Technical Approach

In order to leverage expertise and technologies located at NASA Ames, UMT, CMU, UWF, and Fetch Technologies, this research is utilizing a distributed architecture to generate real-time nowcasts and forecasts of ecosystem conditions. Data processing, autonomous discovery, and model validation and application occur in ten distinct steps:

1. Retrieve satellite data (AVHRR, MODIS, SeaWIFS), weather data (DAYMET, RUC) and ancillary data (DEM, soils) on a daily basis from national data centers and store at NASA / UMT; total data ~ 1 GB/day.
2. Data selection and processing plan generated by the IMAGEbot Planner at Ames to optimize processing and system execution.
3. Data processing coordinated by Theseus at Fetch and IMAGEbot at Ames.
4. Data products generated by TOPS, returned to Ames, delivered to UWF/CMU by IMAGEbot.
5. Bayes net (Tetrad IV) and other machine learning techniques used to analyze distributed heterogeneous datasets (from TOPS/IMAGEbot) to discover new causal relationships (UWF/CMU).
6. Causal models generated by Tetrad returned to NASA Ames.
7. New causal models incorporated into the TOPS biospheric modeling system.
8. New plan generated at Ames and executed by Theseus and IMAGEbot to produce forecasts.
9. Autonomous validation and update of model on continuous basis using machine learning techniques.
10. Validated models incorporated into TOPS for use in generation of daily forecast maps.