Land surface temperature to improve the assimilation of SEVIRI radiances over land

Transcription

1 Land surface temperature to improve the assimilation of SEVIRI radiances over land Stephanie Guedj*, Fatima Karbou and Florence Rabier CNRM/GAME, Météo-France and CNRS, 42 av Coriolis, Toulouse,France * Abstract This work aims to improve the assimilation of low-level SEVIRI (Spinning Enhanced Visible and Infra Red Imager) IR (Infra-red) observations over land to better constrain atmospheric analyses in meso scale models operating at Météo-France. To date, only high-peaking Water Vapour channels are operationally assimilated over land and IR channels are entirely rejected over land surfaces. The assimilation of IR observations over land is possible only if several limitations are accounted for: a reliable description of the surface emissivity, a more accurate estimation of the surface temperature and an effective bias correction scheme. Some feasibility studies have been undertaken in order to assimilate high density IR SEVIRI observations in the ALADIN-France system and also into the operational AROME system. The land surface emissivity was described using climatology from the EUMETSAT Land-SAF (Satellite Application Facilities). The use of these climatology was found very helpful in improving the RTTOV performances when simulating SEVIRI brightness temperatures (Tb) over Europe. The land surface emissivity and SEVIRI Tb were also used as input parameter in the radiative transfer model to retrieve the surface temperature (Ts) over Europe. The retrieved Ts was compared with independent Ts estimates (MODIS, Land-SAF products, ) and was then used within the assimilation process to constrain the analysis of surface temperature. A description of the methods for emissivity/temperature retrievals will be given. An evaluation of the retrieved Ts against independent measurements will be also presented. Finally, we will give an overview of assimilation and forecast experiment results when SEVIRI IR observations are assimilated. 1. Introduction The geographical domains of limited area models (LAM) covers mainly continental surfaces and to make the best use of as many satellite observations as possible, it is essential that they should be adequately treated during the data assimilation process. For instance, surface sensitive satellite observations are generally rejected over land because the direct models are unable to perform radiative transfer simulations close to observations. The Spinning Enhanced Visible and Infrared Imager (SEVIRI) measures top-of-atmosphere radiances and reflectances using 12 spectral channels every 15 min with a 3 km horizontal resolution at nadir (Schmetz et al, 2002). With regard to other atmospheric observations, SEVIRI data are particularly suitable for LAM thanks to their excellent spatial and temporal resolution. Information from SEVIRI have been assimilated in numerical weather prediction (NWP) systems for many years but remained restricted to water vapour (WV) channels, which are not sensitive to the surface, and to SEVIRI infrared (IR) channels over oceans (Kopken et al., 2004; Szyndel et al., 2005; Montmerle et al., 2007; Kelly, 2008 and Stengel et al., 2009). EUMETSAT Meteorological Satellite Conference, 5-9 September 2011, Oslo, Norway 1

2 a) b) Figure 1: Locations of a) synoptic observations and b) SEVIRI observations which were available over the ALADIN- France model on 02/08/2011 at 12UTC It is commonly admitted that the description of the IR land surface emissivity (LSE) and temperature (LST) fails to meet requirements in NWP models. Short of improvements, it will be difficult to assimilate more satellite observations over land surfaces (Borbas and Ruston, 2010; Li et al., 2010). LSE is usually set to 0.98 despite its strong variability with wavelength, surface condition, roughness and moisture (Borbas et al., 2007; Zhou et al., 2008, Seemann et al., 2008; Trigo et al., 2008). In addition, satellite observations are commonly used to estimate LST parameter and retrieval methods usually use a single IR channel or a combination of IR window channels (Jiang, 2007 and Pinker et al among many others). Large errors in LST are frequent in NWP models (Borbas and Ruston, 2010; and Edwards, 2010). In most LAMs, LST analysis is based on the use of: i) information from the past (short range forecast), ii) together with surface observations (two meter temperature and humidity), climatology atlases and a land surface scheme. The accuracy of such LST analysis relies mainly on the density of the surface observation network which usually is very uneven, as displayed in Figure 1a) for the ALADIN-France domain. As a consequence, using many more relevant remote sensing observations should help to better constrain LST analyses where few surface observations are assimilated (Figure 1). In this paper, for the first time, a selection of surface-sensitive IR SEVIRI observations are assimilated over Europe. Results of two assimilation experiments performed within the ALADIN- France forecast system are being examined. These developments were made possible after an adequate characterization of the LSE and LST during assimilation (Guedj et al., 2011a). This study aims to document assimilation experiments conducted during the summer-autumn period to assess the impact of incoming data from 3 SEVIRI channels over land on the quality of analyses and forecasts of the ALADIN-France LAM (Guedj et al., 2011b). Section 2 gives a description of the method used for the evaluation of the SEVIRI radiance impact. Validation of LST retrievals is presented in section 3. Impacts obtained in ALADIN-France are presented in section 4 and discussed in section Method and experiment design The operational ALADIN-France forecast system covers Western Europe at a resolution of 7.5 km and 70 vertical levels (Radnóti et al, 1996). The ALADIN-France model takes its lateral boundary conditions from the global model ARPEGE and its initial state from a cycled 3D-VAR data assimilation system (Fischer et al., 2005). ALADIN-France assimilates a substantial amount of heterogeneous observations including conventional measurements and remote sensing data. To date, the use of SEVIRI data in NWP systems is far from optimal. The radiometer SEVIRI, onboard the geostationary satellite of METEOSAT-9 provides full earth s disc images every 15 minutes in 12 EUMETSAT Meteorological Satellite Conference, 5-9 September 2011, Oslo, Norway 2

3 channels (Schmetz et al., 2002). They include three visible channels, one near infrared channel, and eight infrared (IR) channels. The spatial resolution of the IR channels is about 4 to 5 km over central Europe. ALADIN-France makes use of several SEVIRI channels at full resolution but, only data that are located over the sea or that are not sensitive to the surface (WV6.2 and WV7.3). Note that WV channels can be assimilated above low clouds as depicted in the Cloud Type product developed by CMS/SAF-NWC (Derrien and LeGléau, 2005). VIS channels and IR3.9, IR9.7 and IR13.4 are discarded in operations. Moreover, only one out of 5 pixels is selected and thinned to 70 km to avoid observational error correlations between pixels (Liu and Rabier, 2003). In order to extend the assimilation of SEVIRI IR channels over land, the general methodology is organized in 3 steps : Step 1 : To retrieve of LST using SEVIRI window channel IR10.8 Single-Channel method (inversion of the radiative transfer equation) Step 2 : To allocate retrievals of LST to other SEVIRI Bt for radiative transfer simulations Evaluation of Observations departures to Simulations Step 3 : Perform assimilation experiments of surface-sensitive IR SEVIRI observations over land using LST retrievals at channel IR10.8 Impacts on analyses and forecasts Assimilation experiments and bias correction : Two assimilation experiments were run over 12 weeks in 2009 (From August 1 to October 30) using the ALADIN-France system. The control run (called CTL hereafter) is based on the operational configuration do not consider IR SEVIRI channels over land. The other experiment was based on the first one with the assimilation of IR SEVIRI data over land (called EXP hereafter). As summarized in Table 1 and displayed, WV channels are used over land and sea surface and IR channels IR8.7, IR10.8 and IR12.0 are assimilated over sea surfaces in both experiments. IR8.7, IR10.8 and IR12.0 channels are additionally assimilated over land in EXP only, channel IR10.8 being exclusively used to retrieve LST. Table 1: Surface conditions for the assimilation of SEVIRI channels in EXP and CTL WV6.2 WV7.3 IR8.7 IR10.8 IR12.0 IR13.4 CTL land/sea land/sea sea sea sea sea EXP land/sea land/sea land/sea sea land/sea land/sea Besides assimilated channels, another main difference between the two experiments lies in the boundary conditions of the RTTOV model. RTTOV uses in CTL a constant value of LSE (0.98), the LST is taken from analyses, whereas in EXP, the radiative transfer model uses LSE atlases based on the Land-SAF emissivity maps and, whenever possible, LST deduced from SEVIRI channel IR10.8 using a single channel approach (LST IR10.8 ). In ALADIN-France, Variational Bias Correction (VarBC) system is used to correct the bias for satellite observations (Auligné et al., 2007). Four predictors are used for the regression to bias correct SEVIRI observations: 2 air-mass thickness (1000 to 300 hpa and 200 to 50 hpa), the Total Column Water Vapour (TCWV) and surface temperature (over land, LST from ALADIN-France in CTL and LST IR10.8 in EXP). Furthermore, observation error standard deviations have been assigned to 1.5K and 1.7K for WV6.2 and WV7.3 channels, 1.7K for IR13.4 and 1.2K for IR window channels. EUMETSAT Meteorological Satellite Conference, 5-9 September 2011, Oslo, Norway 3

4 3. Results 3.1. On the LST retrievals As mentioned earlier, a single-channel approach detailed in Guedj et al. (2011a) was used in clear sky conditions to retrieve LST using SEVIRI IR10.8 window channel (LST IR10.8 ). With this method, we use LSE information (called LSE-SAF hereafter) coming from the EUMETSAT LSA-SAF (Satellite Application Facility on Land Surface Analysis, Dacamara et al., 2006) combined with estimates of the atmospheric contribution to the measured radiance using the radiative transfer model RTTOV. The used LSE-SAF maps at SEVIRI channels are presented on the Figure 2 (Trigo et al., 2008). The atmospheric component is estimated using short range forecasts. Figure 2: SAF Land Surface Emissivity (LSE) atlas estimated at SEVIRI a) channel IR8.7, b) channel IR10.8, c) channel IR12.0. Atlases are compared with the default static emissivity map of 0.98 d). Daily SAF LSE data have been averaged to produce 1 atlas per windows channels for the period July 15 to August 15 of 2009 However, our method can produce inaccuracy in the computed LST, being sensitive to errors from input parameters namely emissivity, temperature and humidity profiles... It is therefore of prime importance to evaluate the computed LST prior to their use in data assimilation. Such a thorough evaluation study has been performed in Guedj et al. (2011a) with success against independent estimates of the surface temperature. In Figure 3, we show frequency histograms of LST from the Land-SAF with those calculated using channel IR10.8 data and also from analysis of the ALADIN- France model. Results are for a period of 76 days (from 1 August to 15 October 2009) over the entire ALADIN-France domain. There is a very good agreement between the Land-SAF LSTs and the retrieved one. As expected, the agreement is not as good for LST from the analysis of the ALADIN-France model as only 2 meter surface measurements are used to perform this analysis. The disagreement between ALADIN-France and the two other products is maximum at 12UTC for which the ALADIN model produces colder LST (not shown). Conversely, ALADIN-France produces warmer LST during nigh-time (not shown). This reflects the difficulty of reproducing the diurnal cycle in the model. EUMETSAT Meteorological Satellite Conference, 5-9 September 2011, Oslo, Norway 4

5 Figure 3: Frequency histograms of surface temperatures over 76 days (from 1st August to 15th October 2009) using LSTs from the Land-SAF, calculated using IR10.8 data, and those from ALADIN-France analysis. The use of LST retrievals combined to LSE-SAF have been shown to improve the performance of the radiative transfer model RTTOV. In particular, improvements are mainly located over the southern part of Europe where clear sky pixels are more frequent. As displayed in Figure 4, the clear sky observations and simulations are in much better agreement when the radiative tranfer model takes advantage of LST IR10.8 retrievals and the LSE-SAF (EXP). Figure 4: Observed Tb minus simulated Tb histograms of SEVIRI channel a) IR8.7, b) IR12.0 and c) IR13.4 for the period of July 15 to July 31 using CTL and EXP configurations Impacts on the analyse Significant differences were noted between humidity analyses from CTL and EXP (Guedj et al., 2011b). As displayed on Figure 5, the description of integrated parameters at the surface such as TCWV (Total Column Water Vapour) indicate that the assimilation of SEVIRI contribute to decrease moisture in the EXP analysis with regard to the CTL. The maximum impact is in Southern Europe which is associated with the highest number of assimilated SEVIRI observations. To assess the appropriateness of this changes, we used GPS data originating from the Villafranca site in Spain. These data are distributed by UCAR / COSMIC ( Ware et al. 2000). Time series of TCWV displayed on Figure 5 indicate that on overall the GPS measurements are EUMETSAT Meteorological Satellite Conference, 5-9 September 2011, Oslo, Norway 5

6 more consistent with EXP than with CTL at 12h and 18h. It can be noted that the representation of the diurnal cycle appears to be improved in EXP (Figure 6). Figure 5: Mean analysis difference in TCWV between EXP and CTL at a) 12h and 18h. TCWV maps are averaged over 45 days (1 September - 5 October 2009). Red (blue) colors means that EXP contribute to decrease (increase) moisture in the analysis with regard to the CTL. Time series of daily TCWV from GPS measurements (dotted black line), analysis of CTL (gray line) and those of EXP (black thin line) at (c) 12h and at (d) 18h. Figure 6: Average values of TCWV from GPS (Villafranca)(dotted black line), EXP (black thin line) and CTL (gray line) according to the assimilation cycle (0h-6h-12h-18h).TCWV values are averaged over 45 days (September 1 to October 15 of 2009) EUMETSAT Meteorological Satellite Conference, 5-9 September 2011, Oslo, Norway 6

7 3.3. Impacts on forecasts Forecast impacts were studied by examining the short range verification scores (up to 48h) improvements for several meteorological parameters as TCWV, precipitation, wind, relative humidity, and temperature fields. Overall, the forecast impact of EXP is positive over Southern Europe and neutral in other parts of model domain (Guedj et al., 2011b). Forecast scores of 6h cumulative precipitation were calculated for different forecast ranges (up to 48 hour range) and RMS were calculated by taking in-situ raingauges as reference. A big improvement on scores is observed around the 19th of September. To better understand the reasons for this change we exploited in-depth analysis and forecasts around the 19th of September. Figure 7 compares observed cumulated 6h rainfall valid the 19th of September at 6UTC with the corresponding simulated values from EXP and CTL 18h forecasts. The patterns of rainfall in EXP seem to be closer to observations than CTL. In particular, rainfall cells seem to be realistically spread over land in south west and north east France. One can speculate that if the humidity analysis is more realistic at 12UTC, it can be propagated by the 3DVAR into more realistic forecast of relative humidity fields and thus precipitations. More details about this study can be find in Guedj et al., 2011b. Figure 7: Maps of 06 hour total rainfall using RADAR observations and rain gauges for the 19th of September (called ANTILOPE analyses) and 18h-forecast precipitation fields from EXP and from CTL valid on the 19/09/2009 at 06h EUMETSAT Meteorological Satellite Conference, 5-9 September 2011, Oslo, Norway 7

8 4. Conclusion In this work, we have shown that SEVIRI surface-sensitive channels can be assimilated over land in a regional NWP system. Provided good estimates of the land surface emissivity (LSE) and land surface temperature (LST), have been shown to improve radiative transfer simulations and also to bring valuable information for analysed and forecasted fields. Firstly, the LSE produced by the EUMETSAT Land-SAF has been used to calculate the LST using SEVIRI channel IR10.8 data though the inversion of radiative transfer equation (as describe in Guedj et al., 2011a). Retrievals were successfully evaluated using independent LST estimates. Then, these two surface parameters were used at a later stage to describe the boundary conditions of the RTTOV model when calculating the model equivalent to the observations. Significant positive impacts were obtained over simulations when LST retrievals are used as input. Two assimilation experiments were run from the 1st August to the 15th October The control run was representative of the operational ALADIN-France NWP system and the second one assimilated SEVIRI channel IR8.7, IR12.0 and IR13.4 over land surfaces using LST retrievals and LSE atlas. The assimilation of these additional data showed positive impact on the analyses and forecasts of the ALADIN-France model. The main change in the analysis was about the humidity, primarily in southern Europe. The realism of this change in humidity has been successfully evaluated, thanks to independent measurements from GPS ground stations, radiosondes and synoptic data. There was also an improvement in forecasts, especially those of precipitations. We performed two additional assimilation experiments, on a smaller geographical domain (covering mainly France and northern Spain) with the AROME mesoscale model. We obtained similar modifications in moisture and temperature analysis. The evaluation of this change in moisture, in comparison with GPS data and 2 meter temperature, give preliminary positive results. Even if these changes appeared to be weaker in AROME than in ALADIN, a comprehensive study is underway to the understand the impact of SEVIRI data by studying possible interactions with the physical parameterizations of the models, seasonal impact, the use of radars... The impact studies will be the subject of a dedicated publication. Operational implementation of these developments have been undertaken for a future cycle. 5. References Auligné T, McNally T and Dee D Adaptative bias correction for satellite data in a numerical weather prediction system. Quart. J. Roy. Meteor. Soc. 133: Borbas E, Knuteson RO, Seemann SW, EWeisz, Moy L and Huang HL A high spectral resolution global land surface infrared emissivity database. In: Joint 2007 EUMETSAT Meteorological Satellite Conference and the 15th Satellite Meteorology and Oceanography Conference of the American Meteorological society. Borbas E and Ruston B The RTTOV UWiremis IR land surface emissivity module. NWPSAF Tech. Memo. AS Derrien, M. and H. LeGléau 2005, MSG/SEVIRI cloud mask and type from SAFNWC, International Journal of Remote Sensing, 26 (21), Edwards J Assessment of numerical weather forecasts using satellite land surface temperatures. In: 19th Symposium on Boundary Layers and Turbulence. Keystone, Colorado. Fischer C, Montmerle T, Berre L, Auger L and Stefanescu S An overview of the variational assimilation in the ALADIN/France numerical weather-prediction system wave driven circulation of the mesosphere. Quart. J. Roy. Meteor. Soc. 131: Guedj S, Karbou F and Rabier F. 2011a. Land surface temperature estimation to improve the assimilation of SEVIRI radiances over land. Journal of Geophysical Reasearch 116, doi: /2011jd Guedj S, Karbou F, Rabier F, Guidard V and Montmerle T. 2011b, Improved assimilation of observations from SEVIRI over land, Quart. J. Roy. Meteor. Soc, submitted. EUMETSAT Meteorological Satellite Conference, 5-9 September 2011, Oslo, Norway 8

9 Jiang G Retrievals of land surface emissivity and land surface temperature from MSG1-SEVIRI data. PhD thesis, Université Louis Pasteur de Strasbourg. Kelly G Preparations and experiments to assimilate satellite image data into high resolution NWP. Technical Report 522, Met Office Meteorology Reasearch and Development. Li Z, Li J, Jin X, Schmit T, Borbas E and Goldberg M An objective methodology for infrared land surface emissivity evaluation. Journal of Geophysical Research 115(D22): D Liu ZQ and Rabier F The potential of high-density observations for numerical weather prediction: A study with simulated observations.quart. J. Roy. Meteor. Soc. 129(594). Montmerle T, Rabier F and Fisher C Relative impact of polar-orbiting and geostationary satellite radiances in the Aladin/France numerical weather prediction system.quart. J. Roy. Meteor.Soc.,133. Pinker R, Sun D, Hung MP and Li C Evaluation of satellite estimates of land surface temperature from GOES over United States. Applied Meteorology and Climatology 48(1): Radnóti G, Ajjaji R, Bubnová R, CaianM, Cordoneanu E, Von der Emde K, Gril JD, Hoffman J, Horányi A, Issara S, Ivanovici V, Janousek M, Joly A, Lemoigne P and Malardel S The spectral limited area model ARPEGE-ALADIN. In: PWPRR Report, vol. n.7. World Met. Org., pp Schmetz J, Pili J, Tjemkes S, Just D, Kerkmann J, Rota S and Ratier A An introduction to Meteosat Second Generation (MSG). Bulletin of Amer. Meteor. Soc. 83. Seemann S, Borbas E, Knuteson R, Stephenson G and Huang HL Development of a global infrared emissivity database for application to clear sky sounding retrievals from multi-spectral satellite radiances measurements. Journal of Applied Meteorology and Climatology 47: Sobrino J, JC Jimenez-Munoz JEK, Gomez M, Romaguera M and Soria G Single-channel and two-channel methods for land surface temperature retrieval fromdais data and its application to the Barrax site. Int. J. of Rem. Sens. 25. Stengel M, Unden P, Linskog M, Dahlgren P, Gustafsson N, and Bennartz R Assimilation of SEVIRI infrared radiances with HIRLAM 4D-Var 135(645). Szyndel M, Kelly G and Thépaut JN Evaluation of potential benefit of assimilation of SEVIRI water vapor radiances data from METEOSAT-8 into global numerical weather prediction analyses. AtmosphericSciences letters 6. Trigo I, Peres L, DaCamara C and Freitas S Thermal land surface emissivity retrieved from SEVIRI/Meteosat. IEEE Transactions on Goescience and Remote Sensing 46(2). Ware R, Fulker DW, Stein SA, Anderson DN, Avery SK, Clark RD, Droegemeier K, Kuettner JP and Minster JB Suominet: A real-time national GPS network for atmospheric research and education. Bulletin of Amer. Meteor. Soc. 81: EUMETSAT Meteorological Satellite Conference, 5-9 September 2011, Oslo, Norway 9