This poster (or slides) describes the main concepts which relate data format (HDF5), metadata, and data organization, as they are being applied to NPOESS. It includes the motivation for selecting HDF5; the motivation and general implementation of FGDC base metadata and remote sensing extensions; and the current attempt to use best experiences with HDF-EOS swath and netCDF Climate & Forecast conventions to guide NPOESS product definition.
2. Motivation
NPOESS will collect an unprecedented quantity and variety of satellite environmental data from a
constellation of satellites carrying multiple remote sensing and in situ sensors.
NPOESS realized the need for standardization and good metadata in connection with
environmental data sets. Global data sets, such as those produced by satellites or used in
modeling and simulation, are at the leading edge of information stewardship challenges.
Data sets will be delivered to operational users by pre-subscription through Interface Data
Processors (IDPS) at the four US processing Centrals, and through direct readout Field
Terminals (FTS) worldwide. Data will be retrieved by near- and long-term researchers
through NOAA’s data archives.
NPOESS committed early to using existing standards in a rational manner to support our mission
goals.
– Standards-based data products ease product usability.
– Comprehensive metadata contributes to maximizing the continuing value of large,
complex data sets. Standards-based metadata ease product understanding.
– Working from a clean sheet of paper, NPOESS can serve as a “workshop” for best
practices in data stewardship.
This paper presents the results of planning for the NPOESS data product delivery system. It
presents a view of the optimal data product design, much of which is being implemented in
NPOESS.
4. Architectural Considerations
Sizing
•
Overall data rates have increased by two orders of magnitude
•
more products
•
impact the on existing user systems and applications
•
decision to “loosely” couple
•
simple data interface with flexibility on both sides
•
decided to deliver relatively short-duration data granules, containing typically 30 seconds of data
Users and use patterns: designed to serve operational users, current science users, and future archival
researchers
•
Operational users need effectively all of the data as soon as possible
•
Research users need current environmental information, but usually have time and resources to improve
product quality with post-processing
•
Archival researchers look for highly selective data sets
•
Users via Centrals or via field terminals
Sensor complexity
•
multiple versions of the data
•
the raw bitstreams originating from the sensors (RDRs)
•
calibrated fluxes measured by the sensor (SDRs)
•
environmental variables estimated at the source (EDRs)
•
sensors themselves produce a wide variety of data types
•
various techniques to maximize performance within bandwidth
Anticipating change
•
detailed formats, contents, product lists, and interfaces that must be accommodated by NPOESS data
product framework format through the mission lifetime
5. SENSORS
CCSDS (mux, code, frame) &Encrypt
Delivered Raw
Packetization
Compression
Aux.
Sensor
Data
Cal.
Source
ENVIRONMENTAL
SOURCE
COMPONENTS
Filtration
A/D Conversion
Detection
Flux
Manipulation
Comm
Processing
C3S
Comm
Receiver
RDR
Production
RDR Level
TDR Level
SDR
Production
SDR Level
EDR Level
EDR
Production
IDPS produces mission data sets which recreate or estimate signals at 4 points in the sensing
chain. Ancillary data, brought from other systems and used in EDR processing, is captured.
Auxiliary data, produced within the NPOESS system to support processing, is kept with the
mission data or incorporated in documentation. Metadatata provided for all.
IDPS
Comm
Xmitter
Data
Store
OTHER
SUBSYSTEMS
SPACE SEGMENT
Data Products Are More Than Scenes
6. Typical Data Organization
RDRs
SDRs
EDRs
Ancillary
Time-Series of
Packet Types
Binary
Headers
u
v
w
Multispectral
Imagery
Vector
Flux
Data
Volume
Geolocation
Calibration
Table
Slit
Spectra
Quality
Column Data
Thematic Layers
Geolocation
?
Abstract
External Data
Sounding
FT
Spectra
Imagery
7. First Decision – File Format
The NPOESS program identified several key characteristics for the file format within which data products
would be delivered. Hierarchical Data Format ver.5 (HDF5) was found to be the best solution within
technical and programmatic constraints:
•
A single format with proven ability to handle environmental data products (EDRs); more abstract data
structures in RDRs, TDRs & SDRs; and other products delivered to users and archives
•
Capability to incorporate full metadata
•
Supported by the user community and other institutional support, with an adequate practical lifetime;
interoperable with DoD standards
•
Ability to handle large data sets (such as full orbits) and small data sets (individual granules) with
acceptable efficiency
•
Ability to handle multiple arrays and array types within the same granule, such as observational data arrays
and geolocation arrays
•
High efficiency for reading and writing; built-in compression function; capability to “chunk” large data
arrays to access prestructured subsets
•
Sufficiently self-documenting to permit variable formats
•
Available with development tools to expedite file definition and applications
•
Acceptable licensing terms
•
Supported on all likely user platforms & operating systems
•
Support for all likely atomic data types
•
Simple data objects and groups which permit application-specific structures to be created
8. File Structure Implemented in HDF5
B
Dataset
A
File
file attributes
Header
Granule
Granule
Data
Dataspace
int16
current = 12e-9
Datatype
2
Rank
Dim_3=2
Dim_2=4
Dim_1=5
Dimensions
Dataset
Dataset
temp = 56
time = 32.4
Dataset
Dataset
Attributes
chunked; compressed
Storage layout
granule attributes
granule attributes
HDF5 provides a simple, logical file structure based on a Dataset, comprising a data array and a header
which describes the dataspace. Datasets can be structured hierarchically. NPOESS datasets and
additional attributes – incorporating the metadata – combine into granules, and granules combine into
product files. Granules are concatenated in a file in such a way that they can be addressed either
collectively or individually. Individual datasets will be created for elements such as mission data, quality
assessment, geospatial location, time, illumination, and viewing geometry. Users may access subsets of
the full data using HDF5 utilities.
9. Second Decision - Metadata
Based on extensive prior experiences in the earth sciences and other sciences, NPOESS highest
level operational requirements specified that comprehensive metadata would be delivered
with the data. In our context, metadata incorporates the following “data about data”:
• Identification
• Content summary
• Content meaning
• Content structure and format
• Acquisition and processing history – provenance
• Distribution and availability
The National Spatial Data Infrastructure (NSDI) provides basic guidance for metadata. The
Federal Geospatial Data Committee (FGDC) sets content standards. The program establishes
compatible extensions to the standard. The program also defines the representation. Over one
hundred metadata items have been defined for NPOESS.
Basic identification metadata is duplicated in a User Block at the front of the file, where it can be
read without HDF software.
10. FGDC Metadata Base & RS Standards
Metadata
1
2
Identification
3
Spatial
Data
Organization
Data
Quality
6
Distribution
Information
Base Standard
RSE Standard
4
7
Metadata
Reference
Information
5
Spatial
Reference
8
Platform &
Mission
Information
Entity &
Attribute
Information
9
Instrument
Information
11. Quasistatic
NSDI
metadata
Complete
granule
metadata
Dynamic
NSDI
metadata
Dynamic
detail
metadata
eHandbook
external reference
Identification (UB)
Aggregate metadata
Common metadata
file
Unique metadata
granule
Concept
Analysis
Repositories
NPOESS will create all applicable metadata identified in the FGDC Base Standard and Remote
Sensing Extensions, plus mission-unique. Comprehensive metadata at the granule level will be
captured in different ways, consistent with usage patterns. Metadata which changes infrequently will
be collected in an external online reference. Metadata which changes with each granule is saved with
it. Metadata which describes a file or all the granules in the file is extracted and stored with the file.
12. Third Decision – File Organization
Primary characteristics of a design
•Systematic data organization
•Simplifies data maintenance, enhancement, documentation, retrieval, visualization, and exploitation
•NPOESS evaluation looked at lessons learned from EOS, and best practices elsewhere
•Must work with abstract, generalized scientific, and specifically geospatial data sets
Result: Specific derived requirements closely match the “Climate & Forecast Conventions”, primarily
developed for use with netCDF. It provides guidelines for consistent, complete, and clear data entity and
attribute definition. NPOESS is attempting to implement this data design.
Need clear identification of truly independent and dependent variable contents:
•For most unresampled data, the natural independent variables are the index attributes defining the discrete
points in space and time at which the data samples were collected. Index might be a time-series sample number,
detector number, energy or spectral bin.
•Usually include an associated independent variable for each index attribute, such as time, direction, position, or
energy level. These relate to indices by calibration, known to be very accurate.
•Some associated independent variables may be multidimensional. E.g., polar geolocation or solar elevation is
a deterministic function of spatial indices.
•Always a primary dependent variable: function of independent variable(s). May be an abstract binary object,
function of time or place. Usually, well defined variables which are a function of temporal, spatial, or spectral
indices.
•Often 1+ associated dependent variable arrays, such as quality estimates or telemetry values, associated by
design with the primary independant variable
•Optional supplementary arrays, such as calibrations, are functions of 1+ index attributes used in the primary
and associated data arrays.
Concatenation: One index attribute, usually time or time-like, can often be defined as the index which
establishes continuity from one granule to the next.
13. Clear Index & Array Definitions
Primary
Index
Primary Array
e.g., Flux, Brightness,
Counts, NDVI
Index
Attribute
n-Dimensional Dependant
Variable (Entity) Array
2-D Independent
Variable Arrays
e.g., lat/lon, sun
alt/az, land mask
Associated
Independent
Variable(s)
14. Putting Together the Framework
Granule Boundaries
Time
Supporting Timeline Data
Sensor Data Quality Flag Bits
Sensor Data Arrays
Geolocation Arrays
15. Putting Together the Framework
With careful design, and based on lessons learned from previous programs, a comprehensive data
product design can be achieved. The design eases development and maintenance, by
providing a common approach to data and metadata.
Granules form the basic unit of production and cataloguing. They are essentially self-contained.
Each granule contains the primary data arrays, associated data arrays, and descriptive attributes –
including metadata – needed to understand the information content.
To facilitate efficient delivery, multiple granules of the same type are combined into product files.
Common attributes of all granules in a file may be extracted to the file level as common
metadata. Summary and identification attributes are created and added to the file. Basic
identification metadata is extracted from the HDF format and saved as an ASCII ‘user block’
at the physical start of the file.
Finally, metadata which changes only rarely is maintained separately as an electronic handbook,
and is incorporated in the granules by reference.
16. eHandbook
HDF File
Block
Root
File
Attributes
User
Block
File Metadata
Common Metadata
Data
Product
Other File Attributes
Granule
Granule
Granule
Granule
Attributes
Granule
Metadata
Datasets
Index Variables
Primary Array
Associated Arrays
Other
Granule
Attributes
Granule
Attributes
Granule
Metadata
Datasets
Associated Arrays
Granule
Metadata
Other
Granule
Attributes
Index Variables
Primary Array
Granule
Attributes
Datasets
Index Variables
Primary Array
Associated Arrays
Other
Granule
Attributes
Editor's Notes
The National Polar-orbiting Operational Environmental Satellite System (NPOESS) is managed by the NOAA Integrated Program Office, on behalf of the U.S. Dept. of Commerce, Dept. of Defense, and NASA. Northrop-Grumman Space Technology is prime contractor, with Raytheon the subcontractor for ground processing & operations
IDPS produces mission data sets which recreate or estimate signals at 4 points in the sensing chain. Ancillary data, brought from other systems and used in EDR processing, is captured. Auxiliary data, produced within the NPOESS system to support processing, is kept with the mission data or incorporated in documentation. Metadatata provided for all.
HDF5 provides a simple, logical file structure based on a Dataset, comprising a data array and a header which describes the dataspace. Datasets can be structured hierarchically. NPOESS datasets and additional attributes – incorporating the metadata – combine into granules, and granules combine into product files. Granules are concatenated in a file in such a way that they can be addressed either collectively or individually. Individual datasets will be created for elements such as mission data, quality assessment, geospatial location, time, illumination, and viewing geometry. Users may access subsets of the full data using HDF5 utilities.
NPOESS will create all applicable metadata identified in the FGDC Base Standard and Remote Sensing Extensions, plus mission-unique. Comprehensive metadata at the granule level will be captured in different ways, consistent with usage patterns. Metadata which changes infrequently will be collected in an external online reference. Metadata which changes with each granule is saved with it. Metadata which describes a file or all the granules in the file is extracted and stored with the file.
Conceptual representation of the relationship among granule data components. Mission or scientific data is kept as arrays organized by the collection indices. Auxiliary data (e.g., attitude, ephemeris, and scanner angle encoder readings), much of as a time sequence, are kept in a similar manner. Derived parameters (e.g., polar and cartesian geolocation, and illumination and viewing geometry) and processing flags and status bits are added. Each granule is complete by itself. When desired, multiple granules can be concatenated along the time dimension. This structure can be retained until the observed or processed data is resampled onto another coordinate system.