JDJean-Marc Delouis

Jean-Marc Delouis

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Jean-Marc Delouis has a total of 104 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Grid4Earth: An Open-Source Python Ecosystem for Geospatial Data Integration Using an Ellipsoidal HEALPix DGGS

    by Tina Odaka, Jean-Marc Delouis, Justus Magin, Camille Gueguen, Etienne Cap, Anne Fouilloux, Benoit Bovy, Pablo Richard, Wai Tik Chan, Kajetan Marcin Chrapkiewicz, et al.

    The rapid growth in Earth observation (EO) and Digital Twin Earth data volumes creates a need for global, reproducible, and cloudnative spatial representations. Conventional latitude–longitude grids and projected tiling systems remain useful, but they introduce projection boundaries, non-uniform cell areas, and repeated reprojection costs when combining multi-source products. Grid4Earth addresses this problem through an open-source Python ecosystem built around an ellipsoidal HEALPix Discrete Global Grid System (DGGS) representation and Zarr-based data handling. The ecosystem consists of four composable packages: healpix-geo for WGS84-aware indexing and coverage queries, healpix-resample for CPU/GPU-capable regridding, healpix-plot for visualisation, and healpix-analyse for diagnostics and analysis. We situate Grid4Earth in relation to previous DGGS comparisons, XDGGS, and OGC API – DGGS work, focusing on the implementation layer required for EO workflows. When WGS84 geodetic latitude is passed directly to spherical HEALPix, local cell areas vary by up to approximately 0.9% because of Earth’s non-spherical shape. Grid4Earth preserves the HEALPix equal-area property on WGS84 through an authalic-latitude mapping. The ecosystem has been exercised in HEALPix/Zarr workflows for Sentinel-2, Sentinel-3, ERA5, CAMS, and DestinE Climate Digital Twin outputs. This paper describes the ellipsoidal geometry, the CPU/GPU-capable resampling architecture, and the implementation of metadata fields defined by CF 1.13 and version 1 of the Pilot Zarr DGGS convention, establishing ellipsoidal HEALPix/Zarr as a common representation that connects climate and Digital Twin Earth datasets with ellipsoidal EO datasets.

  • Open Access

    Untangling dust emission and cosmic infrared background anisotropies with the scattering transform statistics

    by Srijita Sinha, Tuhin Ghosh, Erwan Allys, François Boulanger, Jean-Marc Delouis

    A template-fit approach is often used to separate the Galactic dust emission and the cosmic infrared background (CIB) anisotropies in low column density regions using the observational fact that the 21,̊m cm H I H I line emission from neutral atomic hydrogen and dust are tightly correlated. However, in some regions with molecular hydrogen, diffuse ionised gas, and dark gas, the same approach fails to trace the excess Galactic dust emission. We developed and tested a statistical component-separation method to extract the dust signal from the contaminated 353, Planck GHz observations using the scattering covariance (SC) statistics, which is a subclass of scattering transform statistics. We first obtained a set CIB maps over 25 square patches, each with a sky area of 222, column density map valid in low column density regions using the template-fit approach. We then constructed from these 25 maps a generative model of CIB using SC statistics. We finally relied on this generative model to perform a component-separation of dust and CIB in the data for different sky regions. These separations were achieved by sampling an ensemble of dust maps through pixel-based optimisation, which, when added to the CIB contamination model, verified all the statistics and cross statistics constraints that were estimated directly from the data. using the linear correlation of dust and Galactic H I H I Planck We validated our algorithm and separated the dust emission from the contamination in the 353, ̊m GHz observations. We show the results of the recovered dust map for a test sky region where there is a significant difference between the Planck Planck dust map and CSFD map. We found that the dust map has more structure than the CSFD map. We compared the power spectrum of the recovered dust map and map and found a difference in slope (Δ α ≈ 0.4) by fitting a power-law model to the two spectra. To explain Δ α, we decomposed the recovered dust map into two gas phases: dust associated with neutral atomic hydrogen, and dust associated with molecular hydrogen. We provide a clear pathway to mapping the Galactic interstellar reddening over intermediate and high Galactic latitudes. Planck H I

  • Open Access

    pangeo-fish: A Python package for studying fish movement using biologging and earth science data

    by Justus Magin, Quentin Mazouni, Etienne Cap, Marine Gonse, Mathieu Woillez, Anne Fouilloux, Jean-Marc Delouis, Tina Odaka

  • Open Access

    Towards universal building blocks for cloud-native digital-twins

    by Alexander Kmoch, Wai Tik Chan, Guillaume Ameline, Justus Magin, Jean-Marc Delouis, Tina Odaka, Benoit Bovy, Anne Anne, Evelyn Uuemaa

    The exponential growth of Earth Observation (EO) data presents significant challenges for efficient data access, processing, and analysis. Current approaches often involve disparate data formats, coordinate systems, and access patterns, limiting interoperability and scalability. Firstly, the Zarr and Parquet data storage formats have seen wide adoption as a unifying cloud-native foundation for various domains in recent years, including climate, EO, bio-imaging, and genomics. Secondly, Discrete Global Grid Systems (DGGS) such as HEALPix, or ISEA-based hexagonal DGGS are being increasingly used to enable spatial data indexing beyond traditional grids, by using equal-area pixels and location- and refinement level encoding indices. Lastly, the recently published OGC API DGGS standard specifies a lightweight web service API for clients accessing data organised according to Discrete Global Grid Reference Systems (DGGRS). We describe a scalable, interoperable, and extensible FOSS architecture for a modern geospatial data ecosystem, based on DGGS. As an example, we introduce pydggsapi, a Python server implementing this new OGC standard to serve large geospatial datasets from cloud-native Zarr and Parquet data stores indexed by a DGGS. This novel architecture combines the DGGS data cube paradigm, standardised OGC API DGGS web service access, and cloud-optimised data formats such as Zarr and Parquet as universal building blocks for geospatial data management, enabling seamless transitions between high-performance computing environments and lightweight client applications.

  • Open Access

    JWST imaging of the Pleiades: Anisotropy of turbulence in the cold neutral medium

    by G. Vigoureux, N. Flagey, F. Boulanger, A. Noriega-Crespo, V. Guillet, A. J. Alvarez-Castro, N. de Jesus-Rivera, E. Allys, J. M. Delouis, E. Falgarone, et al.

    Interstellar medium studies rely on magnetohydrodynamic turbulence as a framework for interpretation. In this context, the statistical characterization of interstellar observations is of prime importance. We present a new perspective on diffuse interstellar matter by analyzing James Webb Space Telescope (JWST) observations of the Pleiades nebula with NIRCam. These observations are remarkable in that they provide a “microscope” view of the cold neutral medium (CNM) with a spatial resolution of 0.2 mpc (40 au). A 2D Fourier analysis was used to characterize the structure of polycyclic aromatic hydrocarbon (PAH) emission in regions near and far from the Pleiades star Merope. To produce maps of the interstellar emission, stars and galaxies were filtered out. The final step in the data cleaning involved subtracting a component, in Fourier space, which we inferred to be a residual of the near-infrared cosmic background. The PAH emission power spectra are highly anisotropic. This power anisotropy is constant across scales. The magnetic field orientation, as derived from the Planck dust polarization data, aligns with the PAH anisotropy. The spectra are well fitted with a break-free power law, suggesting that we do not observe a specific scale for energy dissipation. Power-law indices are −3.5 near Merope and −3 in the more distant field. These findings are discussed in relation to interstellar turbulence that may be driven by the Pleiades stars. The JWST observations of the Pleiades offer a new viewpoint for comparing observations and theoretical models, as they examine physical scales at which turbulence in the CNM is subsonic and decoupled from the thermal instability. The observations may indicate that the turbulent energy cascade in the CNM is anisotropic.


  • Open Access

    Modeling orthorectification and PSF distortions on a HEALPix grid

    by Pablo Richard, Jean-Marc Delouis, Justus Magin, Tina Odaka

    The quality of analysis-ready Earth Observation (EO) products strongly depends on the ability of the processing chain to accurately model the mapping from the Earth’s surface to the detector geometry. This mapping involves a convolution with the instrumental Point Spread Function (PSF) and an orthorectification step that corrects for terrain-induced geometric distortions using a Digital Elevation Model (DEM).These distortions are challenging, as they can lead to spatially varying PSF deformations. Even worse, some areas with strong topographic gradients lead to an effective PSF with multiple modes and may cause the failure of standard operational orthorectification algorithms.To anticipate these failures, we introduce critical incidence maps. For a given point on a DEM, such a map provides the maximum sensor incidence angle at which this point remains visible, i.e. not occluded by surrounding terrain in any azimuthal direction. We show that, for moderate incidence angles (typically below ~20°), rugged areas responsible for orthorectification failures cover only a very small fraction of Earth’s surface, leaving therefore ample room for more robust algorithms to tackle these thorny points.For smooth regions, we introduce and compare various semi-analytical orthorectification schemes that achieve appealing trade-offs between computational cost and geometric precision. We then combine two distinct orthorectification strategies, tailored respectively to smooth and rugged terrain, and express the overall ground-to-detector mapping as a sparse linear operator.This formulation yields an efficient forward model that accurately captures terrain-induced PSF distortions, including multimodality. Finally, we apply this model to invert the data projection, from the detector to the Earth’s surface, in the context of a HEALPix discrete grid.

  • Open Access

    Gauge-Equivariant Spherical U-Nets on HEALPix for Global SST Forecasting: Encoding local orientation on the sphere

    by Jean-Marc Delouis, Tina Odaka, Sébastien Tétaud

    Many climate variables are naturally defined on the sphere and exhibit strong anisotropy and directionality (e.g., fronts, jets, boundary currents). Yet most deep-learning forecasting models still rely on planar projections and Euclidean convolutions, which introduce geometric distortions and artificial discontinuities. Graph-based spherical models alleviate some of these issues, but typically remain isotropic and do not explicitly represent local orientation, a key ingredient to model directional transport-like patterns.Here we introduce and evaluate a gauge-equivariant spherical U-Net implemented directly on the HEALPix grid, designed to encode local orientation consistently across the sphere. Our approach leverages gauge-equivariant convolutions that transform predictably under changes of local reference frame, allowing the network to learn directional filters while preserving spherical geometry. This provides a principled alternative to both planar U-Nets (with longitude-periodic padding) and graph U-Nets, and addresses a core limitation of most spherical models: the lack of explicit orientation handling. This work benchmarks this model against two strong baselines: a planar U-Net with longitude-periodic padding and a spherical graph U-Net defined on the same HEALPix discretization.We apply this architecture to multi-horizon forecasting of global sea-surface temperature (SST) anomalies at NSIDE=32, using a controlled experimental design with matched training protocols and comparable parameter budgets, with emphasis on low-capacity regimes relevant to data-limited climate settings (≈30–40 years of monthly observations). We report quantitative metrics across horizons and analyze qualitative error modes, showing how gauge-equivariant spherical convolutions mitigate projection artefacts while enabling orientation-aware feature extraction on the sphere. Our results highlight when and why encoding orientation through gauge equivariance provides added value beyond “spherical-but-isotropic” baselines, and offer practical guidance for deploying spherical equivariant models in climate forecasting pipelines.

  • Open Access

    BigEarthNet-HEALPix: Spherical CNNs for Land Cover Classificatiom

    by Sébastien Tétaud, Jean Marc Delouis

    Remote sensing datasets for land cover classification are mostly distributed in UTM projection which introduce significant geometric distortions—particularly at high latitudes—and fail to respect the spherical geometry of Earth. These distortions propagate into deep learning models trained on such data, leading to latitude-dependent biases, edge artifacts in tile-based processing, and poor generalization across geographic boundaries. While convolutional neural networks (CNNs) have achieved state-of-the-art performance on benchmark datasets like BigEarthNet, they operate on Euclidean grids and cannot naturally handle the structure of a sphere.Here we introduce a comprehensive pipeline for transforming the BigEarthNet dataset—comprising 549,488 multispectral image patches from its original UTM projection into the HEALPix (Hierarchical Equal Area isoLatitude Pixelization) representation. HEALPix, originally developed for cosmic microwave background analysis, offers equal-area partitioning of the sphere, ensuring uniform statistical treatment of pixels regardless of latitude, and provides a natural hierarchical structure for multi-resolution analysis.We implement and evaluate spherical CNNs architectures designed for data on spherical manifolds—against traditional planar CNN baselines (Unet/Resnet50) trained on the HEALPix-transformed data, benchmarking classification performance for multi-label land cover prediction using the 19-class BigEarthNet nomenclature with metrics suited to imbalanced settings (F1-macro/micro, precision, recall, average precision).This work represents the first large-scale application of HEALPix projection to Remote Sensing classification and validates the effectiveness of spherical deep learning for real-world remote sensing beyond traditional climate science domains. Our experimental design employs matched training protocols and comparable model capacities, demonstrating that spherical representations eliminate projection-induced artifacts, enable seamless cross-boundary analysis, and provide rotation equivariance that reduces the need for extensive spatial data augmentation—key advantages for global-scale Earth observation applications.

  • Open Access

    Pangeo-Fish and the Global Fish Tracking System: Scaling Biologging Analytics with Earth System Digital Twins for evidence-based policy support

    by Tina Erica Odaka, Etienne Cap, Quentin Mazouni, Corentin Hue, Jean-Marc Delouis, Mathieu Woillez, Anne Fouilloux, Benjamin Ragan-Kelley, Daniel Wiesmann

    The Global Fish Tracking System (GFTS) and Pangeo-Fish integrate biologging data with high-resolution environmental data in a digital-twin framework to address key challenges in marine conservation and fisheries management. Linking fish movement models with climate projections from Europe’s Destination Earth (DestinE) Climate Change Adaptation digital twin yields an evidence-based tool for decision support in habitat conservation and fisheries management under climate change. The implementation is built on the open-source Pangeo ecosystem and deployed on the DestinE platform.Pangeo-Fish is an open-source package that ingests multiple biologging data types—including archival tags, pop-up satellite archival tags (PSATs), and acoustic telemetry detections. It processes time series observed by fish (e.g., depth, temperature, and light) together with geolocation constraints derived from external sources such as acoustic receiver networks and tag-based positioning. These heterogeneous observations are harmonised in a cloud-native workflow to support scalable track reconstruction and downstream habitat-relevant products.Tracks are reconstructed using a Hidden Markov Model (HMM) geolocation approach that combines tag-recorded time series (e.g., depth, temperature, and light) with external geolocation constraints (e.g., acoustic detections), together with priors such as bathymetry and release/recapture information. Processing leverages cloud-native tools (Jupyter, Dask, Xarray) and chunked cloud-optimised storage (Zarr) for scalable analysis. A key design choice is the use of HEALPix as the base spatial grid and indexing scheme from ingestion to visualisation, enabling efficient path-likelihood evaluation on an equal-area, iso-latitude grid while avoiding distortive resampling. Environmental reference fields are primarily sourced from the Copernicus Marine Service, but the workflow can also ingest user-defined datasets. In addition, in situ observations (e.g., Argo float temperature) can be incorporated to represent uncertainty in the ocean physics fields used by the geolocation model and better account for model–data discrepancies.The Pangeo-Fish workflow yields most-probable tracks and daily presence-probability maps, while GFTS supports aggregation of distributions for management-relevant analyses. GFTS intersects these outputs with climate projections from the DestinE Climate Change Adaptation Digital Twin to assess potential habitat exposure under climate change. GFTS has so far been demonstrated for Atlantic applications, while Pangeo-Fish has been extended to the Pacific Ocean, enabled by portable cloud-native processing and the availability of cloud-accessible datasets. As an early DestinE platform use case, this work illustrates how Earth system digital twins can be operationalised for reproducible, scalable biologging analytics to inform marine conservation and sustainable fisheries management.

  • Open Access

    Ellipsoidal HEALPix for the Earth sciences: healpix-geo and xdggs integration into the Pangeo ecosystem

    by Justus Magin, Benoît Bovy, Pablo Richard, Jean-Marc Delouis, Tina Odaka

    Discrete Global Grid Systems (DGGS) and in particular HEALPix have become increasingly popular in the Earth sciences over the past few years, mainly due to its equal-area nature and readily available python libraries. However, this adoption and the ever increasing amounts of data come with its own set of challenges. In particular, the existing python libraries are written for use in astronomy and thus only work on a sphere, resulting in small but often non-negligible variations in the cell areas when applied to the surface of the earth.Additionally, a large spatial coverage at high resolutions require a large amount of memory just to represent the spatial information in memory (e.g. roughly 100GB for full Earth coverage at 100 m resolution).Finally, the storage format of HEALPix was not standardized until very recently with the release of the CF conventions version 1.13 and the upcoming zarr DGGS convention.We present healpix-geo, a HEALPix implementation library for python built on top of the cdshealpix, moc, and geodesy rust crates with minimal python dependencies (numpy and optionally shapely). It supports the most common HEALPix indexing schemes (nested, ring, zuniq), allows the conversion of cell indices to and from ellipsoidal coordinates, and contains a range-based data structure suitable to index a large amount of cells with a small memory footprint.We further show how healpix-geo integrates with xdggs, an xarray extension that enables high-level interaction with DGGS datasets, including efficient subsetting, analysis-ready representations, and visualization within Pangeo workflows. xdggs also provides an extensible mechanism to easily import/export DGGS data from/to a variety of models or conventions, with built-in support of the CF HEALPix conventions and the zarr DGGS conventions. Together, healpix-geo and xdggs provide an end-to-end, standards-aligned pathway for scalable HEALPix-based geospatial analysis on the ellipsoid.