analysis, which can turn qualitative contents into different GIS layers. The layers can be easily linked and operated with different calculations. The methodology includes three parts:
First, textual descriptions are converted into topologically validated GIS layers. Layer 1 represents the "Primeval Mound" (elevated, unflooded zones), Layer 2 represents "Marshes & Lakes" (low-lying retention basins), and Layer 3 delineates the "Western Mountains" (the absolute safety boundary).
Second, we execute rigorous Topographic Surface Modeling. Based on terrain stability, we use QGIS terrain analysis algorithms to compute critical geomorphological variables from the historical DEM. By calculating slope, aspect, Topographic Position Index (TPI), and the Topographic Wetness Index (TWI), we quantitatively define the physical landscape characteristics, isolating potential ancient mounds from natural depressions.
Third, we perform Hydro-conditioning and Simulated Routing. Since modern hydrology is severely disrupted, we reverse-engineer the ancient floodpaths through the terrain. We apply open-source algorithms to hydro-condition the DEM—executing pit-filling, flow-direction, and flow-accumulation routing—to establish a hydrologically correct surface. A terrain-based inundation algorithm, such as the Height Above Nearest Drainage (HAND) model, is then deployed. By routing simulated water levels across this stable topography, the inundation results are meticulously calibrated to match the spatial descriptions in the ancient texts.
4-Preliminary Results and Archaeological Implications
Currently, the DEM-driven spatial analysis successfully bridges the text-to-terrain gap. Our analysis of Layer 3 (the "Western Mountains") exposes a significant divergence between mythological narratives and physical geography. While the inscriptions describe this western margin as an impassable, absolute safety zone, the DEM analysis reveals it to be a modest ridge with an average elevation of only 200 to 300 meters. Furthermore, extracting elevation profiles from our decadal remote-sensing time series confirms that this topography has remained geomorphologically static, ruling out historical degradation. This discrepancy suggests that the ancient characterization was a phenomenological exaggeration, likely stemming from limited mobility and the imposing visual perspective of looking westward from the low-lying Esna basin. Identifying this spatial hyperbole is archaeologically significant; it demonstrates a broader tendency for cognitive exaggeration within the temple texts, providing a critical, data-driven foundation for reinterpreting other geographic claims in the inscriptions.
Our geomorphological computations explicitly identified elevated landforms and topographical depressions that align flawlessly with the text-derived topological layers. The hydro-conditioned HAND model effectively simulates the historical flood recession zones, demonstrating how rising waters would inundate the natural basins (Layer 2) and expose the structurally sound primeval mounds for temple construction (Layer 1).
5-Conclusion and Future Work
This study establishes a robust methodological blueprint for landscape archaeology. By pivoting from a purely hydrological focus to a terrain-driven analysis, we demonstrate how open-source GIS tools can bypass modern infrastructural disruptions to reconstruct ancient environments. The calibrated, DEM-based inundation model not only contextualizes the past but also serves as a predictive tool.
Future work will focus on:
First, we will broaden the temporal depth of our spatial database by incorporating and georeferencing early historical cartography (e.g., 18th- and 19th-century expedition maps). This process will cross-validate our findings and further corroborate the millennial topographic stability of the Esna geomorphology.
Second, we will expand our textual dataset—currently focused on the Festival of Seizing the Staff—to include inscriptions from other major local events, such as the Festival of Raising the Sky. Cross-referencing these distinct mythological narratives will allow us to generate additional topological layers and conduct rigorous statistical evaluations to assess their spatial validity and accuracy.
References:
[1] Sauneron, S. (1962). Le temple d'Esna. Tome V: Les fêtes religieuses d'Esna aux derniers siècles du paganisme. Le Caire: IFAO.
[2] Abdel-Raham, A. M. (2009). The Lost Temples of Esna. BIFAO, 109, 1-8.
[3] Assmann, J. (1996). The Mind of Egypt: History and Meaning in the Time of the Pharaohs. Metropolitan Books.
Xiandong Cai; TBD (oral presentation or poster)
Bare-earth Digital Elevation Models (DEMs) or Digital Terrain Models (DTMs) are fundamental to geospatial applications, from flood modelling and landslide assessment to infrastructure planning and environmental management. However, original publicly accessible global elevation products, such as SRTM (Shuttle Radar Topography Mission), ASTER GDEM (Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Model), and Copernicus DEM, represent Digital Surface Models (DSMs). DSMs are elevation models that include canopy heights and building structures, rather than true bare-earth topography. Their vertical accuracies (how closely elevations match the real ground height) typically range from 4–15 m RMSE (root mean square error), and their spatial resolutions (size of the smallest discernible detail) are constrained to 30 m or coarser. Airborne Light Detection and Ranging (LiDAR) derived DTMs achieve sub-meter vertical accuracy and spatial resolution because lasers/photons penetrate vegetation to measure ground elevation. However, high-cost LiDAR surveys lead to fragmented coverage, even in developed countries. For example, about 16% of New Zealand’s land surface still lacks airborne LiDAR mapping, resulting in critical data gaps in the very remote, rugged, and densely vegetated terrain where elevation information is most important.
Recent deep learning approaches have the potential to enhance the vertical accuracy and spatial resolution of global DEMs through super-resolution techniques. For example, JSPSR (Joint Spatial Propagation Super-Resolution) networks improve Copernicus GLO-30 DSM from 30 m to 3 m spatial resolution by utilising high-resolution remote sensing imagery, reducing elevation RMSE by over 70% across diverse sites. However, their performance drops in dense forest canopies. Optical sensors cannot penetrate vegetation, so models must infer ground elevation from indirect cues (such as estimating canopy height or shape). Spaceborne LiDAR missions, such as ICESat-2’s ATL08 product, provide global terrain measurements that can penetrate vegetation. However, incorporating these measurements poses two main challenges. First, after rigorous data quality filters, ATL08 photons can cover as little as 0.1–0.2% of dense-forest mountain areas. Second, spaceborne LiDAR provides sparse point data, whereas optical imagery and DEMs are dense raster grids, resulting in a data-geometry mismatch that most conventional network architectures cannot inherently accommodate.
This study describes an open-source deep neural network framework that tackles these two challenges using a triple-branch multi-modal fusion network. It processes three complementary data streams: high-resolution remote sensing imagery via a Swin Transformer encoder, interpolated Copernicus GLO-30 DEMs via a parallel Swin encoder, and serialised ATL08 along-track data via a sparse encoder that preserves measurement features without rasterisation or interpolation. All code and trained models will be released under an open-source license to support open-science principles.
The main innovation is a multi-scale deformable cross-attention mechanism that enables effective fusion of sparse LiDAR measurements with dense raster data. At each scale, individual ATL08 photons independently query the surrounding image and DEM features through learned deformable sampling patterns, allowing each photon to adaptively sample contextual information most relevant to elevation prediction at its location. Inspired by deformable DETR, the deformable cross-attention mechanism implements bidirectional information flow: image and DEM features inform the interpretation of photon measurements during downscaling, while photon-derived elevation features are injected back into the feature maps during upscaling. This design provides that sparse yet accurate LiDAR measurements guide feature extraction, while dense image context enriches photon representations, addressing the key challenge of multi-modal feature fusion in open geospatial data science.
Ultimately, the Spatial Propagation Network (SPN) transforms sparse DSM grids into dense predictions by conditioning content-adaptive kernels on fused multi-modal features. Through multiple propagation iterations, corrections propagate along paths guided by image content, following ridgelines, thalwegs, or areas with similar vegetation characteristics, while re-injecting precise photon measurements at each iteration to retain accuracy at known locations.
Sparse-to-dense progressive supervision computes elevation loss only at ATL08-confirmed locations (typically fewer than 64 per 256×256 training patch), whereas multi-scale deep supervision heads propagate gradient information throughout the network, even in areas without direct elevation constraints. This strategy prevents learning invalid correlations in unobserved areas while retaining end-to-end differentiability.
We evaluate our approach using the DFC30 dataset, which we augment with spatially matching ATL08 measurements. The training set contains 12,728 image-DSM-photon tuples, and the test set contains 3,196, both with airborne LiDAR ground truth. Our results show considerable improvements versus the baseline method, JSPSR. Across all test sites at 3 m spatial resolution, elevation accuracy (RMSE) improves by 8% in open terrain (from 1.1 m to 1.01 m) and 28% in dense forest canopies (from 6.7 m to 4.8 m). Overall, we achieve a vertical accuracy of 3.2 m for all vegetation classes. The largest improvements occur where optical-only methods perform weakest. In indigenous forests with dense understory and complex terrain, our method decreases systematic bias from 15.1 m to 8.8 m.
All code, trained weights, and preprocessing tools will be public under an open-source license. This ensures complete repeatability and allows community adaptation. The trained model enables end-to-end 3 m DTM generation for unmapped areas of New Zealand using global DEMs. It will produce a seamless bare-earth elevation product at a national scale, covering about 268,000 square kilometres. This supports applications such as landslide mapping, hydrological analysis for freshwater management, carbon stock assessment in forests, and infrastructure planning in rural areas.
For the FOSS4G community, this work makes three key contributions. First, it delivers a practical, open-source solution for generating high-resolution bare-earth DTMs by fusing global datasets (Copernicus DEM, remote sensing imagery, and ICESat-2 ATL08) using reproducible methods. Second, it provides an architectural strategy that respects the geometry of different data modalities, e.g., constructed tables, sparse point clouds and dense rasters, by using a template for multi-modal fusion in open geospatial science. Third, it shows that open data and software can solve real-world data gaps to produce operational products that benefit communities, environmental management, and disaster resilience. More broadly, our work supports ongoing efforts in the open geospatial community to improve global terrain characterisation by fusing diverse Earth observation assets, making high-quality elevation data accessible to all.
TonyLiu, Shichao Wang; TBD (oral presentation or poster)
1-Introduction and Study Area
Cladophora is a filamentous green alga native to the North American Great Lakes. Its excessive proliferation not only causes foul odors and impairs public beach recreation but also triggers severe ecological issues, including avian botulism outbreaks. Since the 1990s, the filtering effect of invasive species such as dreissenid mussels has significantly increased water clarity, allowing sunlight to penetrate to greater depths. This has led to massive Cladophora blooms even under relatively low nutrient concentrations. The study area of this research focuses on the nearshore waters along the southern shore of Lake Ontario (the United States side). To achieve precise calibration of remote sensing observations, the spatial scope of the study is strictly defined as two independent 6 km × 6 km square regions, centered respectively around two key hydrological and biological monitoring stations established by the United States Geological Survey (USGS): the OIR station (Irondequoit, near Rochester) and the OOL station (Olcott).
These two core USGS stations provide substantial, highly valuable ground-truth data for this study. These comprehensive datasets encompass multi-depth water flow velocities, water turbidity, and various critical chemical constituents in the water column (such as nutrient concentrations). More importantly, the stations provide net weight data of Cladophora samples collected in situ across different depth gradients. These multi-dimensional, high-precision ground truth indicators not only serve as an irreplaceable validation foundation for evaluating and calibrating various spectral remote sensing indices within our open-source computational architecture, but also enable us to deeply investigate the complex mechanisms underlying the relationships between micro-environmental physicochemical variables and nearshore benthic algal outbreaks.
2-Evaluation of Traditional Indices and Experimental Derivation of a Novel Index
In the preliminary remote sensing analysis phase, we developed a Python-based workflow to extract Sentinel-2 image bands and automatically calculated various traditional spectral indices, including NDVI, FAI, NDAVI, and SABI. Statistical analysis of multi-temporal imagery (from May to August 2023) revealed that the mean and median values of these indices were frequently negative or extremely low, accompanied by disproportionately large standard deviations. For instance, across multiple summer observation dates, the median values for NDVI and FAI consistently hovered near zero (ranging from -0.012 to 0.025). At the same time, NDAVI and SABI exhibited even deeper negative medians (often between -0.05 and -0.09). Furthermore, the high standard deviations—frequently exceeding 0.25 for NDVI and 0.50 for SABI—demonstrated massive signal noise. This statistical analysis demonstrates that vegetation indices based on the Near-Infrared (NIR) band exhibit severe absorption failures in aquatic environments, rendering them inadequate for precise mapping of submerged benthic Cladophora.
To address this optical challenge and identify the optimal spectral response, we designed a controlled physical experiment. A 3m × 3m water tank was used, with an incandescent light source simulating solar irradiance. A receiver simulated the satellite sensor to capture reflectance from a green surrogate representing benthic algae. Strikingly, the experimental results revealed that the strongest reflectance signals emerged in the Blue and Short-Wave Infrared (SWIR) bands, significantly diverging from the band selections of traditional vegetation indices. Based on these empirical findings, we are currently conducting rigorous mathematical derivations utilizing the Blue and SWIR bands to formulate a novel, water-penetrating spectral index specifically optimized for Cladophora detection.
3-Automated Open-Source Cloud-Masking Algorithm to Bypass API Limitations
To achieve high-frequency monitoring of Cladophora, we aimed to build a fully open-source, automated data acquisition architecture. However, querying the Copernicus Data Space API inevitably encounters strict request frequency limits and download volume quotas. Furthermore, the official API only provides the average cloud cover percentage at the full-scene level. For our small 6 km × 6 km Region of Interest (ROI), this macroscopic cloud assessment is highly inaccurate. A scene with a low average cloud percentage might still have dense clouds completely obscuring our study area, leading to massive invalid downloads and wasted bandwidth. Additionally, a single remote sensing image rarely covers the target area perfectly without clouds, necessitating the seamless mosaicking of multiple images and stricter screening for high-quality data.
To overcome this core bottleneck, we designed and implemented a regional cloud-masking algorithm based on image Quicklooks (previews) within our workflow. Since Quicklook files are extremely small and consume negligible download bandwidth, the program automatically prioritizes retrieving them. Given that Quicklooks do not inherently contain geographic coordinates, the algorithm first extracts the boundary coordinates of the scene's footprint polygon from the metadata. Subsequently, it correlates and standardizes the ROI's geographic coordinates against this footprint boundary. Based on this geometric translation, the system can precisely reverse-engineer the specific pixel rectangle corresponding to the study area on the unreferenced Quicklook image. Ultimately, the algorithm computes the proportion of white pixels exclusively within this localized bounding box to accurately assess the true cloud cover rate within the ROI. Only when the ROI's cloud cover meets strict clear-sky thresholds does the system automatically trigger the API to download the heavy, high-resolution original imagery. This algorithm successfully achieves precise "on-demand downloading," effectively circumventing API bandwidth restrictions while dramatically improving the efficiency of acquiring the cloud-free data required for subsequent image mosaicking.
4- Conclusion and Future Works
This study successfully established a highly efficient, Python-based open-source remote sensing download architecture that practically circumvents API limitations. It also highlighted the severe shortcomings of traditional vegetation indices through both satellite data statistics and controlled physical experiments. Future research will focus on advancing two primary tasks:
First, further refining the Quicklook-based cloud-masking algorithm to automate the acquisition of extensive multi-temporal imagery for seamless spatial mosaicking. To ensure complete reproducibility, this process will be integrated into an end-to-end Python pipeline, with the full source code made freely available on GitHub.
Second, finalizing the mathematical formulation of our novel Blue-SWIR spectral index based on the water tank experiment, and deploying it within our open-source pipeline to precisely map the spatial distribution and evolutionary dynamics of Cladophora during peak summer blooms.
References:
[1] Howell, E. T. (2018). A decadal-scale perspective on the occurrence of Cladophora on the north shore of Lake Ontario. Environmental Monitoring and Assessment.
[2] Wright, N., et al. (2024). CloudS2Mask: A novel deep learning approach for improved cloud and cloud shadow masking in Sentinel-2 imagery. Remote Sensing of Environment, 306, 114122.
[3] Copernicus Data Space Ecosystem. (2024). Quotas and Limitations Documentation.
Dr.Victor N.Sunday, Grace Martins-Ateli; TBD (oral presentation or poster)
Coastal deltaic environments are highly dynamic landscapes where human activities interact with climate-related processes, including shoreline change, flooding, and wetland degradation. The Niger Delta of southern Nigeria contains one of Africa’s largest mangrove and wetland systems, supporting extensive estuarine networks and rapidly expanding coastal settlements. Over recent decades, increasing urbanization and agricultural expansion have significantly altered land use and land cover (LULC) patterns across the region. Despite numerous regional studies, reproducible workflows that integrate multi-decadal change detection, landscape fragmentation analysis, spatial driver assessment, and scenario modelling within a unified open-source geospatial environment remain limited. This study, therefore, applies a QGIS-based open-source remote sensing framework to analyze long-term coastal LULC dynamics in the Niger Delta between 1986 and 2026, identify spatial drivers of land conversion, and simulate possible land-cover trajectories for 2050 to support climate-resilience planning. Multi-temporal Landsat surface reflectance imagery from Landsat 5 TM (1986), Landsat 7 ETM+ (2000), Landsat 8 OLI (2013), and Landsat 9 OLI-2 (2026 composite) was obtained from the USGS Earth Explorer archive. Image preprocessing procedures included cloud masking and band stacking of Landsat imagery within the QGIS Semi-Automatic Classification Plugin (SCP) to support supervised classification and improve thematic separability of landcover classes. Supervised classification was implemented using the Random Forest algorithm with a stratified training and validation sampling strategy. Six LULC classes were mapped: mangrove forest, freshwater wetlands, built-up areas, agricultural land, bare surfaces, and water bodies. Classification accuracy was evaluated using overall accuracy and the Kappa coefficient to ensure the reliability of multi-temporal change detection. All analyses were performed using open-source geospatial tools and publicly available datasets to ensure methodological transparency and reproducibility consistent with FOSS4G principles. The classification results indicate strong model performance across all epochs. Overall accuracy increased from 86.2% (κ = 0.83) in 1986 to 90.2% (κ = 0.88) in 2026, confirming the robustness of the classification workflow implemented within the open-source QGIS environment. Quantitative LULC analysis reveals significant transformation across the coastal landscape over the 40-year study period. Mangrove extent declined from 2456 km² in 1986 to 1978 km² in 2026, representing a loss of approximately 478 km² (−19.5%). Freshwater wetlands decreased from 3789 km² to 3098 km², corresponding to a loss of 691 km² (−18.2%). In contrast, built-up areas expanded substantially from 457 km² in 1986 to 1290 km² in 2026, representing an increase of 833 km² (+182.3%). Agricultural land also expanded from 1235 km² to 1935 km², corresponding to a 56.7% increase. These trends indicate sustained conversion of natural coastal ecosystems into urban and agricultural landscapes across the Niger Delta. Landscape fragmentation analysis further reveals structural degradation of mangrove and wetland ecosystems. Mangrove patch density increased from 0.51 patches/km² in 1986 to 1.13 patches/km² in 2026, while mean patch size declined from 1.97 km² to 0.89 km². Freshwater wetlands exhibit similar fragmentation patterns, with patch density increasing from 0.62 to 1.12 patches/km² and mean patch size decreasing from 1.62 km² to 0.90 km². Increasing edge density and decreasing patch size indicate growing spatial fragmentation of coastal ecosystems against coastal flooding and environmental disturbance. Spatial driver analysis was conducted using terrain, accessibility, demographic, and climatic variables derived from open datasets. Elevation and slope were derived from the Shuttle Radar Topography Mission (SRTM) digital elevation model, while accessibility gradients were represented using Euclidean distance to roads, shoreline boundaries, and urban centers derived from OpenStreetMap data. Population density data from WorldPop and rainfall data from the CHIRPS dataset were incorporated to evaluate potential socio-environmental influences on land conversion. Correlation analysis indicates that population density (r = 0.88), proximity to urban centers (r = 0.82), and accessibility to roads (r = 0.78) exhibit the strongest positive associations with built-up expansion. Conversely, mangrove and wetland losses show negative associations with population density (−0.62 and −0.58 respectively), suggesting strong anthropogenic pressure on coastal ecosystems. Future LULC scenarios for 2050 were simulated using the Cellular Automata–Markov (CA–Markov) model implemented through the MOLUSCE plugin within QGIS. Transition probabilities derived from the 1986–2026 period were combined with suitability surfaces generated from spatial driver variables. Under a business-as-usual scenario, built-up areas are projected to increase from 1289 km² in 2026 to approximately 2157 km² by 2050, representing a 67.2% expansion. Concurrently, mangrove and wetland ecosystems are projected to decline by approximately 15.2% and 13.6%, respectively. Alternative scenario simulations indicate that conservation-oriented interventions could partially stabilize mangrove and wetland systems, while development-intensive trajectories could accelerate ecosystem loss. Beyond regional findings, this study demonstrates that multi-decadal LULC change detection, landscape fragmentation assessment, spatial driver modelling, and land-use scenario simulation can be implemented entirely within a reproducible open-source geospatial framework. By integrating SCP, GRASS GIS tools, and the MOLUSCE plugin within QGIS, the research provides a transparent analytical workflow aligned with the principles of Free and Open-Source Software for Geospatial (FOSS4G). The results highlight accelerating coastal transformation in the Niger Delta and emphasize the importance of integrating historical land-change analysis with forward-looking modelling to provide spatial evidence relevant to long-term coastal management and planning.
Keywords: Coastal LULC change, Niger Delta, QGIS, Random Forest classification, Landscape fragmentation, CA–Markov modelling, MOLUSCE plugin, open-source GIS, FOSS4G.
Narumasa Tsutsumida, Irhamillah; TBD (oral presentation or poster)
Biological invasions represent one of the most significant threats to global biodiversity and agricultural systems, causing substantial ecological and economic damage worldwide. Among emerging invasive pests in East Asia, Aromia bungii, commonly known as the red-necked longhorn beetle, has become a serious threat in Japan. The species attacks several Prunus species, including ornamental cherry trees (Cerasus spp.), peach (Prunus persica), and plum (Prunus salicina). Because cherry trees play an important ecological and cultural role in Japan, the spread of this invasive beetle has raised growing concerns for landscape management and biodiversity conservation.
Since its first detection in Aichi Prefecture in 2012, A. bungii has expanded rapidly across urban and peri-urban areas. Understanding its spatial pattern is therefore essential for effective monitoring and early intervention. Spatial analysis can be applied to these processes. However, such analyses fundamentally depend on how spatial data are defined, including the geometry of the spatial grid, which can influence the results.
Thus, to examine how grid shape influences spatial analysis results, this study evaluates spatial autocorrelation measures using different tessellations of invasive species occurrence data and environmental variables. Specifically, we compared rectangular and hexagonal grids for analysing spatial patterns in A. bungii occurrence records and density of rivers in Saitama Prefecture, Japan.
Volunteer-based occurrence data for A. bungii were compiled from field surveys across Saitama Prefecture from 2017 to 2023, yielding 2,412 confirmed presence records. Records were classified as confirmed presences if either adult beetle observations (Adult-yes = 1) or evidence of tree damage (Tree_damage = 1) was recorded. All records were georeferenced using latitude–longitude coordinates (WGS84) and then reprojected to UTM Zone 54N (EPSG:32654) to ensure metric accuracy in spatial calculations. We assembled the occurrence data into predefined grid cells to explore spatial patterns across the study area and to enable consistent spatial aggregation and neighbourhood-based analyses.
We calculated density of rivers networks in Saitama, such original data were obtained by Digital Map (Basic Geospatial Information) of Geospatial Information Authority of Japan.
Two grid tessellation schemes were constructed over the study area. The rectangular grid consisted of 6,372 cells at a 1 km × 1 km resolution. Each rectangular cell contained pre-calculated directional river and road connectivity values (normalised lengths: 0–1) in four directions: top, bottom, left, and right. This configuration corresponds to rook contiguity, where only four directly adjacent neighbours are considered. To evaluate the effect of diagonal bias, the same rectangular grid was also analysed using queen contiguity, which includes eight neighbouring cells by incorporating both direct and diagonal neighbours. The hexagonal grid was generated using the H3 hierarchical spatial indexing system, producing hexagonal cells with an equivalent spatial resolution to the rectangular grid. To keep the overall grid coverage comparable to the rectangular representation, 6,392 hexagonal cells were generated to cover the study area. The rectangular grid had a side length of 1.0 km, whereas the hexagonal grid had a slightly larger side length of 1.074 km. H3 provides a discrete global grid system based on hexagonal indexing, enabling consistent spatial aggregation and neighbourhood relationships.
We found that grid type and spatial-weight configuration influenced the global Moran’s I results. For spatial autocorrelation, the rectangular grid produced Moran’s I values of 0.5373 (p = 0.001) under rook contiguity and 0.4636 (p = 0.001) under queen contiguity. Meanwhile, the hexagonal grid produced a lower value of 0.4144 (p = 0.001). The hexagonal grid yielded the lowest Moran’s I because all six shared edges are equidistant.
The higher value observed in the rectangular grid reflects inflated clustering due to diagonal adjacency, where diagonal neighbours are treated as equivalent to directly adjacent cells despite being farther apart. In contrast, the hexagonal grid provides a more geometrically balanced structure because all neighbouring cells are equidistant.
To further investigate the influence of weight configuration, we designed a distance-weighted queen scheme by assigning diagonal neighbors a weight of 1/√2. Moran’s I was 0.4732 (p = 0.001). Comparing Moran’s I across configurations in descending order (rook, queen, distance-weighted queen, and hexagonal), we found that the degree of spatial autocorrelation can be influenced by the choice of spatial grid, although all values were positive and statistically significant.
For river network density, all grid configurations produced high Moran’s I values (rectangular rook: 0.8365; rectangular queen standard eight: 0.8345; rectangular queen distance-weighted: 0.8349; and hexagonal: 0.5170), with p = 0.001 in all cases, indicating strong spatial clustering characteristics.
In both cases, the rectangular grid produced higher Moran’s I values under both rook and queen contiguity, indicating stronger spatial clustering. Theoretically, rook contiguity captures neighbourhood effects directly, but limits spatial association to four directions. Queen contiguity considers eight neighbours, but treats diagonal neighbours, which are farther away (1.414 km), as equivalent to direct neighbors at 1.0 km. Even when a distance-weighted scheme was applied to the queen configuration, the Moran’s I values changed little. In contrast, the hexagonal grid produced the lowest Moran’s I values in both experiments, suggesting greater stability because it uses six neighbors at equal distances.
In conclusion, we found that the choice of grid system is important for quantifying spatial autocorrelation. The rectangular grid tended to yield higher Moran’s I values, whereas the hexagonal grid tended to yield lower values. Further analyses should be conducted, specifically to investigate spatial clustering patterns using local Moran’s I under different spatial configurations.
Keywords: hexagonal grid, rectangular grid, tessellation, spatial autocorrelation, Moran's I, invasive species, Aromia bungii, Saitama, Cellular Automata
Arissara Sompita; TBD (oral presentation or poster)
Geographic Information System (GIS) data is inherently multidimensional, encompassing spatial extent, temporal dynamics, and diverse attributes. Understanding and presenting this complex data through cartographic visualization requires both comprehensive geospatial knowledge and sophisticated map design skills. However, a significant portion of geographic platform users—including web developers, data analysts, and professionals without formal cartography training—frequently encounter substantial challenges throughout the map visualization workflow. These difficulties span from initial data interpretation and selection of appropriate visualization methods to the configuration of critical cartographic elements such as color schemes, symbol sizes, opacity levels, and overall compositional layout.
In contemporary web mapping systems, data visualization is predominantly controlled through Map Style JSON, a structured format that governs data rendering in mapping libraries such as MapLibre GL JS, which operates according to the MapLibre Style Specification standard. While this specification is open and highly flexible, creating appropriate styles from actual datasets remains a task demanding both technical expertise and design experience. Consequently, many users invest considerable time manually experimenting with style adjustments, often through trial and error, which can be both frustrating and inefficient.
This presentation proposes an innovative approach to simplifying the map design process through the development of a Model Context Protocol (MCP) for automated Map Style JSON generation from users' spatial vector data. This data can be sourced through database connections or API service integrations. The fundamental concept underlying MCP is the creation of a "context layer" that enables language models to systematically understand the structure and semantic meaning of GIS data before applying this understanding to map style generation.
The proposed architecture integrates MCP with open-source Large Language Models (LLMs) capable of operating locally through Ollama. The system analyzes users' spatial data characteristics, including geometry types, attribute structure, and data distribution patterns. Subsequently, it automatically generates MapLibre-compliant Map Style JSON that can be immediately deployed in web mapping applications. This local processing capability addresses both performance and data privacy concerns that often arise with cloud-based solutions.
The distinctive advantage of this approach lies in how MCP extends beyond merely ensuring structurally correct JSON generation. The protocol fundamentally incorporates cartographic design principles based on the "Perceptual Properties of Linear and Spatial Systems" into the decision-making process. This integration manifests in several critical ways: the selection of color schemes aligned with data semantics, the assignment of appropriate symbols corresponding to geometry types, and the strategic application of color tones and opacity levels to enhance user perception and readability. The system also accommodates datasets with multiple classification classes and leverages modern color palettes to ensure that spatial data visualization achieves both clarity and accessibility.
The technical implementation combines several key components working in concert. First, the MCP server acts as an intermediary layer that processes incoming spatial data, extracting relevant metadata and structural information. This includes analyzing coordinate reference systems, identifying attribute data types, detecting statistical distributions, and recognizing spatial patterns that inform styling decisions.
The language model component, running locally through Ollama, receives this contextualized information and applies learned cartographic principles to generate appropriate styling rules. The model has been trained to understand the relationships between data characteristics and visual representation best practices. For instance, when encountering categorical data with distinct classes, the system automatically selects qualitatively different colors that maximize perceptual distinction. For continuous numerical data, it applies sequential or diverging color schemes appropriate to the data's semantic meaning.
The generated Map Style JSON adheres strictly to MapLibre specifications, ensuring immediate compatibility with MapLibre GL JS and other compliant rendering engines. The output includes properly structured layers, sources, paint properties, and layout configurations that reflect both the data's inherent characteristics and established cartographic conventions.
A critical innovation of this approach involves embedding cartographic design expertise directly into the generation process. Traditional automated styling systems often produce technically correct but cartographically naive outputs. This MCP-based system incorporates several levels of design intelligence:
Perceptual hierarchy: The system understands which data elements should be visually prominent and adjusts styling properties accordingly, considering factors such as feature importance, scale-dependent visibility, and visual contrast.
Color theory application: Beyond simple color assignment, the system applies principles of color harmony, considers color blindness accessibility, and ensures adequate contrast ratios for legibility across different display conditions.
Symbolic representation : The selection of point symbols, line patterns, and fill styles reflects both the semantic meaning of the data and established cartographic conventions, making maps intuitively interpretable even for non-expert users.
Scale responsiveness: Generated styles include appropriate zoom-level dependencies, ensuring that map elements appear at suitable scales and with appropriate levels of detail.
The objectives of this presentation extend beyond merely demonstrating automated Map Style JSON generation. Fundamentally, this approach democratizes quality map production, enabling individuals without GIS backgrounds to create professional-quality cartographic visualizations. This democratization has significant implications for data journalism, civic participation, educational applications, and small organizations that lack dedicated GIS expertise.
Furthermore, the utilization of open-source models and architecture capable of local execution through Ollama aligns perfectly with the principles of the Open Geospatial Ecosystem. This approach ensures data sovereignty, eliminates dependency on proprietary cloud services, and facilitates integration with other open-source tools prevalent in the geospatial community. The system can be extended and customized by users, fostering innovation and adaptation to specific domain requirements.
The protocol-based architecture also enables future enhancements and integrations. As language models continue to evolve, the MCP layer provides a stable interface that can leverage improved capabilities without requiring fundamental system redesign. Additionally, the approach can be extended to incorporate user feedback, learning from styling preferences and iteratively improving recommendations.
In this presentation, our goal is not only to automatically generate Map Style JSON from your data but, more importantly, to empower individuals without GIS expertise to create high-quality map visualizations with ease. Additionally, by utilizing an open model and architecture that can operate on users' devices via Ollama, we align with the principles of an open geospatial ecosystem and enable integration with other open-source tools within the community.