A bare ground evaporation revision in the ECMWF land-surface scheme: evaluation of its impact using ground soil moisture and satellite microwave data

In situ soil moisture data from 122 stations across the United States are used to evaluate the impact of a new bare ground evaporation formulation at ECMWF. In November 2010, the bare ground evaporation used in ECMWF's operational Integrated Forecasting System (IFS) was enhanced by adopting a lower stress threshold than for the vegetation, allowing a higher evaporation. It results in more realistic soil moisture values when compared to in situ data, particularly over dry areas. Use was made of the operational IFS and offline experiments for the evaluation. The latter are based on a fixed version of the IFS and make it possible to assess the impact of a single modification, while the operational analysis is based on a continuous effort to improve the analysis and modelling systems, resulting in frequent updates (a few times a year). Considering the field sites with a fraction of bare ground greater than 0.2, the root mean square difference (RMSD) of soil moisture is shown to decrease from 0.118 m 3 m −3 to 0.087 m 3 m −3 when using the new formulation in offline experiments, and from 0.110 m 3 m −3 to 0.088 m 3 m −3 in operations. It also improves correlations. Additionally, the impact of the new formulation on the terrestrial microwave emission at a global scale is investigated. Realistic and dynamically consistent fields of brightness temperature as a function of the land surface conditions are required for the assimilation of the SMOS data. Brightness temperature simulated from surface fields from two offline experiments with the Community Microwave Emission Modelling (CMEM) platform present monthly mean differences up to 7 K. Offline experiments with the new formulation present drier soil moisture, hence simulated brightness temperature with its surface fields are larger. They are also closer to SMOS remotely sensed brightness temperature.


Introduction
Soil moisture plays a fundamental role in the partitioning of mass and energy fluxes between the hydrosphere, biosphere and atmosphere because it controls both evaporation and transpiration fluxes from bare soil and vegetated areas, respectively. In addition, it is a key variable in hydrological processes ( i.e. runoff, evaporation from bare soil and transpiration from the vegetation cover) and has an impact on plant growth and carbon fluxes (Dirmeyer et al., 1999;Entekhabi et al, 1999). Its initialisation is of crucial importance for Numerical Weather Prediction (NWP) models and this topic has been extensively explored in the literature (e.g., Shukla and Mintz, 1982;Dirmeyer, 2002;Douville et al., 2001); soil moisture might play a role in meteorological forecasting (Seneviratne et al., 2010).
The second phase of the multi-institutional numerical modelling experiment GLACE-2 (Global Land-Atmosphere Coupling Experiment) led to several insights about how the realistic initialisation of soil moisture can have a significant impact on the skill of precipitation and air temperature forecast skill at the sub-seasonal scale (Koster et al., 2011). Notably, while both wet and dry land-surface-model initialisation are likely to generate skill in different areas of the world, dry initialisation provides more skill at the transition between soil-moisture and energy-availability-controlled evaporation. The first phase of GLACE (Koster et al., 2004) focused on the atmospheric response to soil moisture variations. Meteorological variables such as precipitation and air temperature, were found to be particularly affected by soil moisture variations in specific areas: the transition zones between arid and humid areas. Dry-land has recently received more attention both in literature (Wang et al 2012) and in several initiatives on the Drought Early Warning (DEW) information services world-wide (WMO, Drought monitoring and early warning, 2006). Many studies have focused on the coupled bare soil-canopy processes and the ability of land surface models to simulate bare soil processes has also been of interest (Desborough et al., 1996). Albergel et al, (2012a, b) demonstrated the good quality of the European Centre for Medium-Range Weather Forecasts (ECMWF) soil moisture products with respect to global-ground based in situ observations. They found good level of correlations despite the high values of root mean square difference (RMSD) which indicate that ECMWF products tend to overestimate soil moisture, particularly over dry areas. To overcome this problem an improved bare ground evaporation scheme over dry land  was implemented in ECMWF's Integrated Forecasting System (IFS), in November 2010. It is expected that the new scheme will reduce the soil moisture over bare soil by enhancing evaporation, resulting in more realistic soil moisture when compared to in situ data. Because the improved bare ground evaporation was implemented in 2010 along with other modifications affecting soil moisture (e.g., an Extended Kalman Filter for soil moisture analysis, de  it is difficult to isolate the impact of the new evaporation scheme. For this reason, offline experiments were carried out with and without the new bare ground evaporation to analyse the impact of the specific modification. This study aims at evaluating the impact of this new evaporation scheme on soil moisture. In the framework of the Soil Moisture and Ocean Salinity (SMOS, Kerr, 2007;Kerr et al., 2010) mission, ECMWF is implementing the direct assimilation of near real time brightness temperature (TB) in the L-band (Sabater et al., 2012). It will only be effective if realistic and dynamically consistent fields of TB are simulated as a function of land-surface conditions. At ECMWF the Community Microwave Emission Modelling platform (CMEM, Holmes et al., 2008;Drusch et al., 2009a, de Rosnay et al., 2009) is used to simulate TB. Surface soil moisture is, amongst other surface fields, coupled with CMEM to produce ECMWF's first-guess TB. The main objective of this study is to evaluate the new evaporation formulation on soil moisture, however as the improved bare ground evaporation is expected to affect surface fields such as soil moisture, it also assesses its impact on simulated TB.
After a description of the ECMWF's analysis and soil moisture products used in this study, the new bare ground evaporation formulation is presented followed by a description of the in situ observations required to evaluate soil moisture analyses. Next the CMEM platform is briefly described along with the remotely-sensed SMOS TB data set. Then the impact of the improved bare ground evaporation is assessed using (i) two offline experiments and (ii) ECMWF's operational IFS and in situ soil moisture data for the period 2010-2011 (only 2010 for the offline experiments). Finally, surface soil moisture and soil temperature fields, as well as snow depth and density fields from the two offline experiments, are coupled with the CMEM platform to simulate the TB data set at a global scale; they permit the study of the sensitivity of CMEM to the new bare ground evaporation. Finally the two TB data sets are compared to SMOS TB observations for the year 2010.

Material and methods
In situ soil moisture observations are important for evaluating soil moisture products. In this study use was made of soil moisture data from the NCRS-SCAN network (Natural Resources Conservation Service -Soil Climate Analysis Network) in the United States (Schaefer and Paetzold, 2000). They were obtained through the International Soil Moisture Network (ISMN, Dorigo et al., 2011, http://www.ipf.tuwien.ac.at/insitu/), a new data hosting centre where globally available ground based soil moisture measurements are collected, harmonized and made available to users. Data at 148 stations in 2010 and 2011 were used to evaluate soil moisture from offline experiments and the operational IFS. The soil moisture data sets used in this study are presented in Table 1.

ECMWF's land surface analysis
Data produced at ECMWF include a large variety of surface parameters that describe the atmosphere as well as ocean-wave and land-surface conditions (more information at: http://www.ecmwf.int/products/). The core atmospheric assimilation system at ECMWF relies on the four-dimensional variational (4D-Var) data assimilation scheme described in Rabier et al. (2000) and Mahfouf and Rabier (2000). It has an observation time window of 12 hours (Bouttier, 2001). Data provided by satellite sensors (from microwave and infrared radiometers) as well as conventional observations (e.g., radiosonde network) are ingested within the 4D-Var. Use is also made of surface observations such as surface pressure, humidity and wind.
For several decades NWP initialisation has relied on data assimilation approaches which use satellite data to analyse atmospheric variables. Land-surface initialisation is generally independent from the atmospheric system and is based on ground measurements of screen-level variables as a proxy for soil moisture. In recent years, major upgrades have been implemented in the land-surface modelling and analysis systems of the IFS with respect to soil moisture: (i) an improved soil hydrology model , (ii) a new snow scheme (Dutra et al., 2010) and (iii) a multi-year satellite-based vegetation climatology (Boussetta et al., 2010) together with enhanced bare ground evaporation . Also a new soil moisture analysis scheme, based on a point-wise Extended Kalman Filter (EKF) for the global land surface, has been developed and this was implemented in the IFS (Drusch et al., 2009b;de Rosnay et al., 2011de Rosnay et al., , 2012 in November 2010. The model forecast for the land surface analysis is provided by the TESSEL land surface scheme (Van den Hurk et al., 2000) which was then upgraded to H-TESSEL (Van den Hurk and Viterbo, 2003;Balsamo et al., 2009) with an improved soil hydrology. H-TESSEL development was a response to weaknesses in the TESSEL hydrology; a Hortonian runoff scheme hardly producing surface runoff and the choice of a single global soil texture was not able to characterize different soil moisture regimes. So, for HTESSEL the formulation of the soil hydrological conductivity and diffusivity was revised to be spatially variable according to a global soil texture map (FAO/UNESCO Digital Soil Map of the World, DSMW, FAO, 2003). H-TESSEL was implemented in the IFS in November 2007. It was verified in various ways including field site comparison, data assimilation and modelling experiments by Balsamo et al. (2009). Also Albergel et al. (2012b) provide a detailed evaluation of HTESSEL soil moisture. Analyses are available at four depths (0-7, 7-28, 28-100 and 100-289 cm). The soil heat budget follows a Fourier diffusion law, modified to take into account soil water freezing/melting according to Viterbo et al. (1999). The energy equation is solved with a net ground heat flux as the top boundary condition and a zero flux at the bottom. The water balance at the surface (i.e. the change in water storage of the soil moisture, interception reservoir and accumulated snowpack) is computed as the difference between the precipitation and (i) the evaporation of soil, vegetation, interception water, (ii) surface and subsurface runoff. First precipitation is collected in the interception reservoir until it is saturated. Then, excess precipitation is partitioned between surface runoff and infiltration into the soil column. HTESSEL's formulation of the soil hydrological conductivity and diffusivity is spatially variable according to a global soil texture map (FAO/UNESCO Digital Soil Map of the World, DSMW, FAO, 2003). Surface runoff is based on variable infiltration capacity. At the end of each data assimilation cycles an adjustment to the model forecast (e.g. soil moisture) is produced, it usually referred to analysis increment and represents the net response of the variational data assimilation to all observations used.
Three analysis schemes for the surface (and near-surface) variables are currently used in operations. They are based on: spatial Optimal Interpolation (2D-OI, for snow depth and screen-level analyses), column Optimal Interpolation (1D-OI, for soil and snow temperature analysis), and an EKF (for soil moisture analysis, Drusch et al., 2009b;de Rosnay et al., 2011de Rosnay et al., , 2012. Analysis of surface parameters is decoupled from the main atmospheric analysis. Firstly an OI scheme produces estimates of screenlevel temperature and relative humidity by combining synoptic observations over land with background estimates (short-range forecasts) from the most recent analysis . Analysed fields of screen level temperature and relative humidity are then used to update estimates of soil moisture (and soil temperature) for the layers of the model using the EKF analysis. While producing the forecast, the model estimates a wide variety of physical variables including precipitation. Even if not directly observed, the model estimates are constrained by the observations (in situ measurements of temperature and humidity) used to initialise the forecast and their accuracy relies on the quality of the model physics as well as that of the analysis.

Soil moisture products
In this section a description is given of the major differences between the deterministic operational suite and the offline experiments with respect to soil moisture.
The version of IFS used in operations at ECMWF from January 2010 to December 2011 spans from 35r3 to 37r3 (more information at: http://www.ecmwf.int/research/ifsdocs/). There are continuous efforts to improve the analysis and modelling schemes (i.e. changes in spatial and vertical resolutions, data assimilation, parameterizations and sources of data), resulting in frequent updates of the IFS (few times a year). Before the implementation of cycle 36r4 in November 2010, the assimilation technique used was the OI (Mahfouf, 1991;Mahfouf et al., 2000b). The EKF was implemented in operations to optimally combine model data with conventional observations and satellite measurements. In its current configuration, the EKF soil moisture analysis uses meteorological observations of screen-level parameters close to the surface, as with the previous OI method. However due to the flexibility of Kalman-based techniques, the EKF can handle different sources of observations (Mahfouf et al., 2009) and offers a wide range of development possibilities including the use of remotely-sensed data such as ASCAT (Advanced Scatterometer, Wagner et al, 2007) and SMOS (Kerr, 2007;Kerr et al., 2010).The operational IFS soil moisture analysis is produced four times each day (i.e. at 00:00, 06:00, 12:00 and 18:00 UTC); it has a spatial resolution of about 25 km (T799) until 26 January 2010 and then it was about 16 km (T1279). Analyses at 00:00 UTC are considered in this study.
The offline experiments used in this study are based on IFS cycle 36r4. They are produced daily at 00:00, 06:00, 12:00 and 18:00 UTC at a spatial resolution of about 80 km (T255). Offline experiments are a response to reproduce the land-surface model state in between two reanalyses (e.g., the two latest reanalyses of ECMWF are ERA-Interim, Dee et al., 2011 andERA-40, Uppala et al., 2005). Reanalysis such as ERA-Interim are produced by a fixed version of the IFS (for the main component of the atmospheric model and data assimilation) and have the advantage of being consistent over a long period. Offline experiments are based on the same principle and take into account specific improvements implemented in the operational IFS (e.g., the ERA-Interim reanalysis land-surface scheme is based on the TESSEL scheme while HTESSEL is used for offline experiments in this study). Offline experiments can be considered as add-on before future generations of reanalysis are produced at ECMWF. They are driven by ERA-Interim (Dee et al., 2011) global atmospheric reanalysis.The difference between the two experiments concerns only the bare ground evaporation. For one, the bare ground evaporation over dry land has been enhanced by adopting a lower stress threshold than for the vegetation, allowing a higher evaporation (BEVAP_NEW). Its control experiment without the new bare ground evaporation is called BEVAP_OLD.

New bare ground evaporation
This section gives a description of the new bare ground formulation. It is in agreement with the experimental findings of Mahfouf and Noilhan (1991) and results in more realistic soil moisture values for dry land . Indeed the evaporation from non-vegetated areas responds to a different physical mechanism compared to densely vegetated areas. Over bare soil the vaporisation of water in the soil pores takes place in a thin layer close to the surface-atmosphere interface as a direct effect of incoming solar radiation providing the latent heat requirements. Atmospheric conditions such as air temperature, humidity, wind velocity and radiation, as well as soil conditions (e.g. water content and roughness length) play a role in modulating the evaporation processes (Hillel, 1980). The relationship between soil moisture and bare soil evaporation is generally parameterized in land-surface models. Mahfouf and Noilhan (1991) compared several of these formulations; in these studies, bare soil evaporation formulations was halted when the soil is completely dry (for soil moisture close to zero). In the previous TESSEL scheme linking of soil moisture and evaporation was assumed to be linear between the permanent wilting point and the field capacity values for soil moisture. With the introduction of a tiling approach, the same stress function was applied to both vegetated and nonvegetated tiles, neglecting the fact that wilting point is a soil moisture threshold that applies uniquely to vegetated areas. The formulation of the bare soil evaporation has been revisited in the latest HTESSEL version to allow a smooth transition between vegetated and non-vegetated areas and to realign the formulation of bare ground evaporation with studies in the literature.
In BEVAP_OLD experiment, Eq.3 is used while it is Eq.5 in BEVAP_NEW.

In situ soil moisture observations: the NCRS-SCAN network
The SCAN network (http://www.wcc.nrcs.usda.gov/scan/) is a comprehensive, nationwide soil moisture and climate information system designed to provide data to support natural resource assessments and conservation activities. It is administered by the United States Department of Agriculture (USDA) Natural Resources Conservation Service (NRCS) through the National Water and Climate Centre (NWCC), in cooperation with the NRCS National Soil Survey Center. The system focuses on agricultural areas of the U.S.A. The observing network monitors soil temperature and soil moisture at several depths, soil water level, air temperature, relative humidity, solar radiation, wind, precipitation and barometric pressure amongst others. SCAN data are used for a variety of purpose ranging from global climate modelling to agricultural studies. Data are collected by a dielectric constant measuring device; typical measurements at 2 inches (about 5 cm) are used. The vegetation cover at those sites consists generally of natural fallow or short grass. In this study, all the stations of the NCRS-SCAN network providing data in 2010 and 2011 are retained (leading to a total of 148 stations). Stations from this network were already been used to validate SM products from ECMWF (Albergel et al., 2012a, b) . The location of the stations of the NCRS-SCAN network are shown on Figure 1.

Statistical Comparison between analysis and in situ observations
A usual step for evaluating soil moisture products from model is to determine whether their behavior matches the observations. Hence in situ measurements of soil moisture are a highly valuable source of information. For all stations, correlations (R, Eq.7), bias (in situ minus analysis), root mean square difference (RMSD, Eq.8) and p-value (a measure of the correlation significance) are calculated. The latter indicates the significance of the test; the 95% confidence interval is used in this study (as in Rüdiger et al., 2009;Albergel et al., 2009Albergel et al., , 2010; only configurations where the p-value is below 0.05 (i.e. the correlation is not a coincidence) are retained.
As in situ data may contain errors (instrumental and representativeness), they are not considered as 'true' soil moisture. This is emphasised by using the RMS difference terminology instead of RMS error. In situ observations of soil moisture at the NCRS-SCAN stations are associated with soil temperature measurements. The observations of soil moisture were flagged for temperature below 4 degrees Celsius to avoid frozen conditions. When considering TB, the standard deviation (STD) is also computed.
To avoid seasonal effects, monthly anomaly time-series are calculated. The difference from the mean is produced for a sliding window of five weeks (if there are at least five measurements in this period), and the difference is scaled to the standard deviation. For each surface soil moisture estimate at day (i), a period F is defined, with F=[i-17, i+17] (corresponding to a five-week window). If at least five measurements are available in this period, the average soil moisture value and the standard deviation are calculated. The Anomaly (Ano) is then given by: The same equation is used to compute in situ anomaly time-series which can be compared with that from ECMWF's analyses.

2.4
The terrestrial microwave emission modelling

The Community Microwave Emission Modelling (CMEM) platform
The CMEM platform has been developed by ECMWF as the forward operator for low frequency passive microwave TB (from 1GHz to 20 GHz) of the surface in the framework of the SMOS mission. CMEM is one of the ESA (European Space Agency) SMOS tools and it is available to the entire community through the ECMWF web pages: http://www.ecmwf.int/research/ESA_projects/SMOS/cmem/cmem_doc.html.

SMOS brightness temperature
SMOS consists in a microwave imaging radiometer with an aperture synthesis collecting top of atmosphere full polarized radiances coming from the scene viewed by its antennas through their power patterns. It is a Y-shaped instrument with several elementary antennas regularly spaced along the arms (69 in total) which provide, at each integrations step, a full image (circa 1000x1200 km) at either two polarisations or full polarisation, of the Earth's surface (Kerr et al., , 2010. The spatial resolution is about 40 km and the globe is fully imaged at least twice every three days (ascending and descending orbits). Any points at the surface are viewed frequently at different angles and polarisations. The angular information is used to separate the different contributions (soil-vegetation) to the signal (Wigneron et al., 2000). The signal measured at satellite level is a TB for the L-band consisting of four main contributions: (i) the up-welling atmospheric emission, (ii) the Earth's surface emission, attenuated by the atmosphere, (iii) the atmospheric down-welling atmospheric emission reflected at the surface and attenuated along the upward path by the atmosphere, and (iv) the cosmic background emission attenuated by the atmosphere, reflected at the surface and attenuated again along the upward path by the atmosphere. SMOS Near Real Time (NRT) products are processed at the European Space Astronomy Centre (ESAC) and sent to ECMWF via the SMOS Data Processing Ground Segment (DPGS) interface. ECMWF is involved in global monitoring and data assimilation of the SMOS mission data. The development of a data monitoring system for the SMOS near real time product provides a timely quality check for the European Space Agency (ESA) and the SMOS calibration and validation teams. More information and comparison between SMOS data and modelled TB are available through ECMWF web pages: http://www.ecmwf.int/research/ESA_projects/SMOS/index.html.

Impact of the new bare ground evaporation on soil moisture
3.1.1 Using Offline experiments Figure 2 illustrates the mean soil moisture for both BEVAP_OLD and BEVAP_NEW for August 2010. A simple look at Figures 1 and 2 show that areas with a high fraction of bare soil are drier with BEVAP_NEW than with BEVAP_OLD. The statistical scores for the comparison between either BEVAP_OLD or BEVAP_NEW and the stations from the NCRS-SCAN network are presented in Table 2. As indicated in NCRS-SCAN website, data are provisional and subject to revision, very little control is applied to measurements from NCRS-SCAN. Dharssi et al. (2011) used a simple process to identify stations where sensors might be dysfunctional. Stations are rejected based on the scores obtained when compared to their experiments (in term of correlations, RMSDs and biases). As this study aims to assess the impact of the new bare ground experiment, mostly in term of RMSD, a similar is applied based only on the correlation level. Stations for which either BEVAP_OLD or BEVAP_NEW have a correlation less than 0.3 are rejected (as in Dharssi et al., 2011).  This rather strict process has probably removed some good stations too (e.g. in areas where the model might not realistically represent soil moisture). Also, stations with a non-significant correlation are rejected (p-value < 0.05). This filtering of the NCRS-SCAN stations, results in 122 stations being available (out of 148) for the comparison of the two offline experiments. Table 3 Figure 4 illustrates the two offline runs as well as the in situ observations for one site located in Utah. Minimum values of BEVAP_OLD soil moisture are limited by the dominant wilting point for vegetation types, however ground data indicate much drier conditions, as is clearly observed from May to September 2010. The new bare ground evaporation allows the model to go below this wilting point so the BEVAP_NEW analysis is in much better agreement with the observations than that for BEVAP_OLD. Along with the decrease in RMSD, one may note an increase in the correlation (from 0.63 to 0.65). Also BEVAP_NEW has a more realistic decrease in soil moisture after a precipitation event due to its higher water holding capacity and this explains the slightly better correlations.
Considering the short-term variability, the average correlations for the monthly anomaly time series are 0.54 for BEVAP_OLD and 0.55 for BEVAP_NEW. Correlations of volumetric time series are larger than those for the monthly anomaly time-series. The good level of correlation of the volumetric time series is largely explained by seasonal variations, which are suppressed in monthly anomalies.

Using the operational product
The new bare ground evaporation formulation was implemented in operations in November 2010. Its impact on the operational analysis was assessed for the 2010-2011 period. Results are presented in Table 4 and illustrated by Figure 5. The same threshold of 0.2 for the fraction of bare ground was used. The correlations, bias and RMSD are 0.59, -0.076 m 3 m -3 and 0.110 m 3 m -3 for 2010 and 0.69, -0.010 m 3 m -3 , 0.088 m 3 m -3 for 2011. Figure 5 illustrates the soil moisture time-series for three stations with different fraction of bare ground (~0.46,~0.79 and ~0.15 from top to bottom). Compared to 2010, the station with a fraction of bare ground of 0.79 has a lower value of RMSD in 2011 than a station with a fraction of 0.15. As for BEVAP_OLD and BEVAP_NEW, the operational analysis is in much better agreement with the observations for 2011 (with the new bare ground evaporation) than for 2010; this is particularly clear for the period from May to September 2011 (see Figure 5). If the decrease in RMSD is associated with the new bare ground evaporation, the increase in correlation (from 0.59 to 0.69) is mainly due to the new EKF analysis (also, in situ data are different). As demonstrated in Albergel et al. (2010) the use of the EKF permits to increase the quality of the soil moisture product compare to the former OI method. The higher water holding capacity observed for 2011 also helps in this way (more realistic decrease of soil moisture after a precipitation event). Correlations of anomaly time-series are 0.53 and 0.54 for 2010 and 2011, respectively.

Impact of the new bare ground evaporation on terrestrial microwave emission and comparison with SMOS
The differences between the TB simulated using surface fields from BEVAP_NEW and the one from BEVAP_OLD are computed for each month of 2010, for both H (horizontal) and V (vertical) polarizations (referred to as TBH and TBV) and an incidence angle of 40º, for 06:00UTC and 18:00UTC. It is useful to study the sensitivity of the simulated TB to the bare soil parameterization as a better representation of soil moisture should lead to more realistic TB, a pre-requisite for SMOS data assimilation. In terms of mean difference, BEVAP_NEW soil moisture is drier than BEVAP_OLD, so simulated TB are larger with the BEVAP_NEW surface fields. For TBH (18:00 UTC) the global monthly mean bias between the two data set range from 4.72 K to 7.01 K, with an annual value of 6.2 K. For TBV (18:00 UTC), global monthly mean biases range from 2.94 K to 4.14 K, with an annual mean difference of 3.7 K. Statistical scores are summarized in Table 5 and Figure 6 provides a map of the differences between the simulated TB for one month (August 2010 at 06 UTC) and one polarization (H). For this month, mean differences are 6.87 K and 3.96 K, with STD of 15.58 K and 9.04 K, for TBH and TBV, respectively (at 06 UTC). Positive differences are found in relatively dry areas. A look at the North American continent shows that large differences are found in the western part of the United States, where there is a high fraction of bare ground (accordingly to Figure 1). Figure 7 shows, (i) the global monthly mean sensitivity between the two TB and (ii) the spatial correlations between each TB data sets and the fraction of bare ground. As expected, there is a slight annual cycle due to the larger distribution of the continental areas in the northern hemisphere. Spatial correlations between bare ground and TB mean sensitivity are on average 0.66 and 0.65 for TBH (06 UTC and 18 UTC), and 0.61 and 0.61 for TBV (06 UTC and 18 UTC).        Both TB data sets are compared to SMOS TB observations at 06 UTC, at an incidence angle of 40 degrees (more data available). Radio frequency interference (RFI) disturbs the natural microwave emission observed by SMOS (Zribi et al., 2011). At an acquisition time of 06 UTC SMOS data are mainly observed over Western Europe and Africa, areas known to be less affected by RFI than others (such as Central Europe for instance). Results are presented in Table 6, it shows that BEVAP_NEW TB are in better agreement with SMOS than BEVAP_OLD TB, with large differences however. The mean difference (for 2010) between SMOS and BEVAP_NEW TBH is 10.2 K (STD of 21 K) and 14.5 K (STD of 20.8 K) when considering BEVAP_OLD TBH. Values are higher for TBV than for TBH.

Discussion
While previous studies (Albergel et al., 2012a, b) have demonstrated the good ability of ECMWF analyses to represent the soil moisture annual cycle as well as its short-term variability, they have also shown an overestimation of soil moisture. The modification of the soil moisture inhibition function (Eq.5) in the new bare ground evaporation formulation allows a much lower level of soil moisture to be reached over bare soil areas as a consequence of direct bare ground evaporation under strong insulation. Results are more realistic when compared to in situ soil moisture values observed over dry areas; they have a smaller RMSD but also a slightly better correlation. The larger water holding capacity induced by the new bare ground evaporation provides a more realistic decrease in soil moisture after a precipitation event. This explains the slightly better correlations obtained with the new formulation for the offline experiments for both the volumetric and the monthly anomaly time-series. Indeed, the latter reflects the time-integrated impact of antecedent meteorological forcing (e.g., precipitation). ECMWF analyses do not assimilate ground-based observation of precipitation. Over land the information used by the model to generate rain is strongly constrained by in situ measurements of temperature and humidity. The use of precipitation data in the analysis continues to be studied at ECMWF. Lopez (2011) has demonstrated a positive impact on model performance of the direct 4D-Var assimilation of 6-hourly radar and rain-gauge rainfall accumulations. Considering all the stations (122) the differences in RMSDs between the two offline experiments is not very important, 0.135 m 3 m -3 and 0.124 m 3 m -3 for BEVAP_OLD and BEVAP_NEW, respectively. However, if a threshold of 0.2 for the fraction bare ground is set this difference increases to 0.118 m 3 m -3 and 0.087 m 3 m -3 . At the spatial resolution of BEVAP_NEW and BEVAP_OLD ECMWF considers that about 46% of the land is covered by a fraction of bare ground more than or equal to 0.2. Similar RMSDs are obtained with the operational IFS soil moisture product; in 2010, the RMSD is 0.110 m 3 m -3 and 0.088 m 3 m -3 in 2011 (new bare ground evaporation implemented in November 2010). Results from the recent land surface model developments at ECMWF were evaluated using a land surface benchmarking database gathered for this purpose. Those included field sites from the FLUXNET (http://www.fluxdata.org/) and CEOP (http://www.ceop.net/) observing networks where latent heat, sensible heat and carbon dioxide fluxes measurements are available. For instance, The land surface fluxes results from offline-runs indicated an average improvement of 8%, when adopting the HTESSEL (BEVAP_NEW) scheme in replacement of the former TESSEL scheme evaluated as RMSD reduction on both the latent and sensible heat fluxes measured over 36 FLUXNET and CEOP flux-towers for 2004 . In the same way, future activities will concern the evaluation of the impact of BEVAP_NEW with respect to BEVAP_OLD in term of surface fluxes.
The new bare ground parameterization also shows a consistent signal with the L-band microwave emission. Changes in TB with respect to changes in soil moisture as a result of the new bare ground evaporation is found to be close to 15 K and 10K in H and V polarizations, respectively. As the BEVAP_NEW soil moisture is drier than BEVAP_OLD, the simulated TB are larger with BEVAP_NEW surface fields. The latter are closer to SMOS observations but with still large global mean differences and standard deviation (about 10 K and 20 K, respectively). These residual biases are also related to other factors such as the SMOS instrument or input parameters of the radiative transfer model, which are not straightforward to assess. Radio frequency interferences affecting the SMOS measurements could also be responsible of the bias. CMEM configuration used in this study is based on de Rosnay et al., 2009 andSabater et al, 2011 using AMSR-E C-band data and local L-band data, respectively. Results presented in this study are very preliminary and a full calibration of the CMEM platform, underway at ECMWF for SMOS activities, should lead to more realistic simulated TB, in better agreement with SMOS data.

Conclusions
This study investigated the new bare ground evaporation formulation implemented in operations at ECMWF in November 2010. Bare ground evaporation over dry lands has been increased by adopting a lower stress threshold than for the vegetation, allowing a higher evaporation. Its impact on soil moisture is assessed as well as on the representation of terrestrial microwave emission. The latter is of particular interest for the planned use of SMOS brightness temperature within the new land-surface analyses; it will be effective only if realistic and dynamically consistent fields of brightness temperature are simulated as a function of the land-surface conditions. ECMWF has developed offline experiments to assess the impact of model changes. They are produced by a fixed version of the IFS (and used atmospheric forcing from ERA-Interim) while the operational product is based on a continuous effort to improve the analysis and modelling schemes, resulting in frequent updates of the system (few times a year). So offline experiments make it possible to study the impact of a single modification in the land surface modelling. Even if they are at a coarser spatial resolution than the operational product, they have the benefit of being less time-consuming and are very useful between the completion of future reanalyses of ECMWF.
In situ soil moisture from 122 stations (over 148 available) of the NCRS-SCAN network from all over the United States are used to evaluate the new bare ground evaporation formulation over two periods, 2010 and 2010-2011. It was first assessed using offline experiments to isolate its impact (2010 only) and then using the IFS operational product (2010-2011). The new scheme results in more realistic soil moisture values, particularly for dry areas; a decrease of about 26% in RMSD is obtained between the two offline experiments when considering the fraction of bare ground that has a threshold greater than or equal to 0.2 (from 0.118 m 3 m -3 to 0.087 m 3 m -3 ). Slightly higher levels of correlations were also obtained. The same conclusion is reached with the IFS operational analysis where a better agreement with in situ data was found in 2011 than in 2010. More realistic soil moisture also lead to better initial fields for simulating brightness temperature with the CMEM platform, a pre-requisite for SMOS data assimilation. This preliminary study demonstrated a better agreement between SMOS data and simulated brightness temperature with surface fields from the new bare ground evaporation. Future improvments of the land-surface physics will focus on evaporation from free water surface such as intercepted water on leaves.