NOAA Privacy Policy | NWS Disclaimer
N.O.A.A. logo HWRF banner image National Weather Service logo

April 07, 2011 Meeting Summary

Dr. Takemasa Miyoshi presented his work with Dr. Masaru Kunii on "Data Assimilation Studies on Typhoon Sinlaku (2008) Using the WRF-LETKF System." Takemasa began by giving a brief overview of data assimilation (DA). He demonstrated the impact of using DA by showing a plot comparing forecast track to observations for two cases using the same NWP model and observations: one with the JMA operational system with 4D-Var and one using Local Ensemble Transform Kalman Filter (LETKF). The LETKF forecast track was much closer to observations. Takemasa explained this was likely because LETKF has more of a flow dependence which benefits TCs and their nature. Next, the experiment approach was described. First, Takemasa and Masaru applied LETKF to toy models to make sure their results made sense. Then they used intermediate, more realistic models, and if that was successful, LETKF was applied to real systems or operational models. Ideas for improving the LETKF system included using satellite bias correction, adaptive inflation, and ensemble sensitivity.

After briefly going over a schematic diagram of an Ensemble Kalman Filter (EnKF), Takemasa explained that the analysis in LETKF is provided by a linear combination of the forecast ensemble. Next, Takemasa listed some applications for the LETKF code, which was developed from scratch, specifically WRF. Since WRF and LETKF had different interfaces, some adapters are needed when the two are combined. This was shown in a flowchart. For the experiments conducted by Takemasa and Masaru, LETKF used 20 ensemble members, unperturbed or fixed lateral boundary conditions, adaptive fixed covariance inflation at 20%, and 400 km horizontal and 0.4 log p covariance localization. WRF settings included a 136 by 108 by 39 domain and 60km horizontal resolution. Next, a 6-hr forecast was shown for Typhoon Sinlaku after 9 days of cycled model runs. To demonstrate LETKF performance, satellite observations from GSMAP were compared to an NCEP GDAS forecast, an assimilation done without observations, and LETKF assimilation. While no observations captured larger features, like areas of high pressure, NCEP GDAS and LETKF capture Sinlaku better.

Next, Takemasa described using adaptive inflation to improve the LETKF system. Adaptive inflation is useful for improving ensemble spread, and it can account for things like model errors or limited ensemble size. In an ensemble, the inflation factor is usually fixed and tunable. With adaptive inflation, different inflation factor values can be used to even out areas with sparse observations or dense observations. In plots of bias and RMSE values comparing adaptive inflation to fixed inflation, the RMSE and bias are both reduced using the former. Takemasa then showed plots of bias and RMSE comparing CTRL, which used fixed SST values for all ensemble members, to perturbed SST values. The RMSE and bias were reduced when SSTs were perturbed, compared to fixed values.

Takemasa then described his and Masaru's work with ensemble sensitivity to improve the LETKF. He explained how they calculated observation impact on assimilation without using an adjoint model and applied this method to real observations for the first time. Of the different types of observations that were assimilated, the biggest impact was from upper atmosphere soundings. Takemasa showed a specific example of the impact dropsondes have on a typhoon assimilation. A plot showing dropsondes from a DOTSTAR flight in Typhoon Sinlaku was shown. Blue dots at the sonde locations indicated they improved the forecast while red dots degraded the forecast. To prove certain sondes degraded the forecast, Takemasa described how removing the sondes from the assimilation improved the track forecast. This method is called denying negative impact. A plot showing both upper and lower level dropsondes was shown for Typhoon Sinlaku. This raised the question of at what stages in the lifetime of a TC are upper level dropsondes more important than lower level ones. More work on this branch of sensitivity experiments is ongoing.

Takemasa concluded by describing Masaru's and his future work. The main goal was to focus on TC intensity forecasting. This includes looking into the air-sea coupled covariance around the TC, higher resolution experiments to better resolve TC structures, and predictability studies by ensemble prediction. More TCS-08 case studies will be conducted with greater analysis of aircraft observation impacts, use of AIRS retrievals, use of rapid-scan cloud images, and direct assimilation of best track data. There is a possibility of further ITOP-10 case studies as well.

Please e-mail comments, questions, or suggestions about the contents of this webpage to Janna O'Connor, at

Home | HWRF Main Page
EMC | NCEP | National Weather Service | NOAA | Department of Commerce