Abstract

This paper presents a study undertaken in preparation of the work leading up to the assimilation of Soil Moisture and Ocean Salinity (SMOS) observations into the land surface model (LSM) Interaction Soil Biosphere Atmosphere (ISBA) at Météo-France. This study consists of an intercomparison experiment of different space-borne platforms providing surface soil moisture information [Advanced Microwave Scanning Radiometer for Earth Observing (AMSR-E) and European Remote Sensing Satellite Scatterometer (ERS-Scat)] with the reanalysis soil moisture predictions over France from the model suite of Système d’analyse fournissant des renseignements atmosphériques à la neige (SAFRAN), ISBA, and coupled model (MODCOU; SIM) of Météo-France for the years of 2003–05. Both modeled and remotely sensed data are initially validated against in situ observations obtained at the experimental soil moisture monitoring site Surface Monitoring of the Soil Reservoir Experiment (SMOSREX) in southwestern France. Two different AMSR-E soil moisture products are compared in the course of this study—the official AMSR-E product from the National Snow and Ice Data Center (NSIDC) and a new product developed at the Vrije Universiteit Amsterdam and NASA (VUA–NASA)—which were obtained using two different retrieval algorithms. This allows for an additional assessment of the different algorithms while using identical brightness temperature datasets. This study shows that a good correlation generally exists between AMSR-E (VUA–NASA), ERS-Scat, and SIM for low altitudes and low-to-moderate vegetation covers (1.5–3 kg m−2 vegetation water content), with a reduction in the correlation in mountainous regions. It also shows that the AMSR-E (NSIDC) soil moisture product has significant differences when compared to the other datasets.

1. Introduction

Soil moisture is the governing variable for modeling soil surface-to-atmosphere energy exchanges and land surface model (LSM) initialization, as it controls both evaporation and transpiration from bare soil surfaces and vegetation covers. Consequently, a significant amount of studies have been and are currently being conducted to obtain soil moisture estimates through land surface modeling (e.g., Dirmeyer et al. 1999; Georgakakos and Carpenter 2006) and remotely sensed surface soil moisture observations (e.g., Wagner et al. 1999a,b; Kerr et al. 2001; Njoku et al. 2003).

For the purpose of soil moisture remote sensing, observations in the microwave bands have been found to produce the best results. The optimal wavelength lies within the L-band range (∼1–2 GHz), as interference through vegetation water content at this frequency range is lower than at higher frequencies. However, instruments have been and are currently operated at higher frequencies (above 5 GHz) mainly because none of these missions were dedicated soil moisture missions. The first such dedicated soil moisture mission will be the Soil Moisture and Ocean Salinity mission (SMOS), to be launched in 2009. The first microwave instrument operated for an extensive duration and within adequate wavelengths was the Scanning Multichannel Microwave Radiometer (SMMR) on Nimbus-7 (operational from 1978 to 1987), which operated at bands at and above 6.6 GHz. SMMR was followed by the Special Sensor Microwave Imager (SSM/I; since 1987) and the similar Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI; since 1997), which operate at frequencies above 10 GHz. Instruments that are currently operational at frequencies similar to SMMR (and, therefore, closer to L band), are the Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) on board the National Aeronautics and Space Administration’s (NASA) Aqua satellite, WindSat on board the U.S. Navy’s Coriolis satellite, and the scatterometers (Scat) on board the European Remote Sensing Satellites 1 and 2 (ERS-1 and ERS-2). Finally, a new scatterometer [Advanced Scatterometer (ASCAT)] was launched on board the European Space Agency’s (ESA) meteorological operation (MetOp-A) satellite in 2006, and its data are now available operationally (Bartalis et al. 2007).

Despite almost 30 yr of experience with these microwave remote sensing instruments, it is still necessary to validate the soil moisture products obtained from these instruments through in situ soil moisture observations. However, such ground truthing has only been achieved over small temporal and spatial scales [e.g., the Soil Moisture Experiments (SMEX) or the campaign for validating the operation of SMOS (CoSMOS)], as it is economically and practically infeasible to observe soil moisture at high spatial and temporal resolution over large scales using in situ observations, mainly because of its high spatial variability (Bosch et al. 2006). Only in the present decade have there been attempts to establish long-term and large-scale soil moisture observation networks or data banks such as the Global Soil Moisture Data Bank (Robock et al. 2000), the Goulburn River experimental catchment in Australia (Rüdiger et al. 2007), or the soil moisture observing system–meteorological automatic network integrated application (SMOSMANIA) in southwestern France (Calvet et al. 2007; Albergel et al. 2008). However, these datasets only represent single points in space. This lack of spatial extent limits the usefulness of such datasets for assimilation into large-scale land surface models and also for disaggregation studies, as the large scale but also subpixel variability is not captured with single-point measurements. Moreover, satellite products are generally available at scales of about 0.25° or 25 km, which leads to problems in their validation process due to the different spatial scales (spatially averaged satellite products are compared to point measurements). Consequently, new validation methods complementing the existing soil moisture networks have to be conceived (Wagner et al. 2007). Under the assumption that LSMs—forced with high-quality atmospheric forcing data—adequately represent the surface soil moisture dynamics, the scale issues can be reduced. This assumption in turn will then allow the large-scale and long-term evaluation of the satellite products in terms of their temporal dynamics, as the products considered are essentially independent models.

In this paper, both the need for large-scale ground truthing and understanding of the subpixel heterogeneity of soil moisture are addressed. First, the temporal correlation of satellite products at a large scale with a synthetic high-resolution surface soil moisture database is presented. The high-resolution meteorological observation network throughout France (more than 1000 surface meteorological stations and more than 3500 daily rain gauges) has resulted in a high-quality atmospheric forcing database (Quintana-Seguí et al. 2008) for the operational land surface model Interaction Soil Biosphere Atmosphere (ISBA) of Météo-France, within the modeling system Système d’analyse fournissant des renseignements atmosphériques à la neige (SAFRAN)–ISBA—coupled model (MODCOU; SIM; see Habets et al. 2008). The SIM model simulates the soil moisture dynamics.

The satellite products used for this study were obtained from AMSR-E and ERS-Scat on ERS-2. Furthermore, the recent development of a new retrieval algorithm for AMSR-E (Owe et al. 2008) allowed for the comparison of the official AMSR-E product (Njoku et al. 2003) with this new database. In the first part of this study, the different datasets used are discussed, followed by a brief comparison of those remotely sensed datasets and SIM with in situ observations of the Surface Monitoring of the Soil Reservoir Experiment (SMOSREX) site near Toulouse, France (de Rosnay et al. 2006), to determine their capability for representing the temporal soil moisture dynamics of a point or pixel. The good results of this analysis between the land surface model, in situ observations, and satellite data also show that previous results obtained over Spain (Wagner et al. 2007; one single satellite pixel) or over Australia (Draper et al. 2007; several in situ observations for a number of pixels) can be extrapolated to a national—or even continental—scale, as they show the same tendencies. The differences between the various soil type databases used in the satellite retrieval schemes and the model database make it difficult to compare absolute values. Consequently, the discussion of this paper will focus on the normalized datasets. In the second part, the intercomparison study then presents the correlations and mean differences between all datasets (i.e., also between the different satellite products).

2. Datasets

Because of the limitation in the spatial extent of SIM, this study is limited to the watersheds of mainland France. Nevertheless, the surface and climatic conditions throughout the country are sufficiently variable (ranging from subhumid to alpine) to give a statistically sound data basis for a representative analysis. The years 2003–05 were chosen for this study, as data exists for all sources (SIM: 1970–2006; ERS-Scat: 1992–2006; AMSR-E: 2003 onward; and SMOSREX: 2001 onward). Moreover, this 3-yr period includes both very dry and very wet climatic conditions, which are necessary to determine the dynamic range of the soil moisture observations within each pixel. The following sections briefly outline the various datasets used for this study.

a. SAFRAN-ISBA-MODCOU

The modeled surface soil moisture database was obtained from the modeling system SAFRAN–ISBA–MODCOU (SAFRAN, an atmospheric forcing database; ISBA, a land surface model; and MODCOU, a hydrological routing model). Of these three model chain segments, only SAFRAN and ISBA were of importance for the present study.

SAFRAN is a reanalysis forcing database, initially developed to improve snowfall and avalanche forecasting. Within SAFRAN, the main atmospheric forcing parameters are analyzed. Each atmospheric parameter is analyzed individually using an optimal interpolation method. The final size of a grid cell within SAFRAN is 8 × 8 km2. For precipitation, however, the actual pixel sizes of SAFRAN vary, as they represent zones of climatic conditions rather than regularly gridded areas. Each climatic zone covers about 1000 km2, resulting in about 600 such pixels over France. The forcing parameters are, in principle, assumed to be homogeneous within one such pixel, however, they vary on a subpixel scale with topography. Apart from precipitation the SAFRAN forcing data is available at 6-hourly intervals. Precipitation is obtained using daily observations at the rain gauges and then interpolated into hourly time steps as a function of the relative humidity during the day. The SAFRAN database has recently been validated against in situ observations and found to be well correlated (Quintana-Seguí et al. 2008).

The land surface model used in SIM is ISBA (Noilhan and Planton 1989; Noilhan and Mahfouf 1996), which is used operationally as the land surface scheme within the numerical weather prediction system at Météo-France. The soil layer and soil moisture dynamics are modeled within a three-soil-layer model (Boone et al. 1999), which is based on the force–restore approach, in which the three soil layers are a surface layer of 1-cm depth, forming part of a root zone layer above the third, deep layer.

There is no previous study presenting a verification of the SIM surface soil moisture product. On the other hand, the ISBA model has been extensively validated for various biomes. In particular, a number of studies exist comparing a point-specific calibrated ISBA version to actual in situ soil moisture observations in France (Calvet et al. 1998a,b; Boone et al. 1999; Calvet and Noilhan 2000; Sabater et al. 2007). The latter case corresponds to SMOSREX, and the former to a previous experiment [Monitoring the Useable Soil Reservoir Experimentally (MUREX)] in the same region. The root-mean-square error (RMSE) for those cases is 0.06 and 0.07 m3 m−3, respectively, with a mean difference in the order of 0.01 and 0.03 m3 m−3, respectively. In both cases the Nash efficiency was calculated with 0.65/0.59. Based on the results of Prigent et al. (2005), this level of error between in situ observations and model predictions is expected while maintaining a good correlation despite the different observation and model layer depths. The land surface parameters for ISBA are obtained from the Ecoclimap database (Masson et al. 2003). The parameters provided by Ecoclimap are originally provided at 1-km resolution and are aggregated to the model resolution of 8 km.

b. AMSR-E

AMSR-E is a passive microwave scanning radiometer, operating at six wavelengths within the microwave spectrum (6.925, 10.65, 18.7, 23.8, 36.5, and 89 GHz) in horizontal and vertical polarizations, flown on NASA’s Aqua satellite. The total swath width during an overpass is approximately 1445 km, with footprint resolutions ranging from 56 (6.925) to 5 km (89 GHz). Aqua is a sun-synchronous satellite orbiting Earth approximately 14 times each day, with morning/descending and afternoon/ascending overpasses at around 0130/1330 local time (LT). This configuration results in a repeat coverage of approximately every three days in the equatorial latitudes and more frequent coverage in higher latitudes. For the particular case of France, Aqua overpasses take place at 4 out of 5 days for both ascending and descending orbits.

Currently, two different data products are freely available. The official product can be obtained through the National Snow and Ice Data Center (NSIDC, hereafter AMSR-E (NSIDC)], while a new product has recently been made available through the Vrije Universiteit Amsterdam in collaboration with NASA (hereafter AMSR-E (VUA–NASA)]. Both products are briefly described in the following sections.

1) AMSR-E (NSIDC)

The AMSR-E (NSIDC) data used for this study were obtained from the operational Level 3 B03 AMSR-E dataset (Njoku 2004). While the original resolution at 10.65 GHz is ∼38 km, the data is binned into regular 0.25° × 0.25° pixels through oversampling at 10-km intervals.

The NSIDC method uses two low-frequency, dual-polarized channels to optimize the three parameters (soil moisture, vegetation optical depth, and the effective soil temperature) simultaneously. The method was originally developed and tested for the C- and X-band channels. Unfortunately, severe radio-frequency interference (RFI) was discovered within C band (6.925 GHz) over the United States and Japan and X band over Italy and Great Britain (Li et al. 2004, 2006). For this reason, the retrieval algorithm was applied to the X-band (10.65GHz) and Ku-band (18.7 GHz) brightness temperatures. This has some important disadvantages: 1) the 18-GHz channel introduces atmospheric influences; 2) the observation depth of the soil moisture product is reduced to 5–10 mm, which is approximately half the potential range of C band; and 3) vegetation attenuation effects are more significant than at lower frequencies.

2) AMSR-E (VUA–NASA)

The VUA–NASA retrieval products from AMSR-E are derived according to the Land Surface Parameter Model (LPRM; Owe et al. 2001, 2008). The LPRM is a three-parameter retrieval model for passive microwave data, which uses one dual-polarized channel (either 6.925 or 10.65 GHz) for the retrieval of both surface soil moisture and vegetation water content (VWC). The land surface temperature is derived separately from the vertically polarized 36.5-GHz channel.

The forward radiative transfer model in LPRM is based on one vegetation layer (τω approach), and the vegetation optical depth is parameterized as a function of the microwave polarization difference index (MPDI) and soil moisture according to Meesters et al. (2005). This method is applied globally and requires no regional calibration or fitting parameters to aid the retrieval process.

The main differences with the AMSR-E (NSIDC) soil moisture product lay in the use of a higher frequency band for the retrieval of the land surface temperature (LST) and the parameterization of the vegetation optical depth, leaving only the soil moisture to be optimized.

c. ERS-Scat

The ERS-Scat data are obtained through active microwave remote sensing—that is, an energy pulse is sent to the surface and the intensity of the returned signal is then used within the retrieval algorithm to derive a relative soil moisture state. ERS-Scat is operated at 5.3 GHz (C band), observing only the vertically polarized backscatter within this band, thus resulting in a similar observation depth as AMSR-E. RFI has been found to have little affect on active microwave remote sensing at this frequency. ERS-Scat has a morning/descending and evening/ascending orbit at 1030/2230 LT, with a varying repeat coverage of about 2 to 8 days. The spatial resolution of an ERS-Scat footprint is in the order of 50 km, while the soil moisture product is binned into pixels of 0.25° (north–south extent) and 25 km (west–east extent).

The soil moisture product is provided in relative values, ranging from 0% to 100%. The normalization of the backscatter signal is done using the minimum- and maximum-observed backscatter from the 1992–2000 period as dry and wet references, respectively. The retrieval algorithm is described in detail in Wagner et al. (1999a,b,c).

d. SMOSREX

SMOSREX is an experimental field site for in situ and remotely sensed soil moisture observations, jointly operated by various research institutes in France and located south of Toulouse (43°23′N, 1°17′E) in southwestern France (de Rosnay et al. 2006). The overall size of SMOSREX is approximately 6000 m2, separated into two areas with either bare soil or fallow. The climate is temperate with monthly-mean maximum temperatures of 5°C in winter and 24°C in summer and an average annual cumulative precipitation of about 650 mm. The surface soil consists of a sandy loam, with 16% clay, 47% silt, and 37% sand.

Most instruments installed at the site have been in operation since 2001. The main feature of this site is a tower-mounted L-band radiometer for the production of multiangle brightness temperatures. Other instrumentation includes a weather station, and soil temperature and moisture sensors installed at various points and depths. The soil moisture sensors (i.e., ThetaProbes) used for this study are located at four points within the fallow section of the site (most representative for the overall region and, therefore, the model simulations), with a spacing of only a few meters. The sensors are vertically inserted at the surface—therefore, integrating the soil moisture content from 0 to 6 cm—and a temporally averaged soil moisture content is stored every 30 min individually for all sensors. The calibration of the sensors is presented in de Rosnay et al. (2006).

For the purpose of this study, the in situ observations were aggregated into daily averages and compared to the respective datasets obtained through the model and remote sensing. Spatially averaging the observations of those four probes reduces the effect of spatial variability within and increases the representativeness of the soil moisture observations and also reduces the individual observations to one point in space.

e. Data preparation

The results presented in this paper are based solely on the datasets from descending orbits (nighttime) to avoid overly solar effects in the satellite data due to sun glint and strong temperature gradients between the vegetation and the surface, and also within the surface layer but also due to Faraday rotation and temperature gradients within the sensor, which are more pronounced during daytime overpasses (Kerr and Njoku 1990). Other effects, such as quick dry down or the lack thereof due to local changes in solar radiation, which cannot be adequately represented in an LSM and, in reality, may be affected by cloud coverage and wind, among other factors, play a significant role in the daytime evolution of surface soil moisture. While the in situ observations were spatially and temporally averaged, the soil moisture simulations were extracted for the time steps close to the overpass times of the satellites. A comparison of the differences between the individual measurements of the soil moisture probes and their spatial average at 0600 UTC and also between the daily average with the spatial average at 0600 LT resulted in an RMSE of 0.036 m3 m−3 in both cases. This shows that spatial and diurnal variabilities contribute to the same extent to the uncertainty in the in situ observations. The use of a spatially and temporally daily average is, therefore, justifiable.

All data have been reprojected from their original coordinate systems onto a regular 0.25° × 0.25° grid using a nearest neighbor approach. As the overall footprints of AMSR-E and ERS-Scat are in the order of 50 km with a spacing of about 10 km between the center points, and the gridded products used in this study are binned at 25 km or 0.25°, respectively, a spatial shift in the data due to the reprojection process (a maximum of 12 km) is not expected to add any additional noise to the data or affect the data quality, as a footprint with its center 12 km from the pixel center would still include information from more than half of the land surface corresponding to the pixel area as a result of its size. To obtain an average pixel value within the reprojected pixels, all original pixels with their center falling into one reprojected pixel were averaged to one single value. This average value was then assumed to be the representative soil moisture of the reprojected pixel. In the case of the satellite observations, only one original pixel would generally fall into a reprojected pixel as a result of the similarity in size; therefore, no errors are introduced because of the averaging of two satellite pixels. For all datasets, the same general rule applied for the reprojection process: avoid inconsistencies between the datasets introduced through the reprojection and aggregation process.

In a brief study, it was examined whether the variability between the soil moisture of high-resolution SIM pixels with their averaged low-resolution equivalent resulted in any errors within the analysis. However, no relationship between the subpixel heterogeneity and the spatial distribution of the correlation coefficients between the different soil moisture products presented in the following section was found.

The soil moisture data from the satellites and SMOSREX were normalized following the approach presented by Pellarin et al. (2006), in which the maximum and minimum of the soil moisture range was not determined by the soil type but rather by the observed dynamic range within each individual pixel within the full study period (2003–05). To exclude any abnormal outliers due to observational errors or instrument noise, the 90% confidence interval was chosen to define the upper and lower soil moisture content using (1) and (2), respectively:

 
formula
 
formula

where int+ and int are the upper and lower confidence limits; μ(SM) is the average soil moisture content for the pixel; and σ(SM) is the standard deviation of the soil moisture content for each pixel. With the knowledge of the upper and lower soil moisture content, the absolute soil moisture value is then normalized using (3):

 
formula

where SMobs is the individual soil moisture observation, and θn is its normalized soil moisture value. As a simplification it is assumed that the data are normally distributed, so that 90% of the data lie by definition within a range of μ±1.64σ. All data outside of this range were discarded. Also, pixel values were excluded from the overall analysis in which SIM predicted frozen soil water. Because model simulations as such have no outliers due to instrumentation errors, no screening of extreme values is required. The soil moisture from SIM is, therefore, normalized using the modeled maxima and minima of each individual pixel instead of int+ and int.

Pixels located over major urban agglomerations (i.e., Lille, Paris, Lyon, Bordeaux, Toulouse, and Marseille) were not excluded. However, the correct representation of the soil moisture is doubtful, as SIM is not capable of giving realistic soil moisture conditions over urban (and, consequently, sealed) areas, and moreover, the possibility of pixels subjected to potential radio-frequency interference (Li et al. 2004) is higher in these areas. Nevertheless, the number of these pixels is small (<0.5% of the total) compared to the total over France, and their overall effect on the statistical analyses was found to be negligible.

3. Comparison of the soil moisture products with in situ observations

An evaluation of the surface soil moisture products obtained from SIM and the satellites was undertaken using the same three years of in situ soil moisture observations as for the remainder of this study (2003–05). The in situ data were obtained from the observations at the experimental site SMOSREX. The data from the four surface soil moisture sensors installed at SMOSREX were averaged both spatially and over time, so that one daily averaged observation was obtained for each day. This approach reduced the existing noise levels in the in situ observations, as discussed in the previous section. The model and satellite data used here are the binned and reprojected data as for the large-scale study in section 4, as described above. SIM was not especially calibrated to the conditions at SMOSREX. For this evaluation study, various statistical parameters were calculated: the RMSE, the mean difference or bias between two datasets, the correlation coefficient (r) between two datasets, and the Nash efficiency coefficient (N). All statistics presented in the following sections were calculated for the normalized soil moisture values and are, therefore, dimensionless.

In a first step, the absolute values of the soil moisture products were compared with the in situ data. For this purpose, the already normalized ERS-Scat data were transferred into absolute values using the known maximum and minimum surface soil moisture observations at SMOSREX. While a good correlation exists between SIM and SMOSREX datasets, a severe lack of soil moisture dynamics is observed for the AMSR-E (NSIDC) dataset (not shown). However, the AMSR-E (VUA–NASA) data is well correlated despite an apparent wet bias. Finally, the ERS-Scat observations are also well correlated in terms of their temporal dynamics. In contrast to the AMSR-E (VUA–NASA) data, the ERS-Scat data exhibits a dry bias. Because of the different soil moisture dynamics and biases, it is difficult to compare the various datasets in detail; consequently, all comparison in the remainder of this paper will be undertaken with normalized data (Fig. 1).

Fig. 1.

Time series plots (2003–05) of the normalized values of the in situ observations at SMOSREX (black lines) and the four surface soil moisture products: SIM, AMSR-E (NSIDC), AMSR-E (VUA–NASA), and ERS-Scat (+). The model predictions and satellite observations were obtained from respective low-resolution pixels covering SMOSREX.

Fig. 1.

Time series plots (2003–05) of the normalized values of the in situ observations at SMOSREX (black lines) and the four surface soil moisture products: SIM, AMSR-E (NSIDC), AMSR-E (VUA–NASA), and ERS-Scat (+). The model predictions and satellite observations were obtained from respective low-resolution pixels covering SMOSREX.

The comparison of the normalized SIM and SMOSREX datasets shows a good temporal correlation (r = 0.755; N = 0.478) with a bias (−0.083) toward the model simulations (i.e., the simulations tend to be drier), with the exceptions of very dry conditions when the model has the tendency to overestimate the soil moisture at this site (Fig. 2). Throughout the years, a higher level of surface soil moisture dynamics is observed within the model data (Fig. 1), which results in a RMSE of 0.198. This phenomenon is explained by inaccuracies in the forcing data due to the spatial interpolation process within SIM and the differences in the thickness of the observed soil layers (1 cm for SIM compared to 0–6 cm for the ThetaProbes). However, there are only a few data points causing this noise; therefore, this is deemed acceptable.

Fig. 2.

Scatterplot of the normalized in situ soil moisture observations at SMOSREX (vertical axis) with the four low-resolution datasets for the years 2003–05. Darker regions show a higher density of data points.

Fig. 2.

Scatterplot of the normalized in situ soil moisture observations at SMOSREX (vertical axis) with the four low-resolution datasets for the years 2003–05. Darker regions show a higher density of data points.

The normalized AMSR-E (NSIDC) data display a very high variation, with interchanging peaks and troughs every three months (Fig. 1). Every year, minimal values are reached during winter and maximum values during summer. This recurring negative correlation with the in situ data results in a high RMSE and low overall correlation (r = 0.132; N = −0.734; bias = 0.132; RMSE = 0.356). In contrast to the comparison of the absolute data, the persistent wet bias in the AMSR-E (VUA–NASA) data has been reduced because of the normalization. Similar to the SIM predictions, a strong correlation between in situ and remotely sensed data is found in this case with a wet bias toward the AMSR-E data (r = 0.775; N = 0.471; bias = 0.072; RMSE = 0.194). Finally, the ERS-Scat observations are also well correlated over time with the in situ data (r = 0.618; N = 0.125); however, a dry bias (−0.085) results in a more significant RMSE of 0.244 than for the AMSR-E (VUA–NASA) data. Because ERS-Scat data are only available from August 2003 onward, the identical periods of data (i.e., August 2003–December 2005) for the other surface soil moisture products was undertaken (not shown) to verify that the first seven months did not introduce significant biases in the statistical analyses, which then would not be seen in the ERS-Scat comparisons. The differences in correlation, RMSE, and bias did not change significantly for any of the intercomparisons during either the full three years or the period August 2003–December 2005. Consequently, all comparisons shown in the remainder of this paper are based on the full period. The Nash efficiency coefficient for both SIM and AMSR-E (VUA–NASA) is acceptable. The coefficients are also similar to each other, suggesting that the two datasets perform equally well compared to the in situ observations, while the low Nash efficiency of ERS-Scat is due to the relatively strong bias in the satellite data. In the case of the AMSR-E (NSIDC) data, the negative Nash efficiency suggests, by definition, that an average value of the in situ observations would compare better with the overall observations than the remotely sensed observations. This is an important finding, as it shows the extreme difference between the in situ observations and the satellite product.

Four aspects have to be considered for the cause of the differences observed in this evaluation: (i) the scale difference (8 km and 0.25° for the model and the satellite, respectively, against a single-point observation), as the comparison or validation of soil moisture products at different spatial scales will remain difficult in most cases unless a representative catchment average soil moisture monitoring site (Grayson and Western 1998) can be identified; (ii) the soil database, as the model soil information constitutes an average of the soil particle size distribution within an 8 km/0.25° pixel, which may result in significant differences compared to the soil conditions at the point of observation (the particle size analysis for SMOSREX yielded 16% clay, 47% silt, and 37% sand; the particle size distribution within Ecoclimap is 25% clay, 25% silt, and 50% sand); (iii) the forcing data, as it is obtained by interpolation between observations and atmospheric predictions, which may miss localized events; and (iv) the observation depth, with the model layer of 1 cm and approximately the same depth for the satellite observations against the integrated soil moisture content at 0–6 cm for the in situ observations, may result in different dynamics.

Considering the above four aspects, SIM, AMSR-E (VUA–NASA), and ERS-Scat perform well when compared to the SMOSREX in situ observations and also show a good representation of the dynamic behavior of the soil moisture content. For SIM, an RMSE of 0.198, with a dynamic range of the surface soil moisture at the site of ∼0.3 m3 m−3, can be translated into an absolute error in the soil moisture of just under 0.06 m3 m−3. This result is particularly good, as SIM was not calibrated to the conditions at SMOSREX but rather used the vegetation and soil conditions obtained from Ecoclimap. Moreover, despite the differences in scale, these errors are identical to the performance of the site-specific calibrated model.

Depending on the application, the calculated error may be considered large or acceptable. For atmospheric studies, it is more important to obtain a good representation of the temporal dynamics, while the absolute soil moisture state is less important. On the other hand, an error of 0.06 m3 m−3 exceeds the validation goals of future satellite missions (Kerr et al. 2001). In the first case, the evaluation of satellite data against any benchmark is necessary, shown by the lack of temporal dynamics in the AMSR-E (NSIDC) data. In the second case, two factors influencing the RMSE have to be considered to qualify the above value of 0.06 m3 m−3: (i) the mean difference or bias between SIM and SMOSREX and (ii) the spatial uncertainty of the in situ observation. Biases, in the case of the normalized data 0.084 (or 0.025 m3 m−3), may be removed using various techniques (e.g., Drusch et al. 2005), while the uncertainty in the spatial averaging of the four in situ observations is in the order of 0.036 m3 m−3. In particular, the removal of the bias would lead to a significant decrease of the RMSE. Consequently, it is concluded that SIM may be used with reasonable confidence for a large-scale model intercomparison study, assuming that Ecoclimap provides similarly good information for all other model pixels.

While the correlations derived from Fig. 1 are relatively large for SIM, AMSR-E (VUA–NASA), and ERS-Scat, much of the captured variability is seasonal (dry in summer and wet in winter). To assess the coherence with the in situ observations and to avoid seasonal effects, anomalies are calculated. The difference in the mean is calculated for a sliding window of five weeks, and the difference is scaled to the standard deviation. Table 1 shows seasonal scores, including the Kendall statistics and p value. All the products are significantly correlated to the in situ observations, except for satellite products at specific periods of the year. While SIM presents significant correlations throughout the year, all the satellite products are not significantly correlated to in situ observations in winter (December–February). This may be explained by the sensitivity of the microwave signal (either active or passive) to soil freezing and by the reduced dynamics of the surface soil moisture during winter. Both VUA–NASA and NSIDC products present high correlations of the anomalies for the other seasons. On the other hand, ERS-Scat has significant correlations during spring (March–May) only. The lack of significance of ERS-Scat during the summer (June–August) and autumn (September–November) seasons may be explained by the small number of observations over the SMOSREX site (28 and 41, respectively) compared with AMSR-E (184 and 175, respectively, for the VUA–NASA product).

Table 1.

Comparison of monthly anomalies of surface soil moisture products (SIM, AMSR-E, and ERS-Scat) with in situ 0–6-cm observations at the SMOSREX site for three pooled annual cycles (2003–05).

Comparison of monthly anomalies of surface soil moisture products (SIM, AMSR-E, and ERS-Scat) with in situ 0–6-cm observations at the SMOSREX site for three pooled annual cycles (2003–05).
Comparison of monthly anomalies of surface soil moisture products (SIM, AMSR-E, and ERS-Scat) with in situ 0–6-cm observations at the SMOSREX site for three pooled annual cycles (2003–05).

4. Intercomparison (all datasets)

a. General correlation (all data)

Figure 3 shows correlation maps of the different remote sensing and modeled datasets for all surface conditions and all years. A good correlation exists between the three datasets AMSR-E (VUA–NASA), ERS-Scat, and SIM—in particular, for regions of herbaceous vegetation over regions with little relief—with a range of the coefficient of correlation from 0.2 to 0.9. Areas with denser vegetation, such as the forest of Les Landes in the southwest along the Atlantic coast, show a lower level of correlation, which would have to be expected because of the masking effect on the microwave emissions of the soil moisture through vegetation. Similarly, low correlations are found in regions with strong relief, such as the Massif Central and the Alps. The good correlation of ERS-Scat with SIM in the Italian Alps should be ignored, as only a few data points were available due to the overpass rate of ERS over the region and the filtering of days with frozen soils or snow. Mountainous regions cause errors in both the modeling of soil moisture and its retrieval from satellite observations. First, there exists a high level of uncertainty in the soil depth and its variability in those regions, which affects the predictions of the soil moisture dynamics in the SIM model. Second, relief interferes with the retrieval of low-resolution remotely sensed soil moisture observations and may cause considerable levels of errors (Mätzler and Standley 2000). The AMSR-E (NSIDC) product has virtually only low correlations with any of the other datasets—even producing negative correlations overall (Table 2).

Fig. 3.

Maps of the coefficient of correlation between the various soil moisture products (normalized values) over mainland France. The circles highlight the six major metropolitan areas of France.

Fig. 3.

Maps of the coefficient of correlation between the various soil moisture products (normalized values) over mainland France. The circles highlight the six major metropolitan areas of France.

Table 2.

Statistics of the intercomparison between the different datasets (normalized surface soil moisture data). The values in each cell correspond to the coefficient of correlation, bias, and RMSE, respectively.

Statistics of the intercomparison between the different datasets (normalized surface soil moisture data). The values in each cell correspond to the coefficient of correlation, bias, and RMSE, respectively.
Statistics of the intercomparison between the different datasets (normalized surface soil moisture data). The values in each cell correspond to the coefficient of correlation, bias, and RMSE, respectively.

This analysis also shows that previous results obtained over Spain (Wagner et al. 2007; one single satellite pixel) or over Australia (Draper et al. 2007; several in situ observations for a number of pixels) can be extrapolated to a national—or even continental—scale, as they show the same tendencies. In particular, the lack of soil moisture dynamics within the AMSR-E (NSIDC) dataset is apparent and is shown in all the studies.

The data used to derive the spatial plots of Fig. 3 are summarized in Table 2 as showing the respective coefficient of correlation (r), RMSE, and bias between the datasets. Compared to SIM, the ERS-Scat dataset has the highest overall correlation (r = 0.728) and lowest RMSE (0.201), followed by AMSR-E (VUA–NASA) with an r = 0.491 and an RMSE of 0.297. As mentioned before, AMSR-E (NSIDC) has a negative correlation of −0.014 with an RMSE of 0.370. The RMSE presented here is the RMSE obtained from the normalized results, that is, it represents the relative error of the soil moisture dynamical range. Assuming an average dynamic range of 0.3 m3 m−3 and that SIM gives accurate in situ observations, this would translate into an average error of 0.056 m3 m−3 for ERS-Scat, which is higher than the design accuracy of SMOS (0.04 m3 m−3).

Like the bias between the SIM and SMOSREX datasets, the biases shown between SIM and the three satellite products are all positive. This suggests that a consistent dry bias exists within SIM. An explanation for the bias between SIM and SMOSREX is the different thickness of the observed soil layers (1 cm in the model compared to 0–6 cm in situ), as the deeper profile of the in situ observations is likely to maintain a higher soil moisture content because it is less affected by evaporation than the thin surface layer in the model. Furthermore, other aspects such as erroneous soil type information, biased forcing data, and biases in the soil moisture retrieval for the satellites may result in consistent biases.

b. Correlations specific to land surface cover

A comparison of vegetation maps with the results of Fig. 3 suggested a connection between the accuracy of the remotely sensed soil moisture information and the land cover. Therefore, the dominant land surface cover within each satellite-type pixel was determined using the information from Ecoclimap to identify vegetation specific correlations for each data product. For this purpose, the different vegetation types within each 0.25° pixel were aggregated into three dominant cover types: (i) cultivated soils; (ii) grasslands; and (iii) forests (Fig. 4). Relatively good correlations exist between SIM, ERS-Scat, and AMSR-E (VUA–NASA) for the two herbaceous vegetation covers (Figs. 5a and 5b). As in the analysis of the overall dataset, ERS-Scat and SIM have the highest correlation coefficient and lowest RMSE. Similarly, the pairs SIM/AMSR-E (VUA–NASA) and AMSR-E (VUA–NASA)/ERS-Scat have slightly lower correlation coefficients and higher RMSEs, and AMSR-E (NSIDC) has negative correlations throughout. These results [with the exception of the AMSR-E (NSIDC) data] are not surprising given that remotely sensed soil moisture information should, theoretically, be retrievable with a high level of accuracy over herbaceous vegetation types. In herbaceous vegetation covers, active and passive methodologies are expected to show similar performances, especially when using a similar frequency. The higher correlations of ERS-Scat and SIM, as compared to AMSR-E (VUA–NASA) and SIM, show potential for improvement of the AMSR-E (VUA–NASA) product. Part of the difference might be explained by the limited range of moisture values in the optimization routine for the AMSR-E (VUA–NASA) product (0%–50%). For the retrieval of the current AMSR-E (VUA–NASA) data, the soil moisture content is limited to a maximum of 0.5 m3 m−3. However, it was found that the surface soil moisture states often reached this point of saturation (Fig. 1). Consequently, if this constraint were to be relaxed, and the retrieval process were allowed to produce higher values, a quasi-normalized soil moisture product may be obtained (this aspect has been considered for the next version of soil moisture data, which has recently been made available). However, as a consequence of this constraint, the maximum soil moisture is currently underestimated, which leads to an underestimation of the dynamic range, and, consequently, a wet bias in the AMSR-E (VUA–NASA) data. The methodology behind the ERS product avoids this caveat by scaling between minimum- and maximum-observed signals over the period 1992–2000.

Fig. 4.

Location of pixels with the different dominant land cover types (cultivated soils, grasslands, and forests), based on the fractional covers obtained from Ecoclimap and aggregated to 0.25° resolution.

Fig. 4.

Location of pixels with the different dominant land cover types (cultivated soils, grasslands, and forests), based on the fractional covers obtained from Ecoclimap and aggregated to 0.25° resolution.

Fig. 5.

Vegetation-type specific comparison of the different soil moisture products for the three dominant vegetation types—(a) cultivated soils, (b) grasslands, and (c) forests—using the data from the period 2003–05. The scatterplots and their corresponding statistics are located on opposite sides of each figure, that is, the scatterplot of the data pair SIM–AMSR (VUA–NASA) is in the top left-hand corner, while the respective statistical values are found in the bottom right-hand corner. Darker regions show a higher density of data points.

Fig. 5.

Vegetation-type specific comparison of the different soil moisture products for the three dominant vegetation types—(a) cultivated soils, (b) grasslands, and (c) forests—using the data from the period 2003–05. The scatterplots and their corresponding statistics are located on opposite sides of each figure, that is, the scatterplot of the data pair SIM–AMSR (VUA–NASA) is in the top left-hand corner, while the respective statistical values are found in the bottom right-hand corner. Darker regions show a higher density of data points.

The comparison of the various datasets for forested regions (Fig. 5c) overall shows lower correlations and higher RMSEs. Again, ERS-Scat produces the best correlation with SIM, followed by AMSR-E (VUA–NASA), and AMSR-E (NSIDC). Moreover, the ERS-Scat soil moisture product appears to conserve its good correlation with SIM from the analysis of the herbaceous vegetation types. Under the assumption that SIM is equally valid for forested regions as for regions with low vegetation, it may be concluded that two effects may influence the consistency of ERS-Scat for different vegetation types. First, the retrieval process of ERS-Scat implicitly takes into consideration the vegetation type by scaling the current signal between the wet end dry ends of its long-term database. This statement has significance for other soil moisture missions in both active and passive microwave remote sensing, as the approach taken for the retrieval of ERS-Scat soil moisture may be applied along with more sophisticated radiative transfer models. Second, the ERS-Scat is well calibrated and has a low radiometric noise of about 0.15 dB, which allows estimating soil moisture even in areas where abundant forest cover reduces the effective sensitivity of backscatter to soil moisture.

An aspect of the data visible within the scatterplots of Fig. 5 is the apparent bimodality of the SIM data, with data clouds forming for the lower and upper value ranges. Figures 6 and 7 show histograms of the surface soil moisture from the four different low-resolution data sources—SIM, AMSR-E (VUA-NASA), AMSR-E (NSIDC), and ERS-Scat—for SMOSREX and for France, respectively. The histograms of the various data sources show different patterns at the local scale (Fig. 6). While SIM and ERS-Scat show clear bimodalities, this is not the case for the two AMSR-E products—AMSR-E (VUA-NASA) having several peaks with a saturation at 1, and AMSR-E (NSIDC) data being almost normally distributed—though all the datasets have a minima in the range of 0.4–0.6. The histogram of the in situ data at SMOSREX (Fig. 6e) also shows a bimodality, although with its maximum in the wet spectrum. This would suggest that preferred soil moisture states exist at SMOSREX but that the distribution is not correctly captured by the various models.

Fig. 6.

Histograms showing the relative frequency (vertical axis) of the various normalized soil moisture observations (horizontal axis) and predictions for the years 2003–05 for the SMOSREX site: SIM model, AMSR-E (VUA–NASA), AMSR-E (NSIDC), and ERS-Scat product of University of Vienna, in situ observations.

Fig. 6.

Histograms showing the relative frequency (vertical axis) of the various normalized soil moisture observations (horizontal axis) and predictions for the years 2003–05 for the SMOSREX site: SIM model, AMSR-E (VUA–NASA), AMSR-E (NSIDC), and ERS-Scat product of University of Vienna, in situ observations.

Fig. 7.

Same as Fig. 6 but for France.

Fig. 7.

Same as Fig. 6 but for France.

The nonnormal distribution of the histograms has significance for the normalization process, as it was previously assumed that the soil moisture distribution was sufficiently normal at each point. A violation of the assumption of normality would mean that the 90% confidence interval could not be calculated with Eqs. (1) and (2). To assess this, the distribution of the soil moisture states at the national scale was studied (Fig. 7).

An exception here is SIM with a clear peak in the dry spectrum (0.2) and AMSR-E (VUA–NASA) being skewed toward the wet end (Fig. 7). The overall distributions show that SIM retains its clear bimodality with a peak in the dry spectrum, while the ERS-Scat and AMSR-E data become more normally distributed. For the AMSR-E datasets, the distribution of AMSR-E (NSIDC) data becomes almost Gaussian with a slight skew toward the wet end, while the AMSR-E (VUA–NASA) data is more evenly distributed. Because the normalization procedure of Pellarin et al. (2006) is only applied to the AMSR-E data, it is concluded that the normalization process is still applicable to the majority of the pixels throughout France.

The results shown here are in line with other studies. For example, Teuling et al. (2005) showed that preferred soil moisture states may exist locally. However, they found that this effect could not be observed at all sites studied, and that it could not be linked to local soil conditions and may, therefore, be a random effect. This conclusion is supported by Figs. 6 and 7 in which the histograms for the data at SMOSREX suggest that local preferred wet and dry states exist, while the distribution of all observations over France is not bimodal.

c. Intraseasonal correlation

The bimodality presented in the previous section is unlikely to be caused by differences in the soil types, as the soil moisture data were normalized, and SMOSREX also appears to have this distribution (Fig. 6e). The bimodality is related to the varying soil moisture states, which are caused by either precipitation events or seasons. As the effect of precipitation events on the soil moisture distribution is difficult to obtain, the results obtained for the cultivated soils in Fig. 5 were separated according to the various seasons. This analysis (Fig. 8) clearly shows the different preferred soil moisture states during summer (dry) and winter (wet), which are, consequently, the main reason for the creation of the data clouds in Fig. 5. Similar results of preferred soil moisture states during the various seasons have been shown by Settin et al. (2007), in which they were largely attributed to the precipitation intervals and intensities during the various seasons. Interestingly, the two AMSR-E products have nearly the same correlations with SIM in springtime, which would suggest that the two radiative transfer models work similarly well during this period.

Fig. 8.

Same as Fig. 5 but of scatterplots for pixels with herbaceous vegetation only (cultivated soils and grasslands) for (a) spring, (b) summer, (c) autumn, and (d) winter.

Fig. 8.

Same as Fig. 5 but of scatterplots for pixels with herbaceous vegetation only (cultivated soils and grasslands) for (a) spring, (b) summer, (c) autumn, and (d) winter.

5. Conclusions

In this paper, an intercomparison study of several remotely sensed surface soil moisture products with the reanalysis LSM predictions over France has been presented. First, the LSM predictions and the satellite observations were compared with a 3-yr in situ surface soil moisture dataset from an experimental site in southwestern France (SMOSREX) to determine their capability for representing the temporal dynamics of a point or pixel. A good correlation was found between the model predictions and the in situ data, despite a slight dry bias within the model predictions. Based on this evaluation, it was then assumed that the LSM predictions over France may be used as a credible approximate estimate in the absence of more direct surface soil moisture observations for the entire country.

The analyses of this study have shown that two of the three satellite datasets [AMSR-E (VUA-NASA) and ERS-Scat] generally have a good correlation with the model predictions, while the AMSR-E (NSIDC) dataset did not correlate well with any of the other datasets. Overall, the AMSR-E (NSIDC) data showed a significant lack of seasonal soil moisture dynamics, which was well captured by the other datasets. These results suggest that the AMSR-E (NSIDC) dataset is not correct, as the three other independent models (a physically based radiative transfer model, an empiric soil moisture retrieval scheme, and a land surface model) show a good correlation with each other. This is further supported by the good correlation between SIM, AMSR-E (VUA–NASA), ERS-Scat and the in situ observations at SMOSREX. It is possible that those three models are all wrong and coincidentally produce the same results, although the comparison with SMOSREX suggests that this is not likely. The results of the observations obtained from the scatterometer additionally highlight the potential use of active microwave datasets, which will be continued by the MetOp–ASCAT observations.

The analysis of detrended time series (monthly anomalies) of surface soil moisture over the SMOSREX site shows that short-term variations of SIM and all the satellite products [including the AMSR-E (NSIDC) product] are meaningful. The significance is less for ERS-Scat, which has a high sampling time.

For the moment it has to be acknowledged that there exists a good correlation between some products for densely vegetated areas, but further studies are required to validate their physical meaning or relevance. Given that we present only the temporal dynamics in this paper, it is interesting to learn that some satellite products appear to represent those dynamics better than others—even for forested areas.

While in situ observations averaged to the land surface model or remotely sensed pixel scale may be better suited for the evaluation of both land surface models or radiative transfer models, these observations are still sparse and difficult to obtain. This study presents an alternative to the use of in situ observations for such large-scale evaluations through the intercomparison of independent and apparently similar soil moisture estimates from different models.

Finally, the good correlations between point observations and the low-resolution model predictions and satellite observations also show the importance of single-point observations for the verification of LSM and remotely sensed soil moisture products. They also support the need for the installation of new and the maintenance of existing soil moisture monitoring networks. This is particularly true for forested and mountainous regions, which in the past have been neglected when new soil moisture monitoring sites were established. With the need for the evaluation of land surface model performances and satellite validation campaigns, the relatively few existing networks are not sufficient.

Acknowledgments

The authors acknowledge the NSIDC for the AMSR-E (NSIDC) dataset. Eric Martin and Jean-François Mahfouf from the CNRM are thanked for providing access to the SIM database and for fruitful discussions. During the course of this study, Christoph Rüdiger was financed through a post-doctoral fellowship with the Centre National d’Etudes Spatiales (CNES). The SMOSREX project was co-funded by the Programme National de Télédétection Spatiale and the Programme Terre Océan Surface Continentales et Atmosphère (CNES), and by participants in the experiment—CESBIO (CNES, CNRS, IRD, and UPS), CNRM/GAME (Météo-France, CNRS), INRA, and ONERA—all in the framework of the SMOS science preparatory program.

REFERENCES

REFERENCES
Albergel
,
C.
, and
Coauthors
,
2008
:
From near-surface to root-zone soil moisture using an exponential filter: An assessment of the method based on in-situ observations and model simulations.
Hydrol. Earth Syst. Sci.
,
12
,
1323
1337
.
Bartalis
,
Z.
,
W.
Wagner
,
V.
Naeimi
,
S.
Hasenauer
,
K.
Scipal
,
H.
Bonekamp
,
J.
Figa
, and
C.
Anderson
,
2007
:
Initial soil moisture retrievals from the METOP-A Advanced Scatterometer (ASCAT).
Geophys. Res. Lett.
,
34
,
L20401
.
doi:10.1029/2007GL031088
.
Boone
,
A.
,
J-C.
Calvet
, and
J.
Noilhan
,
1999
:
Inclusion of a third soil layer in a land surface scheme using the force–restore method.
J. Appl. Meteor.
,
38
,
1611
1630
.
Bosch
,
D. D.
,
V.
Lakshmi
,
T. J.
Jackson
,
M.
Choi
, and
J. M.
Jacobs
,
2006
:
Large scale measurements of soil moisture for validation of remotely sensed data: Georgia soil moisture experiment of 2003.
J. Hydrol.
,
323
,
120
137
.
Calvet
,
J-C.
, and
J.
Noilhan
,
2000
:
From near-surface to root-zone soil moisture using year-around data.
J. Hydrometeor.
,
1
,
393
411
.
Calvet
,
J-C.
,
J.
Noilhan
, and
P.
Bessemoulin
,
1998a
:
Retrieving the root-zone soil moisture from surface soil moisture or temperature estimates: A feasibility study based on field measurements.
J. Appl. Meteor.
,
37
,
371
386
.
Calvet
,
J-C.
,
J.
Noilhan
,
J-L.
Roujean
,
P.
Bessemoulin
,
M.
Cabelguenne
,
A.
Olioso
, and
J-P.
Wigneron
,
1998b
:
An interactive vegetation SVAT model tested against data from six contrasting sites.
Agric. For. Meteor.
,
92
,
73
95
.
Calvet
,
J-C.
,
N.
Fritz
,
F.
Froissard
,
D.
Suquia
,
A.
Petitpa
, and
B.
Piguet
,
2007
:
In situ soil moisture observations for the CAL/VAL of SMOS: The SMOSMANIA network.
Proc. Int. Geoscience and Remote Sensing Symp., Barcelona, Spain, IGARSS, 1196–1199
.
de Rosnay
,
P.
, and
Coauthors
,
2006
:
SMOSREX: A long term field campaign experiment for soil moisture and land surface processes remote sensing.
Remote Sens. Environ.
,
102
,
377
389
.
Dirmeyer
,
P. A.
,
A. J.
Dolman
, and
N.
Sato
,
1999
:
The pilot phase of the Global Soil Wetness Project.
Bull. Amer. Meteor. Soc.
,
80
,
851
878
.
Draper
,
C. S.
,
J. P.
Walker
,
P. J.
Steinle
,
R. A. M.
de Jeu
, and
T. R. H.
Holmes
,
2007
:
Remotely sensed soil moisture over Australia from AMSR-E.
Proc. MODSIM 2007 Int. Congress on Modelling and Simulation, Christchurch, New Zealand, Modelling and Simulation Society of Australia and New Zealand. [Available online at http://www.mssanz.org.au/MODSIM07/papers/27_s28/RemotelySensed_s28_Drapper_.pdf.]
.
Drusch
,
M.
,
E. F.
Wood
, and
H.
Gao
,
2005
:
Observation operators for the direct assimilation of TRMM microwave imager retrieved soil moisture.
Geophys. Res. Lett.
,
32
,
L15403
.
doi:10.1029/2005GL023623
.
Georgakakos
,
K. P.
, and
T. M.
Carpenter
,
2006
:
Potential value of operationally available and spatially distributed ensemble soil water estimates for agriculture.
J. Hydrol.
,
328
,
177
191
.
Grayson
,
R. B.
, and
A. W.
Western
,
1998
:
Towards a real estimation of soil water content from point measurements: Time and space stability of mean response.
J. Hydrol.
,
207
,
68
82
.
Habets
,
F.
, and
Coauthors
,
2008
:
The SAFRAN-ISBA-MODCOU hydrometeorological model applied over France.
J. Geophys. Res.
,
113
,
D06113
.
doi:10.1029/2007JD008548
.
Kerr
,
Y. H.
, and
E. G.
Njoku
,
1990
:
A semiempirical model for interpreting microwave emission from semiarid land surfaces as seen from space.
IEEE Trans Geosci. Remote Sens.
,
28
,
384
393
.
Kerr
,
Y. H.
,
P.
Waldteufel
,
J-P.
Wigneron
,
J.
Font
, and
M.
Berger
,
2001
:
Soil moisture retrieval from space: The Soil Moisture and Ocean Salinity (SMOS) mission.
IEEE Trans Geosci. Remote Sens.
,
39
,
1729
1735
.
Li
,
L.
,
E. G.
Njoku
,
E.
Im
,
P. S.
Chang
, and
K.
St. Germain
,
2004
:
A preliminary survey of radio-frequency interference over the U.S. in aqua AMSR-E data.
IEEE Trans Geosci. Remote Sens.
,
42
,
380
390
.
Li
,
L.
,
P. W.
Gaiser
,
M. H.
Bettenhausen
, and
W.
Johnston
,
2006
:
WindSat radio-frequency interference signature and its identification over land and ocean.
IEEE Trans Geosci. Remote Sens.
,
44
,
530
539
.
Masson
,
V.
,
J-L.
Champeaux
,
F.
Chauvin
,
C.
Meriguet
, and
R.
Lacaze
,
2003
:
A global database of land surface parameters at 1-km resolution in meteorological and climate models.
J. Climate
,
16
,
1261
1282
.
Mätzler
,
C.
, and
A.
Standley
,
2000
:
Relief effects for passive microwave remote sensing.
Int. J. Remote Sens.
,
21
,
2403
2412
.
Meesters
,
A. G. C. A.
,
R. A. M.
De Jeu
, and
M.
Owe
,
2005
:
Analytical derivation of the vegetation optical depth from the microwave polarization difference index.
IEEE Geosci. Remote Sens. Lett.
,
2
,
121
123
.
Njoku
,
E.
,
cited
.
2004
:
AMSR-E/Aqua Daily L3 Surface Soil Moisture, B03 data set. National Snow and Ice Data Center, Boulder, CO, digital media.
[Available online at http://nsidc.org/data/amsre/.]
.
Njoku
,
E. G.
,
T. J.
Jackson
,
V.
Lakshmi
,
T. K.
Chan
, and
S. V.
Nghiem
,
2003
:
Soil moisture retrieval from AMSR-E.
IEEE Trans. Geosci. Remote Sens.
,
41
,
215
229
.
Noilhan
,
J.
, and
S.
Planton
,
1989
:
A simple parameterization of land surface processes for meteorological models.
Mon. Wea. Rev.
,
117
,
536
549
.
Noilhan
,
J.
, and
J-F.
Mahfouf
,
1996
:
The ISBA land surface parameterisation scheme.
Global Planet. Change
,
13
,
145
159
.
Owe
,
M.
,
R.
de Jeu
, and
J.
Walker
,
2001
:
A methodology for surface soil moisture and vegetation optical depth retrieval using the microwave polarization difference index.
IEEE Trans. Geosci. Remote Sens.
,
39
,
1643
1654
.
Owe
,
M.
,
R.
de Jeu
, and
T.
Holmes
,
2008
:
Multisensor historical climatology of satellite-derived global land surface moisture.
J. Geophys. Res.
,
113
,
F01002
.
doi:10.1029/2007JF000769
.
Pellarin
,
T.
,
J-C.
Calvet
, and
W.
Wagner
,
2006
:
Evaluation of ERS scatterometer soil moisture products over a half-degree region in southwestern France.
Geophys Res. Lett.
,
33
,
L17401
.
doi:10.1029/2006GL027231
.
Prigent
,
C.
,
F.
Aires
,
W. B.
Rossow
, and
A.
Robock
,
2005
:
Sensitivity of satellite microwave and infrared observations to soil moisture at a global scale: Relationship of satellite observations to in situ soil moisture measurements.
J. Geophys. Res.
,
110
,
D07110
.
doi:10.1029/2004JD005087
.
Quintana-Seguí
,
P.
, and
Coauthors
,
2008
:
Analysis of near-surface atmospheric variables: Validation of the SAFRAN analysis over France.
J. Appl. Meteor. Climatol.
,
47
,
92
107
.
Robock
,
A.
,
K. Y.
Vinnikov
,
G.
Srinivasan
,
J. K.
Entin
,
S. E.
Hollinger
,
N. A.
Speranskaya
,
S.
Liu
, and
A.
Namkhai
,
2000
:
The global soil moisture data bank.
Bull. Amer. Meteor. Soc.
,
81
,
1281
1299
.
Rüdiger
,
C.
, and
Coauthors
,
2007
:
Goulburn River experimental catchment data set.
Water Resour. Res.
,
43
,
W10403
.
doi:10.1029/2006WR005837
.
Sabater
,
J. M.
,
L.
Jarlan
,
J-C.
Calvet
,
F.
Bouyssel
, and
P.
De Rosnay
,
2007
:
From near-surface to root-zone soil moisture using different assimilation techniques.
J. Hydrometeor.
,
8
,
194
206
.
Settin
,
T.
,
G.
Botter
,
I.
Rodriguez-Iturbe
, and
A.
Rinaldo
,
2007
:
Numerical studies on soil moisture distributions in heterogeneous catchments.
Water Resour. Res.
,
43
,
W05425
.
doi:10.1029/2006WR005737
.
Teuling
,
A. J.
,
R.
Uijlenhoet
, and
P. A.
Troch
,
2005
:
On bimodality in warm season soil moisture observations.
Geophys. Res. Lett.
,
32
,
L13402
.
doi:10.1029/2005GL023223
.
Wagner
,
W.
,
G.
Lemoine
,
M.
Borgeaud
, and
H.
Rott
,
1999a
:
A study of vegetation cover effects on ERS scatterometer data.
IEEE Trans. Geosci. Remote Sens.
,
37
,
938
948
.
Wagner
,
W.
,
J.
Noll
,
M.
Borgeaud
, and
H.
Rott
,
1999b
:
Monitoring soil moisture over the Canadian prairies with the ERS scatterometer.
IEEE Trans. Geosci. Remote Sens.
,
37
,
206
216
.
Wagner
,
W.
,
G.
Lemoine
, and
H.
Rott
,
1999c
:
A method for estimating soil moisture from ERS scatterometer and soil data.
Remote Sens. Environ.
,
70
,
191
207
.
Wagner
,
W.
,
V.
Naeimi
,
K.
Scipal
,
R.
de Jeu
, and
J.
Martínez-Fernández
,
2007
:
Soil moisture from operational meteorological satellites.
Hydrogeol. J.
,
15
,
121
131
.

Footnotes

* Current affiliation: Department of Civil and Environmental Engineering, The University of Melbourne, Melbourne, Victoria, Australia

Corresponding author address: Jean-Christophe Calvet, Météo-France, CNRM/GMME/MC2, Toulouse 31000, France. Email: calvet@meteo.fr