Journal of the Meteorological Society of Japan, Vol. 82, No. 1B, pp. 453--457, 2004 453 Impact of GPS and TMI Precipitable Water Data on Mesoscale Numerical Weather Prediction Model Forecasts Ko KOIZUMI and Yoshiaki SATO Numerical Prediction Division, Japan Meteorological Agency, Tokyo, Japan (Manuscript received 5 May 2003, in revised form November 16, 2003) Abstract Observation system experiments with JMA MesoScale model, were performed for precipitable water data derived from TMI (TRMM Microwave Imager) and ground-based GPS observation, by using a fourdimensional variational assimilation method. Since GPS data exists over land only, and TMI data are available only over ocean, use of both data can provide information about water vapor complementally over whole analysis domain. Although the number of experiments is not sufficient yet, the results so far suggest that the complementary use of TMI and GPS precipitable water data, can improve the precipitation forecast of the model. 1. Introduction Prediction of heavy rainfall is one of the most important subjects of a weather forecast. Improvement of the numerical weather prediction (NWP) is essential to achieve a quantitative forecast of heavy precipitation with lead-time. JMA MesoScale Model (MSM) covers Japan, and its surrounding areas (3600 km 2880 km), with a horizontal resolution of 10 km. It is run four times a day, and produces an 18-hour forecast to support very short-range forecast, aiming at the disaster prevention (JMA 2002a). A mesoscale four-dimensional variational data assimilation system (Meso 4D-VAR) was implemented in March 2002, which conducts 3- hour cycle analyses to prepare initial conditions for MSM. The assimilation window is set to be three hours previous to the initial time (JMA 2002b). While the Meso 4D-VAR has greatly improved precipitation forecasts of MSM, it is still not a easy task to make a quantitative forecast Corresponding author: Ko Koizumi, Numerical Prediction Division, Japan Meteorological Agency, 1-3-4 Ote-machi Chiyoda-ku Tokyo, 100-8122 Japan. E-mail: kkoizumi@met.kishou.go.jp ( 2004, Meteorological Society of Japan of heavy precipitation with sufficient lead-time. One of the reasons is that there is very little information about three-dimensional water vapor distribution at the initial time. Therefore, water vapor data of GPS, and TMI observation, are expected to have great value when assimilated to NWP models. In this report, results of observation system experiments (OSEs) of precipitable water data of GPS, and/or TRMM microwave imager (TMI) are presented. Since GPS data exists over land only, and TMI data are available only over ocean, use of both data is expected to give information about water vapor complementally over the whole analysis domain. 2. Data and assimilation method 2.1 GPS and TMI observations In Japan, a nationwide GPS network, called GPS Earth Observing Network (GEONET), is operated by the Geographical Survey Institute (GSI) to monitor crustal deformation over the Japanese Islands. The network has grown from 610 sites in 1996, to about 1000 now. The GEONET data was analyzed with the precise point positioning (PPP) technique, using the GIPSY-OASIS II software developed by the Jet Propulsion Laboratory (JPL) (Release
454 Journal of the Meteorological Society of Japan Vol. 82, No. 1B Fig. 1. Scatter diagram of observation and first-guess of TMI-PW (left) and GPS-PW (right). 2.6 Zumberge et al. 1997). PPP can directly solve receiver clock error, along with other parameters, by using the sophisticated stochastic filtering technique with precise satellite orbit and satellite clock error information. The analysis used precise fiducial-free orbits and satellite clocks, provided by the JPL, and the Niell s mapping function (Niell 1996). The ZTD and the tropospheric delay gradient were modeled as a random walk process, with the scale parameters 5:0 10 8 km/sqrt(s) and 5:0 10 9 km/sqrt(s), respectively, following Bar-Sever et al. (1998). The site positions were estimated daily, but the ZTD and the tropospheric delay gradient were estimated every five minutes. In this experiment, the ZTD data were averaged over one hour and converted to precipitable water (PW) by using surface pressure and vertical temperature profile of first-guess field. Then a quality control process, proposed by Mannoji et al. (1998), was applied to the GPS- PW data. The TMI precipitable water data were retrieved by NASDA /EORC using Shibata (1994). In this experiment one TMI-PW observation point within each 70 70 km area, was used to avoid correlation of observation error. Figure 1 shows scatter diagrams of observation and first-guess of TMI-PW and GPS-PW. Both data show good correlation with the firstguess. Given that root mean square of difference between observation and first-guess is 5.13 mm and 2.53 mm for TMI-PW and GPS- PW respectively, a crude estimation of the observation errors were made as 3 mm and 1 mm, which were about halves of the RMSs of observation-first-guess difference. More sophisticated estimation of the observation errors will be necessary in the future. 2.2 Model and assimilation system The MSM is a hydrostatic spectral model, with a horizontal resolution of 10 km and 40 vertical levels up to 10 hpa. The lateral boundary condition is provided by the regional spectral model (RSM), with a horizontal resolution of 20 km starting from initial conditions at 00 and 12 UTC. The initial condition of MSM is prepared by Meso 4D-Var with 3-hour assimilation windows. The cost function of Meso 4D-Var, consists of a background term, observation terms, and a penalty term for reducing gravity wave noise. The control variables are the initial and boundary conditions of unbalanced wind, temperature, surface pressure, and specific humidity. The background error statistics are obtained by using the NMC method. The horizontal background error correlations are assumed to be homogeneous, and Gaussian type, to significantly reduce the memory requirement. An incremental method is taken for reducing computational time. The forward model
March 2004 K. KOIZUMI and Y. SATO 455 Fig. 2. Left: Analysis increment of precipitable water (mm) at 18 UTC 18 th June 2001 in the case that; (a) TMI-PW data were added to the conventional data, (b) GPS-PW data were added to the conventional data; and, (c) both TMI-PW and GPS-PW data were added to the conventional data. Contour interval is 1 mm. Right: Positions of corresponding data. White circles show points where departure of observation from first-guess is positive, and black ones show those where departure is negative. Diameter of circles varies according to the absolute value of the departure. in this system has the same architecture as the forecast model (viz. MSM), except that its horizontal resolution is reduced to 20 km. The adjoint model has the same dynamical process as the forward model, while its physical processes include moist processes, boundary layer processes, long-wave radiation and horizontal diffusion only. This operational meso 4D-VAR system was employed in the experiments though an observation operator for precipitable water is added since PW data are not assimilated operation-
456 Journal of the Meteorological Society of Japan Vol. 82, No. 1B Fig. 3. Three hour precipitation amount during 12 15 UTC 19 th June 2001 of control run, TMI run, TMI þ GPS run and observation from left to right respectively. Initial time of forecasts is 12 UTC 19 th June 2001. ally. Other than PW data, data from radiosonde, synop, ship, buoy, aircraft, windprofiler, as well as the radar-amedas precipitation data, are assimilated. 3. Experiment design Three sets of three-hourly forecast-analysis cycles were executed starting from 18 UTC on 18 th June 2001. In the experiment period there occurred heavy rainfall over the western part of Japan which caused landslide disasters. The three sets include a control run, a TMI run and atmiþgps run. The experiment with GPS- PW only might well be made, but it is not affordable due to the limitation of the computational resources. In the control run, conventional observation data, JMA wind-profiler data and radar- AMeDAS precipitation analysis data were assimilated. In the TMI run, TMI-PW data were added to the control run, and in the TMI þ GPS run GPS-PW data were added to the TMI run. In each run, 18 hour forecasts were made four times a day (initial times are 00, 06, 12 and 18 UTC), and 3-hour precipitation forecasts were evaluated, using threat scores calculated against radar-amedas precipitation analysis. 4. Results and discussion Figure 2a shows data positions, and analysis increments of precipitable water in the case of assimilating TMI-PW at 18 UTC of June 18 th 2001. Though TMI-PW data exist only over ocean, the analysis increments expanded to inland regions due to the background error covariance. However, those increments of inland regions are in disagreement in some places with the GPS-PW observation departure (Fig. 2b). Such disagreement was seen in several cases in the experiment, which suggests that the model shows different behavior about moisture forecast over land and over ocean, and hence the assumption of homogeneity of background error may corrupt across the land-sea border. Therefore, both observation over ocean and over land are important, and using both TMI-PW and GPS-PW can provide more accurate moisture distribution of the whole analysis domain (Fig. 2c). Figure 3 shows an example. Three-hour precipitation of 0 3 hour forecasts of each run, whose initial time is 12 UTC 19 th June 2001 are shown. A spurious heavy rain area (over 30 mm per 3-hour shown by A in Fig. 3), which was produced by precipitation assimilation, was reduced by using TMI-PW data but the precipitation amount was too much suppressed (below 10 mm per 3-hour shown by B ). Complementary use of TMI-PW (moisture information over sea) and GPS-PW (moisture information over land) gave 10 20 mm per 3- hour precipitation over the area, which was the best result among three experiments. Figure 4 shows threat scores of 1 mm/3-hour precipitation calculated for 15 forecasts of each run. Scores of TMI run surpass those of the control run for all forecast time and TMI þ GPS run, shows the best scores among three experiments for FT ¼ 3 6 and later. Since atmospheric disturbances generally move from western or southern ocean to Japan, precipitation forecasts over Japan are sensitive to the water vapor distribution over the
March 2004 K. KOIZUMI and Y. SATO 457 ocean, of which TMI-PW can provide good information. On the other hand, small-scale disturbances generated in the inland area are controlled by water vapor distribution of that region, of which GPS-PW has more accurate information than TMI data. Hence, it is reasonable that TMI þ GPS run could provide the best forecasts among three experiments. The result is very promising about the complementary use of GPS-PW and TMI-PW, however, it is not clear how each data contributed to the improvement of forecasts. Especially, ground-based GPS-PW data seem to affect to forecasts of longer lead-time, which is a little difficult to explain. Further research, especially experiments with GPS-PW only, should be made in the future. Fig. 4. Threat scores of forecast precipitation over 1 mm/3 hour. Forecast time is 3 0 (within assimilation window), 0 3, 3 6, 6 9, 9 12, 12 15 and 15 18 hour from left to right respectively. Solid bold line shows those of TMI þ GPS run, solid thin line TMI run and dashed control run. Scores are calculated for 15 cases during 18 UTC 18 th June 2001 to 06 UTC 22 nd June 2001 against radar-amedas precipitation analysis data, which are interpolated to the model grid. References Bar-Sever, Y.E., P.M. Kroger and J.A. Borjesson, 1998: Estimating horizontal gradients of tropospheric path delay with a single GPS receiver. J. Geophys. Res., 103, 5019 5035. JMA, 2003: Meso-Scale Model, Outline of the operational numerical weather prediction at the Japan Meteorological Agency, 82 83 (available from Japan Meteorological Agency). JMA, 2003: Meso-scale Analysis, Outline of the operational numerical weather prediction at the Japan Meteorological Agency, 26 32 (available from Japan Meteorological Agency). Mannoji, N., H. Tada, Y. Hatanaka, R. Ohtani and I. Naito, 1998: An impact study of precipitable water estimated from ground-based GPS network over Japan, Proceedings of 12th Conference on Numerical Weather Prediction, 77 80. Niell, A.E., 1996: Global mapping functions for the atmosohere delay at radio wavelength. J. Geophys. Res., 101, 3227 3246. Zumberge, J.F., M.B. Heflin, D.C. Jefferson, M.M. Watkins and F.H. Webb, 1997: Precise point positioning for the efficient and robust analysis of GPS data from large networks. J. Geophys. Res., 102, 5005 5017. Shibata, 1994: Determination of water vapor and liquid water content by an iterative method. Met. Atmos. Phys., 54, 173 181.