The figure illustrates the AMAPS concept: a virtual workspace in which executable "operators" can be invoked on any member data set regardless of actual physical locations of data, operators and users. AMAPS uses the SciFlo grid workflow engine to call remote operators (web services), pass arguments to them, and/or move code and data over the world-wide web in whatever order or configuration is most efficient. This makes access to data easier, allows users to build on and reuse each other's operators, and increases computational power.

Service users access AMAPS capabilities through web pages that display predefined SciFlo workflows. These workflows are XML documents that chain together a sequence of analysis steps or calculations (operators), and are displayed on a web page. Users enter arguments by filling in text boxes, and invoke the process by pushing an "execute" button. Results are put in one or more files on the computer which executed the SciFlo, and a clickable link pointing to output file(s) is returned on the web page.

Power users author their own XML documents, bypassing the web interface. Power users have the flexibility to create their own executables in python (and eventually in any language or package of their choice), and chain them together with existing AMAPS operators, web services, and even other workflows. Power users execute their workflows from the command line on any computer equipped with AMAPS software. Such a computer is called an AMAPS node. Presently there are three AMAPS nodes: one at JPL, one at the University of Michigan, and one at LaRC.

AMAPS provides access to observational aerosol data from MISR (the Multi-angle Imaging SpectroRadiometer), MODIS (the Moderate Resolution Imaging Spectroradiometer), and AERONET (the Aerosol Robotic Network). We also provide output from the IMPACT (Integrated Massively Parallel Atmospheric Chemical Transport) model. In addition to creating the infrastructure for the Aerosol Measurement and Processing System, the AMAPS team is working on comparisons among these data sets as a guiding application.