Interpreting the CRAS45

The Cooperative Institute for Meteorological Satellite Studies (CIMSS) uses the CIMSS Regional Assimilation System (CRAS) to assess the impact of space-based observations on numerical forecast accuracy. The CRAS is unique from other mesoscale models in that its development was guided by validating forecasts using information from the Geostationary Operational Environmental Satellites (GOES). During its 10-year development period, CRAS used observations from both the imager and sounder onboard GOES to validate the accuracy of the CRAS dynamic module, and to assess the accuracy of model physics. This development strategy has produced a forecast model that is proficient at predicting humidity, clouds, and precipitation.

The purpose of this document is to summarize known strengths and biases in the CRAS45, a real-time CRAS forecast currently generating products for the National Weather Service’s Advanced Weather Interactive Processing System (AWIPS).

Unique characteristics of the CRAS45

To accurately transport water vapor and clouds, a forecast model must preserve mass, along with spatial gradients of mass and momentum. The CRAS uses a semi-implicit time-stepping scheme that allows a longer time step. A time filter is applied every time step that is third-order accurate instead of the usual second-order schemes. Explicit three-dimensional advection of cloud and precipitation are included. The model is “pseudo-non-hydrostatic” in that the effects of precipitation drag are parameterized. Most importantly, the CRAS replaces the usual fourth-order diffusion (smoothing) with a sixth-order tangent filter. The CRAS physics contains many unique features. The convective parameterization includes the partitioning of cloud water (cirrus), and a non-local vertical turbulent exchange scheme drives the formation of single-layer cloud fields and allows no mixing across inversions.

Initializing CRAS45

A 12-hour spin-up forecast is used to initialize the CRAS45. This forecast is initialized using the GFS analysis on a 1/2 –degree grid to define all parameters. 3-hour boundary conditions are provided by the corresponding Global Forecast System (GFS) forecast. As the spin-up forecast proceeds, it checks for the availability of GOES observations and pauses to assimilate the information. The CRAS45 uses 3-layer precipitable water retrievals from the GOES sounders to adjust water vapor columns in clear fields of view. It uses cloud-top pressure and cloud effective amount from the GOES sounders to clear clouds or build 3D cloud fields. Combined, these retrievals provide complete coverage across each GOES scan. Moisture parameters from the spin up forecasts are then combined with winds and temperatures from the latest 6-hour forecast from the GFS. A surface analysis is performed using standard METAR (temperature, dewpoint, winds). The daily real-time graphic sea surface temperature (RTG SST) analysis from NCEP is used to define water surface temperature which is held constant throughout each forecast. Snow cover is defined using the Daily Integrated Multi-Sensor (IMS) snow cover from NESDIS.

Observed biases in the CRAS45

When the CRAS and the GFS are in disagreement over the placement of a low pressure system, defer to a location between the two model solutions, especially beyond 48 hours. The CRAS occasionally exhibits a leftward bias on systems undergoing rapid cyclogenesis on the forward side of a trough. There also is a greater tendency to occlude systems than observed. As a result, low-level temperatures may be too warm during and ahead of winter cyclones.
Moisture analysis and precipitation within 36 hours of initialization are generally superior to the North American Mesoscale (NAM) model, and usually the GFS. The CRAS assimilates water vapor and clouds from the GOES sounder. These data are not currently used by NCEP. This translates into precipitation forecasts that should be no worse than the GFS. Furthermore, the CRAS can provide a better estimation of precipitation chances under mid-level drying.
The local turbulent mixing parameterization in the CRAS does a good job of predicting the maximum daytime height of the boundary layer.
With the sky cover percentage and synthetic satellite imagery, CRAS cloud forecasts are excellent out to 36 hours, and often produce good imagery at 60 hours. The CRAS radiation physics is too efficient at attenuating shortwave radiation in the presence of cirrus. Daytime surface temperatures are usually slow to warm when these conditions are present.
On occasion, the gradient-preserving features of CRAS preserve warm layers and inversions too long in the vicinity of high pressure. Anomalously strong drying is occasionally observed in the lower and middle levels. This is particularly evident during the summer months. Defer to the GFS for comparison.
Surface wind speeds tend to be slightly higher than the NAM for speeds greater than 12 knots.
The CRAS currently does not have a soil moisture analysis. Each forecast starts with a soil moisture climatology. CRAS soil is moistened if it rains during a forecast. This may cause surface temperatures, especially during the day, to be too warm in areas where significant precipitation has fallen prior to model initialization. The surface dew point depression is greater than normal during and immediately preceding precipitation events.

If other biases are observed in CRAS forecasts, please contact Robert Aune, NOAA/NESDIS, Robert.Aune at noaa.gov.