[Paper Review] Atlasmaker: A Grid-based Implementation of the Hyperatlas

R. D. Williams, S. G. Djorgovski+2 more|arXiv (Cornell University)|Dec 8, 2003

Astronomical Observations and Instrumentation1 references3 citations

TL;DR

Atlasmaker is a grid-based software system that creates scientifically trusted, multi-wavelength, multi-resolution sky atlases by federating astronomical image surveys through standardized resampling using the Hyperatlas framework. It enables on-demand or batch computation of co-added mosaics via trusted resampling tools (Montage and Swarp), storing results in a distributed virtual file system (SRB) for reuse, thereby enabling new data-centric discovery in astronomy and educational applications.

ABSTRACT

The Atlasmaker project is using Grid technology, in combination with NVO interoperability, to create new knowledge resources in astronomy. The product is a multi-faceted, multi-dimensional, scientifically trusted image atlas of the sky, made by federating many different surveys at different wavelengths, times, resolutions, polarizations, etc. The Atlasmaker software does resampling and mosaicking of image collections, and is well-suited to operate with the Hyperatlas standard. Requests can be satisfied via on-demand computations or by accessing a data cache. Computed data is stored in a distributed virtual file system, such as the Storage Resource Broker (SRB). We expect these atlases to be a new and powerful paradigm for knowledge extraction in astronomy, as well as a magnificent way to build educational resources. The system is being incorporated into the data analysis pipeline of the Palomar-Quest synoptic survey, and is being used to generate all-sky atlases from the 2MASS, SDSS, and DPOSS surveys for joint object detection.

Motivation & Objective

• To develop a scalable, interoperable system for federating multi-wavelength astronomical images into a unified, scientifically trusted atlas.
• To address the challenge of combining images from disparate surveys with different resolutions, wavelengths, and projections into a co-registered, mosaic format.
• To enable both on-demand and scavenged grid computing for large-scale atlas generation, minimizing redundant computation.
• To support provenance tracking and dynamic updates of derived data products through virtual data modeling.
• To establish a new paradigm for data-centric knowledge extraction in astronomy by enabling image federation beyond catalog fusion.

Proposed method

• Utilizes the Hyperatlas standard to define a discrete, distortion-free projection of the celestial sphere onto a plane, with scale specified in powers of two arcseconds per pixel.
• Employs the IVOA Simple Image Access Protocol (SIAP) to discover and retrieve NVO-compliant image data from distributed sources.
• Leverages Grid computing infrastructure (e.g., Teragrid, Condor, PBS) to distribute and parallelize resampling and mosaicking tasks across geographically dispersed resources.
• Integrates two trusted resampling kernels—Montage (NASA/IPAC/JPL) and Swarp (Terapix)—to perform image re-projection, background subtraction, and co-addition with controlled degradation.
• Stores computed mosaics in a distributed virtual file system using the Storage Resource Broker (SRB), enabling persistent caching and efficient retrieval.
• Applies a virtual data model that treats data products as executable recipes, allowing dynamic regeneration when underlying data or calibration changes.

Experimental results

Research questions

• RQ1 How can multi-wavelength and multi-resolution astronomical images be systematically combined into a co-registered, scientifically valid atlas?
• RQ2 What computational architecture enables efficient, scalable, and fault-tolerant generation of terabyte-scale sky atlases using grid resources?
• RQ3 To what extent can image federation reveal new astrophysical insights not apparent in individual surveys or their cataloged data?
• RQ4 How can provenance, reproducibility, and dynamic updates be preserved in large-scale, distributed image mosaicking workflows?
• RQ5 Can on-demand and scavenged grid execution modes coexist effectively in a system designed for both interactive and batch atlas generation?

Key findings

• Atlasmaker successfully generates multi-wavelength, co-registered sky atlases from surveys such as 2MASS, SDSS, and DPOSS using the Hyperatlas standard.
• The system supports both on-demand and scavenged grid execution, enabling efficient use of heterogeneous computational resources for large-scale atlas creation.
• By using trusted resampling tools (Montage and Swarp), Atlasmaker maintains acceptable image quality with controlled degradation in PSF, astrometry, and flux during resampling.
• Computed mosaics are stored in a distributed SRB-based virtual file system, enabling persistent caching and reuse, reducing redundant computation.
• The virtual data model allows downstream data products to be dynamically regenerated when underlying data or calibration changes, improving reproducibility.
• Atlasmaker is being integrated into the Palomar-Quest synoptic survey pipeline, demonstrating real-world applicability for joint object detection and data mining.