Hung Lung Huang (PI)
University of Wisconsin-Madison
allenh@ssec.wisc.edu

FPGA Re-Configurable Computation Demonstration: AIRS/MODIS Co-Registration and Cloud Characterization for Data Assimilation

The AIRS high spectral resolution infrared sensor aboard the EOS Aqua satellite has demonstrated high quality radiometric measurements.  These measurements are available at global scales and are suitable for Numerical Weather Prediction (NWP) use.  Recent studies conducted by Joint Center for Satellite Data Assimilation (JCSDA) have shown that using only approximately 0.45% of clear AIRS data can provide significant positive impact on 3-6 days of forecasts for both northern and southern hemispheres.  The currently limited use of AIRS data is due to the present inability to directly assimilate observations that contain cloud, and these represent the majority of observations.  There is a strong expectation that the optimal use of larger amounts of AIRS data in both clear and cloudy areas will significantly improve model forecasts by providing additional information to the model.  In order to use larger amounts of AIRS data, more sophisticated algorithms, which may be computationally intensive, are needed.  In the algorithm designing process, model forecast improvements should not be sacrificed simply to strike a balance between algorithm accuracy and computational capacity.  This proposal seeks,    to develop the comprehensive algorithms for enabling the optimal use of AIRS observations for model forecast improvement;    and to implement these algorithms using high performance field programmable gate array (FPGA) reconfigurable computer technology. The proposed comprehensive algorithms include the synergistic processing of highly accurate, co-registered MODIS and AIRS data.  The co-registered AIRS/MODIS data will be used for deriving high spatial resolution, cloud-cleared radiances over the ocean as well as over land areas, where the current performance is relative poor.  The associated AIRS sub-pixel cloud information such as cloud mask, cloud hydrometeor phase, and cloud height are also derived to prepare cloud-cleared and cloudy data for assimilation into NWP models. In concert with the algorithm development, this proposed study will also utilize high performance FPGA reconfigurable computer technology for algorithm implementation.  FPGAs are logic devices that offer in-circuit re-programmability and high performance computation with the potential for many parallel computational streams.  It is well documented that the execution of applications offered by re-configurable computing from an FPGA processor can provide orders-of-magnitude higher performance over conventional von Neumann general-purpose processors since the latter spend a lot of machine cycles moving data through a register chain.  FPGA coprocessor cards implement algorithms in hardware which can then be plugged into desktop or larger computer systems. The expected outcomes include 1) enhanced and complementary EOS AIRS/AMSU data usage, 2) optimal use of EOS AIRS/MODIS data in NWP, 3) demonstration of high performance computing power to achieve efficient real-time data processing, 4) preparatory demonstration of optimal use of global high spatial resolution NPP/NPOESS CrIS/VIIRS data, 5) risk reduction for sustained GOES-R HES/ABI high temporal resolution synergistic data processing, and, most importantly, 6) to provide NWP models with processed ultraspectral data that meet time latency in data assimilation to further improve analysis and 3 to 6 day forecast skill.

+ Back to Participant Listing