Monitoring, modeling, and forecasting the Earth system rely on an ever-growing stream of spatially and temporally resolved data—from satellite observations to model outputs . Such Earth system data are getting increasingly larger, higher-dimensional and complex. Many of these data, in particular outputs of Earth system models and regional to global Earth observation products, are based on regular grids in space and time. Higher dimensions of these arrays may correspond to multiple variables or model ensemble versions. Such data are increasingly stored as so called analysis-ready-data-cubes (ARDCs), i.e., data organized along spatial, temporal, and thematically gridded dimensions . Paired with cloud-native file formats, they enable efficient querying of large spatiotemporal datasets.

However, as Earth system datasets grow in size and dimensionality, scientific analysis, data processing and visualization become more challenging . Capable visualization tools for exploring large Earth system datasets are crucial for scientists to still be able to easily access and intuitively understand such data . To address these challenges, we previously developed https://www.lexcube.org (last access: 17 March 2026), a web-based visualization tool for data cubes see Fig. ;, and Lexcube for Jupyter Notebooks, an open-source Python package (https://github.com/msoechting/lexcube, last access: 17 March 2026), that allows intuitive exploration of data cubes as part of scientific workflows . These developments enable users to interact with their data visualized as three-dimensional cubes in a virtual environment, where the side faces display dynamically changing Hovmöller diagrams  to reveal spatiotemporal patterns. This setup enables rapid assessment of data quality and structure, making spatial or temporal gaps, anomalies, and artifacts immediately visible. Key insights, such as the presence of unexpected patterns, can thus be identified early in the analytic workflows or be effectively communicated to students or the general public.

Map-based tools, i.e., two-dimensional visualizations, completely lack an intuitive view on spatiotemporal dynamics as they show only data at single points in time at once. There, the only visualization of temporal dynamics specifically happens with animations, which however have been shown in other fields to be less effective at communicating information than static depictions of the same data . 3D data cube visualization tools like Lexcube offer this static and interactive view on temporal dynamics through the Hovmöller diagrams on the side faces, in addition to animations through time. However, limited to regular two-dimensional computer displays, Lexcube renders the 3D data cube onto a 2D screen, which may present navigation and orientation issues to users and hinder intuitive understanding. In order to have a truly intuitive three-dimensional interaction and understanding, the interactive device needs to have a three-dimensional form, e.g., in the form of a physical installation.

While interactive physical installations featuring Earth observation data and model outputs exist, e.g., in NASA's Earth Information Center

https://science.nasa.gov/science-research/earth-science/nasas-new-exhibit-showcases-our-home-planet-and-climate/

The interactive physical data cube has five square touch-enabled screens, representing five of the six sides of the virtual data cube (see Fig. ). The bottom side is left out for easier construction and improved air circulation. To guide users and allow advanced operations, there is a separate overview monitor mounted on a tripod stand. It shows a virtual 3D data cube, which is synchronized to the physical data cube, to give an overview in an interface similar to our previously developed software at https://www.lexcube.org. The monitor also shows meta data and the currently selected cube boundaries to help users understand what each cube side is currently visualizing. Furthermore, it is also possible to switch the currently visible dataset, make a precise data selection with sliders, play an animation through any dimension as well as adjust the colormap. See also the attached supplementary video file for a video demonstration of the physical data cube and its possible interactions.

On the software side, the interactive data cube uses existing software components from our previous work on https://www.lexcube.org. Inside of the physical data cube, there is a single PC connected to all five cube screens and the overview monitor, with one video output and one USB touch connection each. The PC is running a single Lexcube visualization server as well as multiple Lexcube visualization clients, one for each screen. All five clients synchronize in regards to any interaction or selection being made, achieving the effect of an interactive data cube as all five screens show the same virtual data cube from different angles. To automatically restore this setup whenever the PC is booted, an auxiliary Python script starts all software components and arranges them on the multiple displays correctly, making the interactive physical cube ”plug-and-play” with just a single power cable. The server component pre-processes all ingested datasets into tiles ahead of time. These tiles contain 2D slices of the original data, indexed in all three dimensions, and are available at different levels-of-detail to connected clients. For demonstration purposes, five different Earth system data cubes with over 200 parameters in total from various domains and data sources are accessible. This includes spectral indices from Sentinel-2 as well as data products from the ESA Climate Change Initiative (ESA CCI), the ERA5 and CAMS EAC4 reanalyses by the European Centre for Medium-Range Weather Forecasts (ECMWF) , the Global Land Evaporation Amsterdam Model (GLEAM)  and FLUXCOM  – partially aggregated via the Earth System Data Cube  dataset. In practice, any gridded ARDC stored in a cloud-native data format such as Zarr (https://zarr.dev, last access: 17 March 2026) can be ingested into the system, as long as there is enough disk space in the magnitude of the original dataset size for the saved tiles. See the prior work on https://www.lexcube.org on more details on tiling and the pre-processing algorithm . For all applicable parameters that have multiple years of data, anomalies, i.e., deviations from the mean seasonal cycle, can also be visualized instead of absolute values.

We exhibited the interactive data cube at the Living Planet Symposium 2025 conference held by the European Space Agency (ESA) in Vienna, Austria, during 23–27 June 2025 (see Fig. ). The cube was accessible in a central location in the exhibition space during the whole conference. On the cube, the data sets demonstrated in Table 1 in  were available to explore, amongst others. While we personally interacted with approximately 150 attendees at the cube over the week, many more used it on their own without further guidance. Overall, the feedback was very positive and participants noted that the physical data cube makes the concept of a spatiotemporal data cube immediately clear. With accessible examples such as precipitation and air temperature, the side faces of the cube – effectively interactive Hovmöller diagrams – were generally easily understood, allowing users to then explore typically less widely known Earth surface dynamics such as land-surface energy fluxes and soil moisture dynamics. Next to the increased understanding, users also remarked that the interaction with the cube and the data presented to be very intuitive. This is most likely because the three-dimensional data is presented as such in the real world instead of being projected onto a two-dimensional computer screen. Users particularly noted the fast interaction and responsiveness of the system. Built using the Lexcube software, the interaction speed is effectively in the same order of magnitude (tens to hundreds of milliseconds) as the https://www.lexcube.org benchmarks , and in practice even faster since no network transmission is involved, as all data is hosted locally. Notably, one user discovered a new spatiotemporal pattern in ECMWF aerosol data they had already worked with for years. Furthermore, users enjoyed the playful, social experience of having various large Earth system datasets accessible and explorable in physical space while discussing visible patterns with colleagues.

We conclude that the interactive physical data cube is an effective tool for exploring and visualizing Earth observation data and model outputs for scientific analysis and scientific communication. The data cube intuitively conveys to scientists, students and the broader public how spatiotemporal Earth system data in particular appears and behaves. We see great potential for using the interactive data cube in education, as interactive visualizations have been shown to increase engagement . While we have not yet systematically evaluated the physical data cube for teaching, we anticipate high value particularly in secondary education, where geography instruction still often emphasizes static, atlas-style content and neglects temporal dynamics. Yet, understanding seasonality, trends, and variability is essential for grasping the dynamic nature of the Earth system. Having discussed phenomenons visible and explorable in a physical exhibit allows for deeper engagement and a playful learning environment. Furthermore, the data cube can be exhibited at outreach events or in natural science museums, offering an accessible and playful way of exploring and interacting with Earth system data without any technical knowledge required. For scientists, the physical cube allows to inspect spatiotemporal patterns even more intuitively than previously possible with digital-only tools, enabling a deeper understanding of existing data sets they might already know and use in their research. Extending the current feature set of the physical data cube with analytical tools beyond pure visualization and integrating it deeper into scientific workflows, e.g., to easily visualize in-progress data sets, possibly from running computations, execute analyses, visualize single time series and compute statistical measures, would further strengthen the scientific use case.

A limitation of the proposed interface is that, while many Earth system datasets are gridded arrays and are therefore correctly displayed in the shape of a cube, the Earth itself is not a cube. Therefore, for visualizing global data, displaying them on the planar surfaces of a cube will necessarily introduce distortions. In practice, this means the poles are visually exaggerated, while the area around the equator is under-exaggerated. Scaling the visualized data along latitude to correspond to an equal-area projection could mitigate this issue partially. As the tool focuses on raster array data, non-gridded, sparse or vector data cannot be sensibly visualized. However, a simple vector data overlay showing region boundaries is implemented. Furthermore, the Hovmöller diagrams on the side faces, being orthogonal to the map view of the front and back faces, capture a lot of spatiotemporal dynamics, but may miss phenomena along curved paths, e.g., along coastlines. For this, a voxel visualization allowing arbitrary shapes rather than having a fixed cube shape with orthogonal boundaries would be highly beneficial.

While digital data exploration tools such as https://www.lexcube.org are inherently more accessible than physical cubes requiring construction resources or physical access, physicalizing Earth observation data and model outputs offers significant potential for scientific communication and education. We estimate that physicalizations and tools for intuitively exploring such data sets without technical knowledge will be highly relevant in the future as the amount of available data grows exponentially, e.g., with the Digital Twin initiatives. As the societal importance of Earth system datasets increases and informs policy decisions, e.g., in the biodiversity context via GEO BON , such data visualization tools can transparently and effectively communicate the underlying observed or modeled Earth system data, e.g., global climate projections or spatiotemporal development of essential biodiversity variables, to the general public. We believe that making petabyte-scale data easily available through physical visualizations will be crucial in scientific knowledge gain and science communication. We see further applications across disciplines in Earth sciences dealing with high-resolution, three-dimensional, simulated or recorded data, with possibilities of extending the physicalization to more dimensions beyond spatiotemporal data such as pressure, frequency or height/depth.