THE OFFLINE VERSION OF SURFEX COUPLED TO THE

SURFEX@RMI THE OFFLINE VERSION OF SURFEX COUPLED TO THE OPERATIONAL ALADIN FORECAST OVER BELGIUM: A TOOL TO IMPROVE WINTER SCREEN TEMPERATURE Royal Meteorological Institute, Brussels, Belgium Corresponding author address: R. Hamdi, Royal Meteorological Institute, Avenue Circulaire, 3, B 1180 Brussels, Belgium. E mail: rafiq.hamdi@oma.be. Page 1 of 24

ABSTRACT In this study a state of the art surface only model (SURFEX) is used offline in order to remove the positive feedbacks between the surface and the atmosphere in stable situations. The coupling occurs at ALADIN's lowest vertical level. Variables that must be exchanged are forecasted: air temperature, specific humidity, atmospheric pressure, incoming global radiation, incoming long wave radiation, precipitation rate and wind speed. The purpose of this note is to compare the offline diagnostic of screen temperature by SURFEX with the operational ALADIN forecast during the cool season (2007 2008). Results show a cyclic dependence of the ALADIN errors on the forecast range with large systematic error during the night. This indicates that the ALADIN errors occurs during cold winter night which is in agreement with the findings in the note of Termonia (2001). This underestimation of the winter minimum screen temperature is largely reduced by SURFEX except for January 2008 which is an exceptionally warm month. However, SURFEX seems to overestimate the screen temperature in very stable situations but the occurrence of these situations is marginal. Page 2 of 24

SURFEX@RMI 1. Introduction The ALADIN Be model usually has difficulties to produces a good forecast near the surface in strongly thermally stratified conditions. In a previous study Termonia (2001) calculated the scores for the temperatures during winter 1999 2000. He found that the errors occur during cold winter nights and they are ranging to even an unacceptable 8 C underestimation. However, such scores increasingly play a decisive role in determining the economical value of weather forecasts in social and commercial applications. Post processing techniques, such as model output statistics (MOS), have been developed to improve the meteorological operational forecasts. It consists in correcting the model output based on multiple linear regression by using model forecast variables and prior observations as predictors (Glahn and Lowry 1972). However, MOS techniques presents some problems: (1) the need for a long training datasets, (2) the regression equation is fixed and it does not take into account the fact that the relation between the predictands and predictors changes with the meteorological situations, (3) the regression equation of MOS is not suitable in cases where the model is updated (e.g. given increased resolution or a new parameterization). As a consequence, it needs new historical data in order to develop a new equation. Moreover, this MOS technique has been recently investigated from a dynamical point of view (Vannitsem and Nicolis 2008) in order to evaluate its ability in correcting initial condition and/or model errors. The analysis has revealed that MOS technique corrects systematic initial errors and only a part of model errors. A second conclusion is that the model error correction will strongly depend on the choice of predictors, the better choice being the model observables that strongly correlate with the model error source (Termonia and Deckmyn 2007). Therefore, it is very crucial to understand the source of model errors. The basic cause of the cool bias in ALADIN is that the model physics yields too little near surface vertical turbulent mixing during calm nighttime conditions (i.e. stable nighttime low level temperature inversions, referred to as the stable boundary layer). This problem is greater in the cool season because of the longer nights, the greater tendency for cool season nighttime winds to go calm, and the cooling effect of snow cover yielding even stronger nighttime temperature inversions. Moreover, the nighttime situation has a positive feedback character, because as the low level inversion sets in, the surface vertical turbulent mixing of heat falls off, which in turn acts to strengthen the inversion, etc. Ironically, as the vertical resolution of a model increases, there will be a tendency for the above positive feedback to be enhanced, because the greatest resolution is able to better resolve stronger, more extreme, shallow nighttime inversions. Page 3 of 24

The screen temperature is diagnosticated in the ALADIN model by complex interpolation between the lowest ALADIN level (~17 m) and the surface, making use of the stability functions of the dry static energy and applying the Monin Obukhov similarity theory for the surface boundary layer. However, Best and Hopwood (2001) found that the choice of stability functions at night can have significant impact on both the surface temperature and the sensible heat flux and therefore on the diagnostic of screen temperature in stable situations. It is clear from their results that using the Monin Obukhov similarity theory with the log linear stability functions, as it is done in ALADIN with the Geleyn (1988) interpolation scheme, gives poor agreement with the observations. In fact, using Monin Obukhov similarity theory with log linear stability functions cuts off the flux of heat with increasing stability too quickly, compared to the observations (Best and Hopwood 2001). This leads to incorrect lower surface temperatures as the warmer atmospheric air is no longer mixed down to the surface. In this study we use a state of the art surface only model (SURFEX) offline in order to remove the positive feedbacks between the surface and the atmosphere in stable situations. This model uses the ALADIN's first level forecasted variables as forcing for the surface exchange parameterization. In contrast to the ALADIN interpolation scheme based on log linear stability functions, in SURFEX the diagnostics of screen temperature is done using the formulation of Paulson (1970). The purpose of this note is to to compare the offline diagnostic of screen temperature by SURFEX with the operational ALADIN forecast during the cool season (2007 2008). 2. Data and Model a. ECOCLIMAP The ECOCLIMAP dataset have been designed to provide a complete set of high resolution (1 km) surface parameters for land surface models, including albedo, minimum stomatal resistance, roughness lengths and fraction of soil cover. In addition, the soil texture, merely based on the percentage of clay and sand, is taken from the FAO data (at 10 km resolution, FAO 1988). The dataset is based on a combination of three products: climate maps, land cover map, and one year of AVHRR NDVI data. A classification process per continent is used to assign a homogeneous ecosystem type (214 in total) to each 1 km pixel. All the pixels belonging to a given type have a common land cover. Further details are given in Masson et al. (2003). Within this study, the 1 km ecosystem map is aggregated to the resolution of the operational ALADIN forecast model (7 km). Page 4 of 24

SURFEX@RMI b. SURFEX We use the newly developed surface scheme of Météo France SURFEX (SURFace EXternalisée) (Martin et al. 2007). SURFEX is an externalized surface scheme that can be run either in a coupled mode in which case the atmospheric forcing is provided by the host atmospheric model (e.g. the AROME forecast model), or in a stand alone mode where the atmospheric drivers are derived from observations and fed to the surface scheme. The latter case is possible by relying on the algorithmic structure proposed by Best et al. (2004). SURFEX contains various modules allowing to describe the exchanges of water, momentum, and energy on 4 tiles of surface: sea, lake, vegetation, and the city, a grid value is then simply an area averaged value of the different tiles present in the grid cell. Over vegetated areas, SURFEX includes the Noilhan and Planton (1989) Interactions between Soil, Biosphere, and Atmosphere (ISBA) scheme. ISBA solves simultaneously the energy and water budget of the soil and vegetation. Water budget is forced by precipitation (rain and snow), and takes into account evaporation of the soil, transpiration from the vegetation, interception and evaporation of water on the leaves, runoff and drainage. The energy budget is forced by incoming radiation (both solar and infra red), and computes outgoing radiation (reflected solar and emitted/reflected thermal infra red radiation), heat flux towards/from the ground, and turbulent fluxes (sensible heat and latent heat from the water vapor flux). Vegetation parameters come from the ECOCLIMAP database. Over urban surfaces, SURFEX includes the Town Energy Balance (TEB, Masson 2000) single layer urban canopy module. Urban canopy is assumed to be an isotropic array of street canyons. The advantage over more comprehensive urban surface schemes, which include parameterizations for the canyon orientation and heterogeneous buildings morphology (see Masson 2006 for a review) is that relatively few individual surface energy balance evaluations need to be resolved, radiation interactions are simplified, and therefore computational time is kept low. TEB simulates heat and water exchanges and climate of three generic surfaces (roof, wall, and road), where heat transfers are computed through several layers of materials, generally four. Anthropogenic heat and vapor releases from buildings, vehicles and chimneys can also be added. This urban module is essentially designed to provide canopy heat fluxes for the lower boundary condition of atmospheric models. TEB is forced with literature based surface thermal parameters and observed or simulated atmospheric and radiation data from above roof level. Despite the simplification hypotheses, offline simulations of TEB have been shown to accurately reproduce surface energy balance, canyon air temperature, and surface temperatures observed in dense urban areas: Vancouver Page 5 of 24

and Mexico City (Masson et al. 2002), Marseille (Lemonsu et al. 2004), Basel (Hamdi and Masson 2008) during dry and hot seasons. In the study by Pigeon et al. (2008), the evaluation of TEB is extended to two other seasons, fall and winter, using measurements conducted over the dense urban area of Toulouse (France) instrumented from February 2004 to March 2005. In this study, a validation of the parameterization of anthropogenic heat sources against an inventory of energy consumption was also performed. Within the ECOCLIMAP database, the surface parameters are prescribed for the city centers, the suburban areas, and the commercial or industrial areas, and even for some remaining urban covers. 3. Modeling strategy Fig. 1. The offline version of SURFEX coupled to the operational ALADIN Be. Page 6 of 24

SURFEX@RMI Figure 1 illustrates how we run the offline version of SURFEX coupled to the operational ALADIN forecast model over Belgium. We selected a small domain, (259 km x 259 km) which cover Belgium. Then, we overlaid that domain over a 1 km resolution land cover classification provided by the ECOCLIMAP database. The land cover types contained in this domain are then aggregated at 7 km resolution, which is the resolution of the operational ALADIN forecast model, into 4 tiles (Sea, Lake, Vegetation, and Urban) with the corresponding fractional coverage. The soil texture and the topography are taken from the FAO and GTOPO30 databases respectively: 1. FAO. http://www.fao.org/ag/agl/agll/dsmw.htm 2. GTOPO30. http://edc.usgs.gov/product/elevation/gtopo30/gtopo30.html The forcing parameters necessary to run SURFEX are derived from the ALADIN's first level variable (~17 m Above Ground Level). The atmospheric data consist of 1 hourly: air temperature, specific humidity, atmospheric pressure, incoming global radiation, incoming long wave radiation, precipitation rate and wind speed. These data are then temporally interpolated to get data with the time resolution of the integration scheme of SURFEX (300 s). The initialization of the different prognostic variables is done using the initial values from the ALADIN forecast. For the sea and lake tiles a constant surface temperature was prescribed during the run. The aim of this work is to compare the offline diagnostic of SURFEX with the operational ALADIN forecast during the cool season. 4. Validation methodology To evaluate SURFEX performance, a comparison is made between the diagnostic of screen temperature by ALADIN [using the Geleyn (1988) interpolation scheme] and SURFEX [based on formulas by Paulson (1970)] on one hand, and the routine observations from the Belgian synoptic measurement net on the other hand. The period of validation extend from October 2007 to February 2008 (the cool season) and all model run were performed based on analyzes at 0000UTC. For this validation study a single station approach is followed (Taylor and Leslie 2005). Termonia (2001) showed that four synoptic stations are sufficient to capture more than 90% of the variance in the 2 m temperatures of the Belgian synoptic measurement net. So a small database was created containing the observations collected in the synoptic station of the RMI (Uccle, WMO number 6447) and eight other stations covering Belgium: Zelzat (WMO number 6431), Chievres (WMO number 6432), Deurne (WMO number 6450), Florennes (WMO number 6456), Saint Hubert (WMO number 6476), Bierset (WMO number 6478), Kleine Brogel (WMO number 6479), and Elsenborn (WMO number 6496). The statistical measures used here are based on formulas by Willmott (1982), who Page 7 of 24

describes various methods of quantifying the statistical relationships between an observed (O) and model predicted (P) quantity. These statistics are computed for 2 m temperature and relative humidity, and 10 m wind speed: 1. The Mean Error (ME) or Bias is the difference between the mean of the model predicted variable and the mean of the observed variable O: N ME= N 1 Pi Oi (1) i =1 2. The Index of Agreement (IA) can be used to assess relative model performance. The IA is defined as: N Pi Oi 2 i=1 IA=1 N (2) P ' i O 'i 2 i =1 Where P'i= Pi O and O'i=Oi O. The IA is bounded between 0 and 1 such that the perfect simulation has IA=1. 3. The RMSE is computed from: N 0.5 RMSE=[N 1 Pi Oi 2 ] (3) i =1 it is always positive, and it emphasizes extreme differences between P and O. 4. The Systematic and Unsystematic RMSE (RMSES and RMSEU) are used to quantify the type of error. RMSES and RMSEU are calculated by accounting for the slope and intercept of the regression line that compare observed and predicted values and using: N RMSES=[ N 1 P i O i 2 ] 0.5 i=1 N 0.5 1 2 and RMSEU =[N Pi P i ] i=1 Where the linear estimator P i is defined by: P i =a boi (4) (5) Such that a is the intercept and b is the slope of the least squares regression. RMSES and RMSEU form the following relation: RMSE 2=RMSES 2 RMSEU 2 (6) Thus the proportion of systematic and unsystematic error in the model can be defined from: SYS = RMSES 2 RMSE 2 and UNSYS= RMSEU 2 RMSE 2 (7) In theory systematic errors should account for processes that the model does not Page 8 of 24

SURFEX@RMI routinely simulate well, whereas unsystematic errors could be attributed to randomness or subgrid scale processes. A ''good'' model will have a systematic error that approaches 0 while the unsystematic error approaches the mean square error. Therefore, better models should have a smaller systematic portion of the error (i.e., bias). 5. Results and Discussions Figure 2 6 show the BIAS, RMSE, Index of Agreement, and the SYS error of the diagnostic of 2 m temperature by ALADIN and SURFEX at the Uccle station. Each figure presents the average over one month of the cool season with the forecast range up to +48h. Fig. 2 6 presents also scatter plot which is a very simple qualitative way to get an estimate of the reliability of the diagnostics. Ideal forecasts will be located on the diagonal of the plot (the linear regression equation is also presented). For the eight other synoptic stations of the RMI, an average over each month and over the 48h forecast is presented in Appendix A1 5. From the scatter plots, ALADIN seems to underestimate the low temperatures while SURFEX diagnostics fit better to the observations. Moreover, upon considering the graphs for the BIAS, RMSE, and SYS errors one notices cyclic dependence of the ALADIN errors on the forecast range with large systematic error during the night. This indicates that the ALADIN errors occurs during cold winter night which is in agreement with the findings in the note of Termonia (2001). This underestimation of the winter minimum screen temperature is largely reduced by SURFEX except for January 2008 which is an exceptionally warm month. The observed average temperature of this month [6.5 C (normal= 2,6 C)] is one of the highest since 1833, the beginning of the observations at Uccle. Page 9 of 24

Fig. 2. The scatter plots, BIAS, RMSE, Index of agreement, and the SYS error of the diagnostic of 2 m temperature by ALADIN and SURFEX at the Uccle station during October 2007. Page 10 of 24

SURFEX@RMI Fig. 3. Scores for November 2007. Page 11 of 24

Fig. 4. Scores for December 2007. Page 12 of 24

SURFEX@RMI Fig. 5. Scores for January 2008. Page 13 of 24

Fig. 6. Scores for February 2008. Page 14 of 24

SURFEX@RMI Figure 7 shows the errors in the diagnostics of nocturnal screen temperatures by ALADIN and SURFEX for the cool season (Oct Feb). The bulk Richardson number obtained from SURFEX is used as a measure of stability, with the specific form RiB 1/2 being employed, since this is proportional to the wind speed. Ri B = g z z 0 s U2 T0 (8) where U is the wind speed, z is the height, z 0 is the roughness length, g is the acceleration of gravity, / S is potential temperature of the ALADIN's first level/surface, and T0 is a representative temperature in the surface layer. Fig. 7. Scatter plot of the errors in the diagnostics of nocturnal screen temperatures by ALADIN and SURFEX against the inverse square root of the bulk Richardson number for the cool season (Oct Feb). The results show that the differences between ALADIN and SURFEX diagnostics are large when there is no or little wind speed which correspond to very stable conditions. In fact, while ALADIN underestimates drastically the 2 m temperature with even unacceptable values ( 8 C), SURFEX seems to overestimate the 2 m temperature in very stable situations but the occurrence of Page 15 of 24

these situations is marginal as can be seen from the error distribution (see Fig. 8). For stable situations and weak wind the underestimation of the winter minimum screen temperature is largely reduced by SURFEX. When the wind speed becomes stronger the difference between ALADIN and SURFEX is negligible. Fig. 8. Error distribution of the nocturnal screen temperature diagnosticated by ALADIN and SURFEX for the cool season (Oct Feb). 6. Conclusion In this study a state of the art surface only model (SURFEX) is used offline in order to remove the positive feedbacks between the surface and the atmosphere in stable situations. The coupling occurs at ALADIN's lowest vertical level. Variables that must be exchanged are forecasted: air temperature, specific humidity, atmospheric pressure, incoming global radiation, incoming long wave radiation, precipitation rate and wind speed. The purpose of this note is to compare the offline diagnostic of screen temperature by SURFEX with the operational ALADIN forecast during the cool season (2007 2008). Results show a cyclic dependence of the ALADIN errors on the forecast range with large systematic error during the night. This indicates that the ALADIN errors occurs during cold winter night which is in agreement with the findings in the note of Termonia (2001). This underestimation of the winter minimum screen temperature is largely reduced by SURFEX except for January 2008 which is an Page 16 of 24

SURFEX@RMI exceptionally warm month. However, SURFEX seems to overestimate the screen temperature in very stable situations but the occurrence of these situations is marginal. To pursue this work, we plan to couple SURFEX to a surface boundary layer (SBL) scheme following the methodology described in Masson and Seity (2008). With this new scheme, several prognostic air layers are added from the ground up to the forcing level and the surface boundary layer is thus, resolved prognostically, taking into account large scale forcing, turbulence and, if any, drag and canopy forces. The screen temperature is calculated prognostically at 2 m level since with the SBL scheme there is no need of analytical extrapolation such as Paulson (1970) or Geleyn (1988). SBL scheme has been validated coupled to AROME on an area in south east of France for two months: one in winter and one in summer. The systematic comparison to more than 350 meteorological stations showed that SBL improves significantly the bias of the model on temperature and humidity especially in nighttime stable conditions. ACKNOWLEDGMENTS. The author is very grateful to the three reviewers: P. Termonia, A. Deckmyn, and J. Nemeghaire. Page 17 of 24

Appendix A. 1. Scores averaged over 48h forecast range for October 2007. Page 18 of 24

SURFEX@RMI A. 2. Scores averaged over 48h forecast range for November 2007. Page 19 of 24

A. 3. Scores averaged over 48h forecast range for December 2007. Page 20 of 24

SURFEX@RMI A. 4. Scores averaged over 48h forecast range for January 2008. Page 21 of 24

A. 5. Scores averaged over 48h forecast range for February 2008. Page 22 of 24

SURFEX@RMI References Best, M. J., A. Beljaars, J. Polcher, and P. Viterbo, 2004: A Proposed Structure for Coupling Tiled Surfaces with the Planetary Boundary Layer. J. of Hydromet., 5, 1271 1278. Best, M. J., and W. P. Hopwood, 2001: Modelling the local surface exchange over a grass field site under stable conditions. Q. J. R. Meteorol. Soc., 127, 2033 2052. Geleyn, J. F., 1988: Interpolation of wind, temperature and humidity values from the model levels to the height of measurement. Tellus, 40, 347 351. Glahn, H. R., and D. L. Lowry, 1972: The use of model output statistics (MOS) in objective weather forecasting. J. Appl. Meteor., 11, 1203 1211. Hamdi, R., and V. Masson, 2008: Inclusion of a drag approach in the Town Energy Balance (TEB) scheme: offline 1 D evaluation in a street canyon. J. Appl. Meteor. Clim., 47, 2627 2644. Lemonsu, A., C. S. B. Grimmond, and V. Masson, 2004: Modeling the surface energy balance of an old Mediterranean city core. J. Appl. Meteor., 43, 312 327. Martin, E., P. Le Moigne, V. Masson, and coauthors, 2007: Le code de surface externalisé SURFEX de Météo France. Atelier de Modélisation de l'atmosphère (http://www.cnrm.meteo.fr/ama2007/), Toulouse, 16 18 January. Masson, V., 2000: A physically based scheme for the urban energy budget in atmospheric models. Bound. Layer Meteor., 94, 357 397. Masson, V., C. S. B. Grimmond, and T. R. Oke, 2002: Evaluation of the Town Energy Balance (TEB) scheme with direct measurements from dry districts in two cities. J. Appl. Meteor., 41, 1011 1026. Masson, V., J. L. Champeaux, F. Chauvin, C. Meriguet, and R. Lacaze, 2003: A global database of land surface parameters at 1km resolution in meteorological and climate models. J. Climate, 16, 1261 1282. Masson, V., 2006: Urban surface modeling and the meso scale impact of cities. Theor. Appl. Climatol., 84, 35 45. Masson, V., and Y. Seity, 2009: Including atmospheric layers in vegetation and urban offline surface schemes. J. Appl. Meteor. Clim., (In press). Noilhan, J., and S. Planton, 1989: A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 117, 536 549. Paulson, C. A., 1970: The mathematical representation of wind and temperature profiles in the unstable atmospheric surface layer. J. Appl. Meteor., 9, 857 861. Pigeon, G., M. A. Moscicki, J. A. Voogt, V. Masson, 2008: Simulation of fall and winter surface energy balance over a dense urban area using the TEB scheme. Page 23 of 24

Meteorol. Atmos. Phys., In press, doi: 10.1007/s00703 008 0320 9. Taylor, A. A., and L. M. Leslie, 2005: A single station approach to model output statistic temperature forecast error assessment. Wea. Forecasting, 20, 1006 1020. Termonia, P., 2001: An overview of the verification of the operational ALADIN forecasts during the year 2000 at the RMIB. Publication scientifique et technique, N 016, Royal Meteorological Institute of Belgium, Brussels, Belgium. Termonia, P., and A. Deckmyn, 2007: Model Inspired predictors for model output statistics (MOS). Mon. Wea. Rev., 135, 3496 3505. Vannitsem, S., and C. Nicolis, 2008: Dynamical properties of model output statistics forecasts. Mon. Wea. Rev., 136, 405 419. Wilmott, C. T., 1982: Some comments on the evaluation of model performance. Bull. Amer. Meteor. Soc., 63, 1309 1313. Page 24 of 24