IMPACT STUDIES OF AMVS AND SCATTEROMETER WINDS IN JMA GLOBAL OPERATIONAL NWP SYSTEM Koji Yamashita Japan Meteorological Agency / Numerical Prediction Division 1-3-4, Otemachi, Chiyoda-ku, Tokyo 100-8122, Japan Toshiyuki Ishibashi Meteorological Research Institute 1-1 Nagamine, Tsukuba-city, Ibaraki 305-0052 Japan Abstract To assess the impacts on Numarical Weather Prediction (NWP) of both Atmospheric Motion Vecter (AMV) and scatterometer winds, data denial studies of these data using Japan Meteorological Agancy (JMA) global NWP system were performed. Adjoint sensitivity analyses of in situ and satellite data including AMV and scatterometer winds were also carried out using the system. Results of data denial studies were confirmed as follows. Assimilation of the AMVs had a positive impact mainly in the tropics (TR) on the mean wind analysis at 200 hpa in the 2010 summer and the 2010/11 winter seasons. Its wind speed differences between with all AMV and without all AMV are extend to approximately 5 m/s. Assimilation of the AMVs brought an improvement of winds profiles against radiosonde observations in TR at 500-100 hpa. Assimilation of the polar AMVs had a weak impact in the equatorial and polar regions on the mean wind analysis at 200 hpa in the 2010/11 winter season. Its wind speed differences between with the polar AMVs and without the polar AMVs are less than 1 m/s. But assimilation of the polar AMVs had strong impacts on mean 500 hpa geopotential height analyzed fields in north and south poles areas. They reduced wind biases (BIASs) against radiosonde observations in the southern hemisphere (SH) at 700-400 hpa mainly. Assimilation of the scatterometer winds had a weak impact in TR on the mean wind analysis at 200 hpa in the 2010 summer season. Its wind speed differences between with the scatterometer winds and without the scatterometer winds are less than 1 m/s. But the scatterometer winds brought a strong positive impact in the North Atrantic. Adjoint sensitivity analyses of AMV derived from geostationary satellites show a strong positive impact mainly in the equatorial Indian Ocean and the equatorial eastern Pacific Ocean in the 2010 summer and winter seasons. These patterns are similar to the results obtained from the AMV denial study. 1. INTRODUCTION During the 10th International Winds Workshop (IWW10, February 2010, Tokyo, Japan), the following action: NWP centres to coordinate a joint AMV and scatterometer wind denial study (possibly two 6- week periods), also looking at adjoint sensitivity statistics where available. Aim to summarise in a report to the WMO GOS impact workshop and IWW11, both due to be held in the first half of 2012 has been adopted to learn more about the impact of satellite-derived wind data through a collaborative winds impact study. To examine the impacts to NWP of both AMV and scatterometer data, data denial studies of them using JMA global NWP system were performed also in JMA. Adjoint sensitivity analyses of AMV were also carried out using the system. In this paper, section 2 introduces outline of the global NWP system briefly. Section 3 describes the experimental design. The results of the experiments are discussed in section 4, and a summary is provided in section 5.
2. OUTLINE OF THE GLOBAL NWP SYSTEM AT JMA A low-resolution version of the operational global NWP system was used in experiments of data denial studies. A low-resolution version of the system is listed in Table 1. More details on a operational version system of the global NWP system are found in Nakagawa (2009). Table 1 Outline of the global NWP system at JMA Data Assimilation System for Global Spectral Model (GSM-DA) Method four-dimensional variational data assimilation (4D-Var) Resolution (inner model) T106L60 (hydrostatic, Gaussian grid, horizontal resolution approx. 110 km, model top 0.1 hpa) Assimilation window 6 hours (±3hours, time slots approx. 1 hour) TC bogus data Used Resolution Forecast time Global Spectral Model (GSM) TL319L60 (hydrostatic, reduced Gaussian grid, horizontal resolution approx. 60 km, model top 0.1 hpa) 84 hours/216 hours (00, 06, 18 UTC/12 UTC) 3. EXPERIMENTAL DESIGN (A) AMV and scatterometer wind denial study Experimental design is outlined in Table 2. Two 6-week periods were selected: one during the 2010 North Atlantic hurricane and North Pacific typhoon season (referred to hereafter as 2010SM) and one during the 2010/11 northern hemisphere (NH) winter season (2010/11WN). Results of four experiments (NOAMV(SM and WN), NOSCAT and NOPLR) are compared with CNTL which closely matched the JMA operational NWP system. Period 1 (2010SM): 1200 UTC 15 August 2010 1200 UTC 30 September 2010. Period 2 (2010/11WN): 1200 UTC 01 December 2010 1200 UTC 15 January 2011. Table 2 Experimental design of OSEs for AMVs and Scatterometer winds NOAMV (SM) NOSCAT (SM) NOAMV (WN) NOPLR (WN) CNTL (SM,WN) Geostationary AMVs Not used Used Not Used Used Used Polar AMVs Not used Used Not used Not used Used Scatterometer winds Used Not Used Used Used Used Other observations Used Used Used Used Used (B) Forecast Sensitivity to Observations (FSO) The relative impact of individual observations was evaluated using adjoint sensitivity diagnostics (Langland and Baker 2004) in configuration of Table3. These methods typically estimate the contribution of each observation towards reducing the 15-hour forecast error. Table 3 Configuration of FSO : energy norm, vertical range which the atmosphere is considered, and forecast error evaluation time which the impact range is measured. Energy Norm Vertical Range Forecast Error Dry Surface to 6 hpa 15hr Periods are as follows. Period 1 (2010WN): 1 January 2010 29 January 2010 only 00 UTC. Period 2 (2010SM): 1 August 2010 31 August 2010 only 00 UTC. 4. RESULTS OF THE EXPERIMENTS
11th International Winds Workshop, Auckland, New Zealand, 20-24 February 2012 Figure 1: Mean 500 hpa geopotential height [m] analyzed fields of NOAMV minus CNTL. Left side figure shows a result of 2010SM. Right side figure shows a result of 2010/11WN. Figure 2: Global vector difference of the mean wind analysis at 200 hpa for NOAMV minus CNTL. Left side figure shows a result of 2010SM. Right side figure shows a result of 2010/11WN. Shading indicates the speed of the difference vector [m/s]. Figure 3: U-component wind speed biases and RMSE against radiosonde observations in SH. Left side figure shows a result of 2010SM. Right side figure shows a result of 2010/11WN. Analysis (NOAMV) in red line, background (NOAMV) in blue line, analysis (CNTL) in orange line, background (CNTL) in green line. (A) CNTL VS NOAMV (SM and WN) Experiments of both of 2010SM and 2010/11WN had similar results. Figure 1 shows mean 500 hpa geopotential height analyzed fields of difference between NOAMV and CNTL. Differences are found in north and south poles and TR. Especially, impacts of both poles areas are strong and they are brought by assimilation of polar AMVs. An impact of TR is brought by geostationary satellite AMVs. Especially, the differences are large in the eastern half of Equatorial Pacific Ocean and Indian Ocean. These results are consistent with vector and wind analysis differences of 200 hpa (Figure 2). Its wind speed differences are extend to approximately 5 m/s. Figure 3 shows BIAS and root mean square error (RMSE) vertical profiles of U-component winds for SH against radiosonde observations. Slight reduction of RMSE for CNTL is shown. Reduction of BIASs and RMSEs are also shown in TR (not shown). There are positive impacts mainly on the mean wind analysis at the range of 500-100 hpa for TR and SH. No impacts were found in NH (not shown). Figure 4 shows anomaly correlation of 500 hpa geopotential height in SH. There is significant positive impact on the forecast skills from 5-day forecasts. In the other regions, it also shows same impacts as SH. Figure 5 shows mean eight typhoon track forecast errors. There is slight improvement in mean tropical cyclone (TC) track forecast
Figure 4: Anomaly correlation of 500 hpa geopotential height in SH. Left side figure shows a result of 2010SM. Right side figure shows a result of 2010/11WN. CNTL in blue line, NOAMV in red line. Figure 5: Mean eight typhoon track forecast errors in 2010SM. CNTL in blue line, NOAMV in red line, number of samples in red dots. Figure 6: Forecast improvement rate with regard to RMSEs of NOAMV against CNTL for 1 9 day forecasts. Upper figure shows a result of 2010SM. Bottom figure shows a result of 2010/11WN. The graph labeled Psea shows surface pressure, T850 shows 850 hpa temperatures, Z500 shows 500 hpa geopotential heights, Wspd850 shows 850 hpa wind speeds, and Wspd250 shows 250 hpa wind speeds. Positive values mean better scores. The green, brown, red and blue lines show the forecast improvement rate for the global, Northern Hemisphere (poleward of 20 N), tropical (20 S 20 N) and Southern Hemisphere (poleward of 20 S) regions, respectively. errors for AMVs from 24-hours forecast. Figure 6 shows forecast improvement rate with respect to RMSE for 1-9 day forecasts in 2010SM. Figure 6 shows many positive values or large improvement in CNTL. In particular there is a clear improvement from five-day forecasts ability in all terms except 250 hpa winds of TR. But 250 hpa wind forecast skills in TR against radiosonde observations have positive impact. (B) CNTL VS NO POLAR AMV (WN) Figure 7 shows mean 500 hpa geopotential height analyzed fields of difference between NO- POLAR-AMV (NOPLR) and CNTL. A large difference is found in north and south poles. It is shown by
11th International Winds Workshop, Auckland, New Zealand, 20-24 February 2012 Figure 7: Mean 500 hpa geopotential height [m] analyzed fields of NOPLR minus CNTL in 2010/11WN. (hpa) BIAS (m/s) Figure 8: Global vector difference of the mean wind analysis at 200 hpa for NOPLR minus CNTL in 2010/11WN. Others are same as Figure 2. RMSE (m/s) Figure 9: U-component wind speed biases and RMSE against radiosonde observations in SH in 2010/11WN. Analysis (NOPLR) in red line, background (NOPLR) in blue line, analysis (CNTL) in orange line, background (CNTL) in green line. Figure 10: Anomaly correlation of 500 hpa geopotential height in SH in 2010/11WN. CNTL in blue line, NOPLR in red line. Figure 11: Forecast improvement rate with regard to RMSEs of NOPLR against CNTL for 1 9 day forecasts in 2010/11WN. Others are same as Figure 6. similar difference patterns compared with a result in pole areas of section (A). But the polar AMVs had a weak impact in the tropics and polar region on the mean wind analysis at 200 hpa in the 2010/11WN (Figure 8). Its wind speed differences are less than 1 m/s. Because the polar AMVs are not used above 300 hpa for JMA s quality control, impacts of the polar AMVs in upper level are small. But they slightly reduced wind BIASs against radiosonde observations in mainly SH at 700-400 hpa (Figure 9). Figure 10 shows anomaly correlation of 500 hpa geopotential height in SH. There is a positive impact on the forecast skills from 3-day forecasts. In the other regions, it also shows same impacts as SH. Figure 11 shows forecast improvement rate with respect to RMSE for 1-9 day forecasts in 2010/11WN. They show many positive values in CNTL. (C) CNTL VS NOSCAT (SM) The scatterometer winds had a weak impact in TR on the mean wind analysis at 200 hpa in the 2010SM. Its wind speed differences between NOSCAT and CNTL are less than 1 m/s (Figure 13). Figure 12 shows mean 500 hpa geopotential height analyzed fields of difference between NOSCAT and CNTL. A difference is found in southern latitude of 60 degrees belt (Figure 12). The scatterometer winds bring an impact mainly in atmospheric boundary layer below 850 hpa, although their wind data
11th International Winds Workshop, Auckland, New Zealand, 20-24 February 2012 Figure 12: Mean 500 hpa geopotential height [m] analyzed fields of NOSCAT minus CNTL in 2010SM. Figure 13: Global vector difference of the mean wind analysis at 200 hpa for NOSCAT minus CNTL in 2010SM. Others are same as Figure 2. Figure 14: U-component wind speed biases and RMSE against radiosonde observations in SH in 2010SM. Analysis (NOSCAT) in red line, background (NOSCAT) in blue line, analysis (CNTL) in orange line, background (CNTL) in green line. Figure 16: Difference (NOSCAT-CNTL) in rms 48-hour forecasts error (m/s) for 500hPa geopotential height. Blue cycle shows in North Atrantic. Figure 15: Anomaly correlation of 500 hpa geopotential height in SH in 2010SM. CNTL in blue line, NOSCAT in red line. Figure 17: Mean eight typhoon track forecast errors in 2010SM. CNTL in blue line, NOSCAT in red line, number of samples in red dots. Figure 18: Forecast improvement rate with regard to RMSEs of NOSCAT against CNTL for 1 9 day forecasts in 2010SM. Others are same as Figure 6.
have a weak impact in upper level of the atmosphere, because their wind data belong to surface winds data. U-component winds in SH has improvement of BIAS against radiosonde observations below 850 hpa (Figure 14). Figure 15 shows anomaly correlation of 500 hpa geopotential height in SH. There is slight positive impact on the forecast skills. The scatterometer winds brought strong positive impacts in the North Atrantic associated with the passage of tropical cyclones (Figure 16). On the other hand, there is slight improvement in mean TC track forecast errors from 48-hours forecast in the Northwest Pacific Ocean (Figure 17). Figure 18 shows forecast improvement rate with respect to RMSE for 1-9 day forecasts in 2010SM. They show many positive values in CNTL, but 850 hpa wind and temperature forecast skills in TR are worse. (D) RESULTS OF FSO These figures show 15-hour forecast error contribution ratio (%, Figure 19). Positive values correspond to a decrease in the dry energy norm of forecast error. A left figure shows against each of all observations. A right figure shows against each of one observation. AMSU-A brightness temperature (AMSUA) and radiosonde observations (SONDE) against all observations are larger reduction of forecast error than other observations. Adjoint sensitivity analyses of AMV derived from geostationary satellites (AMV_GEO) show a strong positive impact mainly in the equatorial Indian Ocean and the equatorial eastern Pacific Ocean in the 2010SM (Figure 20) and 2010WN (not shown). These Figure 19: 15-hour forecast error contribution ratio (%) of the components of the observing system (all observations (left) and one observation (right)) against total 15-hour forecast error in 2010WN (WN) and 2010SM (SM). Positive values correspond to a decrease in the energy norm of forecast error. Left side figure shows against each of all observations Figure 20: Global horizontal map of 15-hour forecast error contribution ratio of AMV_GEO of the observing system against total 15-hour forecast error in 2010SM. Negative values correspond to a decrease in the energy norm of forecast error. 15-hour forecast error contribution of each component is a vertical summation.
patterns are similar to the results obtained from the AMV denial study. And best reduction of forecast error is also observed from AMV_GEO per one observation. The polar AMVs (AMV_POL) and scatterometer winds (SCAT) contribute the reduction of forecast errors, which are smaller than AMV_GEO. 5. SUMMARY Five experiments were performed in each of two seasons to assess the impact of AMVs and scatterometer winds. AMVs of geostationary satellites had large impacts of equatorial area in the eastern half of Equatorial Pacific Ocean and Indian Ocean at mean 200 hpa wind analyzed field. The AMVs basically brought a significant improvement from five-day forecast ability in all terms. AMVs of polar satellites had strong impacts in north and south poles in mean 500 hpa geopotential height analyzed fields. The AMVs brought a positive impact on the forecast skills from 3-day forecasts. Scatterometer winds had an impact of southern latitude of 60 degrees belt in mean 500 hpa geopotential height analyzed fields. They lead a slight positive impact on the forecast skills for 500 hpa in SH. They also brought strong positive impacts in the North Atrantic associated with the passage of tropical cyclones. In results of FSO, adjoint sensitivity analyses of AMV derived from geostationary satellites showed a strong positive impact mainly in the equatorial Indian Ocean and the equatorial eastern Pacific Ocean in the 2010SM and 2010WN. There was best reduction of forecast error by AMVs against one observation in the JMA operational NWP system. 6. REFERENCES Langland, R. H. and Baker, N. L. 2004. Estimation of observation impact using the NRL atmospheric variational data assimilation adjoint system. Tellus 56A, 189-201. Nakagawa, M., (2009): Outline of the High Resolution Global Model at the Japan Meteorological Agency. RSMC Tokyo-Typhoon Center Technical Review, 11, 1-13.