Mosaic Dataset or 1 large raster dataset for U.S coverage of USGS DEM?

[MarkZweifler]

I am in an organization that wants to perform analysis using USGS 10m. DEMs for the entire United States( approximately 440Gb of source data stored as approximately 3,000 Arc/INFO GRIDs). We have a relatively small group of users and our primary use of this data would be to combine it withother rasters as an input to multi-criteria models.

After careful reading of the Help documentation my assessment is that we are best served by creating 1 large SDE raster dataset

My preference for the 1 big SDE raster dataset approach is based on the assumptions that:
�?� SDE management of 1 large raster would be more efficient than a mosaic that references source pixels stored in GRIDs on disk.
�?� SDE's raster storage model of spatial indexes negate the probem of dealing with very large rasters because it only fetches the data needed the display extent or analysis function - so if the db has adeqiate space no value in chuncking a national dataset into pieces
�?� All the DEM data is going to have identical properties and there is no value to storing the standard overlap included in each USGS DEM  'tile'

However, some documentation seems to suggest that leaving data in the its source format and using a Mosaic dataset for visulaization and access is the preferred method.

Managing elevation data: Part 2: Design and data management plan
http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#/Part_2_Design_and_data_management_plan...

My sense is that the advantages of a mosaic dataset do not seem to apply to a consistent data product that has been tiled into manageable pieces ( 1/9 arc second quads) by USGS  for public distribution. and if you have the database storage capacity why not load into SDE as a single raster.


In this scenario, does anyone have advice as to what is the best data storage solution and why

Thanks for any help

[MelanieHarlow]

Ultimately, it's your decision. But we recommend the mosaic dataset over SDE Raster for a few reasons:
- Creating a mosaic dataset will be faster than creating the SDE Raster.
- Database overhead/costs.

Regarding your assumptions:
â?¢ SDE management of 1 large raster would be more efficient than a mosaic that references source pixels stored in GRIDs on disk.
- not necessarily true. But there would be only 1 file as opposed to the multiple files used by the mosaic dataset.

â?¢ SDE's raster storage model of spatial indexes negate the probem of dealing with very large rasters because it only fetches the data needed the display extent or analysis function - so if the db has adequate space no value in chunking a national dataset into pieces
- The mosaic dataset only reads the files and pushes the pixels it needs to generate the mosaicked image/DEM
- Except for building overviews (which are really like raster pyramids which you will need with the SDE Raster) there's no chunking into pieces.

â?¢ All the DEM data is going to have identical properties and there is no value to storing the standard overlap included in each USGS DEM 'tile'
- No value ?? Well some day there could be value. And if there were individual properties you wanted to maintain you could (such as accuracies, processing info and dates). Also as individual files they can be querried and accessed individually if users ever wanted to do this.

The SDE Raster and mosaic dataset have similar performance. Internal numbers sometimes show the mosaic dataset as faster depending on the source image formats (e.g. tiff). SDE Raster is still a valuable data source and simple to create.

We highly recommend using ArcSDE when data requires security that can't be achieved outside a database.

If you want to see an example of a mosaic dataset created with the USGS NED data (you're using) and other dataset, join the World Elevation services beta program.
http://resources.arcgis.com/content/imagery/10.0/world_elevation

[MarkZweifler]

Thanks Melanie for the quick reply- valuable advice and correcting my flawed assumptions.

This is an even stronger endorsement of mosaics than I was expecting.

You seem to suggest that there really is no performance or analytical advantages to raster geoprocessing with an SDE raster versus a mosaic dataset that references files on disk (particularly when data is stored as tiff ).

THat the only advantage to the SDE raster dataset relates to the  additional layer of security  afforded by an RDBMS.

Thanks for the info about the World Elevation services beta program. I guess we will just need to comparison test to see if the network requirements of using this image service provides comprable performance as to accessing a Mosaic dataset on a local server.