NWP Modeling | Forward Modeling | Retrieval Science | Wind | Visibility/Stability | Parallel Computing Cluster

NWP Modeling

Nonhydrostatic numerical models, such as the MM5 and WRF, are used to generate realistic, high resolution atmospheric and cloud microphysical data sets for ingest into a radiative transfer model.  The bulk characteristics of clouds are represented in these datasets by the mixing ratios and effective mean diameters of five microphysical species (cloud water, rain water, ice, snow, and graupel). Sophisticated microphysical parameterization schemes in the MM5 and WRF models are capable of providing realistic mixing ratios for each of the required species.  Effective diameters are then calculated using a gamma distribution that incorporates both the mixing ratio of a given species and various assumptions implicit to each microphysics scheme.

Forward Modeling

An accurate and efficient clear sky forward model is developed with the various features required for atmospheric profile and radiance data assimilation in a NWP context. This includes validating the underlying line-by-line absorption and radiative transfer models, efficient and representative parameterization of the line-by-line results, and development of analytical jacobians and adjoints of the model.  Generally, the “fast forward” model is developed with a training set of profiles spanning a large range of atmospheric conditions. Using the profiles, a line-by-line model calculates accurate transmittances. The transmittances are regressed against profile derived predictor values. The resulting coefficients can be applied to any profile to quickly calculate radiances.

For the GIFTS Fast Model, channel-weighted level-to-space transmittances (from NASA/GIFTS Clear-Sky Fast Model) are coupled with models of the spectral reflectance and transmittance of ice and water phase clouds. This extension makes possible the rapid computation of realistic top-of-atmosphere radiances in the presence of clouds. The GIFTS Fast Model simulations operate over large spatial domains, called GIFTS data cubes, that are representative of the instrument footprint - on input, a 128 x 128 array of atmospheric profiles; on output, a 128 x 128 array of GIFTS spectra (radiance, brightness temperature, transmittance). Input data cubes are generated on appropriate spatial scales by mesoscale models, such as MM5 or WRF, to provide condensate profiles (phase, size, number density) in addition to profiles of temperature, moisture and ozone mixing ratio. Presently a model upgrade is underway; to include more ice crystal habits, to allow for two cloud layers, and to incorporate an ecosystem-based land surface emissivity. Model verification is achieved by comparison against LBLRTM/DISORT model predictions at high spectral resolution and integrated over selected GIFTS channel bandpasses.

Retrieval Science

The objective is to develop algorithms for retrieving boundary layer moisture profiles from hyperspectral data. Successive analyses of these profiles will be fed into automated wind tracking algorithms developed at UW-CIMSS. In our approach, hyperspectral, 4-km spatial resolution GIFTS measurements, and time/space co-located numerical model profiles, will be used to retrieve boundary moisture patterns in cloud-free regions and partly cloudy regions. The resulting fields will provide the analyses used for tracking moisture features over time to obtain wind (directions and magnitudes).

The goals of this task are: (1) develop boundary layer moisture profile retrieval algorithms for use with high spectral resolution GIFTS radiances; (2) use simulated high spectral and high temporal resolution GIFTS radiances for experimental retrieval of boundary layer moisture structure with time continutity; and (3) validate and refine the retrieval algorithms, in preparation for their use with actual GIFTS data. AIRS, IASI data are used.

A two-step algorithm, a Principle Component Regression (PCR; Huang and Antonelli 2000) followed by a non-linear physical retrieval method (Li and Huang 1999; Li et al. 2000), will be adopted for processing GIFTS data. The regression retrieval serves as the first guess in the physical retrieval. PCR simply uses the projections of the predictor variables (brightness temperature) onto a subset of principle components. The first guess can be also taken from the numerical model forecast profile if it is available.  In order to make the physical retrieval processing more efficient, an optimal subset of GIFTS water vapor absorption channels will provide the most moisture information content. Retrieval sensitivity to other parameters such as surface skin temperature, surface emissivity, topography, and cloud effect (in partly cloudy regions) is also being investigated to optimize the retrieval research.

Extacting profiles under partly cloudy regions is also under development. There are essentially three ways to extract profile information from cloud contaminated radiances: (1) cloud-clearing using spatially adjacent cloud contaminated radiance measurements, (2) retrieval based upon the assumption of opaque cloud conditions, and (3) retrieval or radiance assimilation using a physically correct cloud radiative transfer model which accounts for the absorption and scattering of the radiance observed. A fast cloudy radiative transfer model and the AIRS data will be used to test those algorithms for sounding from cloudy radiances.

Wind

Traditional methods for tracking motion vectors from satellite data rely on sequential imagery and the motion of clouds and water vapor (WV) features in this sequence. The limitation of this method is that tracers can be identified only at cloud-top levels, and/or when layer-mean WV features can be identified. Height assignment of these tracer vectors is also problematic.

MURI is helping us to explore a new approach to deriving winds from satellite radiances. Using hyperspectral data from emerging new satellite sounders, we can obtain profiles of winds from retrieved, altitude-resolved fields of WV. This approach addresses the two primary limitations above. Case studies from both simulated (GIFTS) and real (NASTI, AIRS) datasets are being explored to optimize the new techniaue.

Visibility/Stability

Proposed geostationary hyperspectral satellite instruments will sample the Earth's atmosphere at high horizontal (4 km), vertical (1 km), and temporal (5-15 min) resolution, thus providing an unprecedented resource for the study of convective weather phenomena. Retrievals of temperature and moisture derived from simulated hyperspectral satellite imagery are being utilized to compute atmospheric stability parameters such as Convective Available Potential Energy (CAPE) and Lifted Index (LI). Strong horizontal stability gradients coupled with localized stability minima are primary locations of thunderstorm initiation. These stability fields are coupled with cumulus cloud growth and cloud-top microphysical analyses to provide 0-1 hour forecasts of future thunderstorm development over both land and oceans.

Parallel Computing Cluster

The science efforts discussed in the sections above rely upon large NWP simulations. Our NWP modeling capabilities are dependent on the available computer resources, which increased significantly in 2004. The Space Science and Engineering Center was awarded a large grant which was used to purchase a shared memory cluster from SGI. An Altix 3700 system with 24 processor and 192 GB of ram is now used primarily for NWP simulations. We will also be expanding the system to 32 processors in the near future. The Altix system has provided a 2.5 times improvement in model run time and a 12 times increase in possible domain size.