Interactive comment on “ Surface soil moisture estimates from AMSR-E observations over an arid area , Northwest China ” by L

We would like to thank the referee for his or her thorough review and for all comments that will help us to improve the paper. The comments are particularly valuable given his or her deep knowledge of the subject of the paper. We will consider with particular care any additional suggestion that the referee may offer. We fully agree with all of the referee’s comments and will improve our paper according to them. A revised manuscript will be uploaded in the Supplement.


Introduction
Surface soil moisture is a key variable of water and energy exchanges at the land surface/atmosphere interface and thus has potential applications in hydrology, meteorology, and climate change studies.For climate applications, soil moisture is considered as the most important indicator of the effect of increase in greenhouse gases on the Introduction

Conclusions References
Tables Figures

Back Close
Full climate (Rowntree and Bolton, 1983).Real-time soil moisture measurements are vital for crop monitoring and drought warning.Previous researches have shown that passive microwave remote sensing sensors can be used to monitor soil moisture over land surfaces (Choudhury and Golus, 1988;Paloscia et al., 1993;Drusch et al., 2001;Jackson, 1993;Lakshmi et al., 1997;Njoku et al., 2005;Owe et al., 2007).Investigations have established the fundamentals of passive microwave remote sensing for monitoring temporal and spatial variations of regional soil moisture (Kerr and Njoku, 1990;Mo et al., 1982).Ground-based soil moisture network can not be built reasonably without consideration of huge cost.However, satellite observations are available in real-time and provide good temporal and spatial coverage.The effects of vegetation cover and surface roughness play a significant role in the microwave emission from the surface.Therefore, good parameterization schemes of these two effects are prerequisite for retrieving surface soil moisture information.Low frequencies are preferable for soil moisture estimates since perturbing factors such as atmospheric moisture, vegetation canopy, and surface roughness are less significant.Consequently, recent advances in science and technology have resulted in space agency commitments to L-band (1.4 GHz) missions in the near future (Kerr, 2001;Kerr et al., 2007).At present, there are several new satellite sensors operating at somewhat higher frequencies than L-band that show promise for soil moisture mapping under some conditions.The Advanced Microwave Scanning Radiometer (AMSR-E) is one of them which has potential for estimating surface soil moisture in regions with low level of vegetation.In the current investigation, an alternative retrieval technique utilizing the AMSR-E 6.9 GHz observations and based upon the work of Njoku and Li (2003) is implemented and tested for an arid area.Some typical techniques available for the estimation of surface soil moisture from the passive microwave observations are given below.Introduction

Conclusions References
Tables Figures

Back Close
Full Several significant examples of this algorithm have been performed by Jackson et al. (Jackson et al., 1995, 1999;Jackson and Le Vine, 1996), which are based on the radiative transfer equation and use a channel that is most sensitive to soil moisture.This algorithm incorporates soil texture, surface roughness and temperature using hpolarized brightness temperature and also includes correction for vegetation, and is then used with a dielectric model to obtain the soil moisture.

Multichannel iterative algorithm
This approach is successfully used for AMSR-E estimates to carry out three parameters simultaneously: surface soil moisture, soil temperature and vegetation water content (Njoku and Li, 1999;Njoku et al., 2003).However, unanticipated radio frequency interference (RFI) problem is encountered at the 6.9 GHz frequency (Njoku et al., 2005;Li et al., 2004), requiring modification of this approach.A radiative transfer model, the ω−τ model, is used for this algorithm.First, the higher frequencies in terms of 18 and 37 GHz are used for surface classification.Next, the ω−τ model is employed in comparing with the measured and the computed brightness temperatures at four channels (6.9 and 10.7 GHz, dual polarization).Finally, the three parameters are obtained until the difference between the computed and the observed brightness temperatures is minimized in a least squares sense.The principal of this algorithm is to discriminate between the three parameters considering the sensitivity differences between horizontal (H) and vertical (V) polarizations, and between frequencies.effects of surface temperature.This algorithm uses a nonlinear iterative procedure in a forward modeling approach to separate the surface emission from the vegetation emission, and then optimizes on the vegetation optical depth and the soil dielectric constant.
Once convergence between the computed and the observed brightness temperatures is achieved, the model uses a global database of soil physical properties together with a soil dielectric mixing model to solve for the surface soil moisture.Arid and semi-arid land accounts for 40% of the earth's surface.From climatological, ecological and anthropogenic perspectives, the spatio-temporal distributions of soil moisture over arid and semi-arid land are very important.Nowadays, climate change in arid areas is one of the main focuses in global warming which the international community is most concerned about.Desertification and arid ecosystem deterioration induced by arid climate variation and their impacts on the natural environment and anthropogenic activities have been given priority in recent years.
The objective of this study is to develop a specialized model for monitoring soil moisture suitable for an arid region in Northwestern China, Xinjiang.Our research into microwave land surface soil moisture for the study area here has been motivated by a need to better understanding the climate change and forecast modeling for this area.The study area is also interesting because of diversity of its surface types and complexity of its topography.So far, no such attempt has been made exclusively for this region from literatures.

Study area and data descriptions
The Xinjiang Uygur Autonomous Region of China is located in the hinterland of the Eurasian continent.Lying in Northwestern China (Fig. 1), Xinjiang covers a huge area of 1 660 000 km 2 with 49.5% mountains, 28% plains and 22.5% deserts, occupying one-sixth of China's total territorial area and larger than any other Chinese province or autonomous region.
The Xinjiang region is characterized by its unique topography with an extremely large In this paper, the AMSR-E brightness temperature (TB) observations are employed for estimating soil moisture.The AMSR-E instruments onboard the EOS/Aqua platform have been a component of the Defense Meteorological Satellite Program since June, 2002.The AMSR-E instruments are a conical scanning total power microwave radiometer system operating at a constant incidence angle of 54.8 • across a 1445-km swath at six frequencies: 6.9, 10. 7, 18.7, 23.8, 36.5, and 89 GHz.All the channels operate in both V and H polarization.Spatial resolution ranges from approximately 60 km at 6.9 GHz to 5 km at 89 GHz.The orbital period is about 102 min, which results in 14.1 orbits per day.For a given satellite, coverage is possible twice a day approximately 12 h apart on the ascending and descending passes.The orbit is sun-synchronous with equator crossings at about 01:30 and 13:30 LT.Additional information can be found in Njoku et al. (2003).
The AMSR-E measurements used here are AMSR-E/Aqua Daily Girded Brightness Temperatures available at the National Snow and Ice Data Center (NSIDC) (Knowles 2006).They are in the north EASE-Grid projection, with all channels having 25 km resolution and equidistant latitude-longitude projection at 0.25 jiang are used for validations.The data are available from the Xinjiang Climate Center of China.The stations, however, are most installed in the well-populated zones on the front plains of the mountains and the outskirts of the deserts (Fig. 1).
For further validation, we take advantage of soil moisture observations, consisting of 2 stations (Aletai: 88.08 • E, 47.73 • N and Shache: 77.27 • E, 38.43 • N) in the study area, for the period 2002-2004, to evaluate the estimated soil moisture from the AMSE.These observations were taken using the gravimetric technique for each 10-cm layer down to a depth of 0.5 m, The data were observed three times a month on the 8, 18 and 28th days, generally for the growing season (April to October).The data were originally recorded as percent wetness by mass of dry soil.Using soil density information available for each layer at each site, we determined the volumetric soil moisture.Only the data of the top layer (0-10 cm) are used for comparison purpose.
Additionally, soil moisture can not be estimated under snow cover or frozen soil conditions because the soil moisture dependence on the dielectric constant become small when surface soil temperature is below 0 • C (Hallikainen et al., 1985).Therefore, days with snow cover or frozen soil are eliminated from the present study.The period of the beginning of April to the end of October is, in general, historically free-frozen and snow-free for entire study area except for mountains.Thus, this period is taken into account for analysis, and hereinafter referred as to an "annual cycle".

Theoretical background and methodology
Passive microwave remote sensing is based on the measurement of thermal radiation from the land surface in the centimeter wave band.This radiation is largely determined by two components: physical temperature and emissivity of the underlying radiating body.Many land parameters have distinct properties that affect microwave emissivity in various ways.Dielectric properties and surface roughness may describe the properties.
The approach for this paper is based on the first-order radiative transfer theory.The influence of atmospheric moisture and the multiple scattering in the vegetation layer Introduction

Conclusions References
Tables Figures

Back Close
Full Screen / Esc Printer-friendly Version Interactive Discussion are assumed to be neglected.Furthermore, if the soil and vegetation temperatures are approximately equal and represented by a single surface temperature T S (K), the primary relationships between surface parameters and microwave brightness temperatures T b can be expressed (Kerr and Wigneron, 1995;Njoku et al., 2003;Mo et al., 1982): where ω is single scattering albedo of the vegetation, and the subscript p is polarization of the measurement, H or V, e s is emissivity of the soil surface which is related to the reflectivity r sp by and Γ is transmissivity of the vegetation and can be expressed as a function of the vegetation opacity along the observation path: where θ is the incidence angle, τ c is the vegetation opacity along the observation path which could be linearly related to the total vegetation water content (VWC) using the so called b P parameter (Jackson and Schmugge, 1991): Results of investigators have shown that the use of lower frequencies results in far less canopy attenuation of the soil emission.At lower frequencies, a value 0.15 of the parameter for b is appropriate for most agriculture crops.Methods for estimating the VWC have also been developed by passive microwave remote sensing.Investigators have explored relationships between the MPDI and vegetation parameters for a given vegetation type (Paloscia et al., 2001;Wen et al., 2005;Choudhury and Tucker, 1987).
The soil reflectivity at a given incidence angle is mainly related to soil moisture through the complex permittivity of the soil (Wang and Schmugge, 1980;Dobson et , 1985).However, in general case, surface roughness also affects the surface reflectivity because of scattering.To account for soil roughness, the h − Q formulation (Wang et al., 1983) is often used, and then the reflectivity of a rough surface, r sp , is related to that of an equivalent smooth surface, r op , as follows: where p and q denote orthogonal polarizations.The surface roughness parameter h is proportional to the quantity (ks) 2 , where k is the wave number (k=2π/λ) and s is standard deviation of the surface height (cm).The parameter Q contains information on both the vertical and the horizontal roughness correlation length, l .Since there are no data available to quantify the regional variability of h and Q at the AMSR-E frequencies and footprint scales, we calibrate h and Q empirically as discussed below.
In general, it is impossible to separate the integral contributions of surface roughness and vegetation layer unless one of them is known a priori.Investigators usually use a fixed value for roughness parameterization and assume the remainder of roughness can be automatically coupled with the vegetation effects (Owe et al., 2001;Njoku et al., 2003;Owe et al., 2007).
In this study, we use a numerical solution for the vegetation optical depth rather than use ancillary data or other surrogates (e.g.NDVI, LAI, and MPDI).The optical depth is directly derived from the MPDI and the dielectric constant of the soil, which improves the accuracy and overall efficiency of the soil moisture estimates.From Meesters et al. (2005), the numerical solution for the vegetation optical depth is: Then, according to Eq. ( 3) where a and d are defines as and The MPDI is defined as: The MPDI is frequently used to minimize or eliminate the effects of variable surface temperature dependence, leading to a close relationship to the dielectric properties of the emitting surfaces.The MPDI decreases with increasing vegetation water content because denser vegetation usually depolarizes the soil emission.For bare soil, the MPDI decreases with decreasing soil moisture because the soil emission is less polarized when the free water content in the soil is reduced.
The assumption that the vegetation single scattering has minimal effect on emissivity is widely accepted (Jackson et al., 1982).For simplicity, we neglect the effect of the single scattering, and then rewrite (1) by Eq. ( 5)-(10) as: This equation can be rework to the expression of the combined surface parameters of the surface roughness and the vegetation opacity, using Eq. ( 5) : In Eq. ( 11), there are three parameters unknown: the two parameters of surface roughness and the soil moisture that affects the soil emissivity.It is evident that if the Introduction

Conclusions References
Tables Figures

Back Close
Full values of surface roughness corresponding to Q and h are available, we can readily estimate the soil temperature.However, the ancillary data of surface roughness are sparse.For lower frequency (at 1.4 GHz), Q would have a minimum effect on the surface moisture calculations, while h also is usually used with a lower value.Therefore, when surface roughness conditions may be unknown, a value of zero is often assigned to Q, and a value between 0 and 0.3 is typically assumed for h (Jackson, 1993).At the 6.9 GHz frequency, Njoku and Chan (2006) used Q as a fixed global calibration factor leaving h to incorporate the roughness spatial variability.For the desert region in Saudi Arabia, with a global minimal topographic variation, a value of Q (0.174) was adopted as a fixed parameter to calibrate the AMSR-E data globally to the microwave model for a baseline minimum roughness level, assuming h=0.
The challenge for us is to separate these variables from the emitting signal by using a minimum of ancillary data.Apart from the brightness temperature data, we have only the ancillary data of soil texture.As previously mentioned, the study area is characterized by its extremely complex topography.Assuming the value of Q assigned as 0.174, if the value of h is assigned a fixed value (e.g., 0.2), the soil moisture estimates could not be made for many places of this study area except for desert regions.The reason for this explanation is that the value of h is uniform and probably much higher for the regions of complex topography.We have therefore realized using a uniform value of h for entire area is not reasonable.For clarity, the theoretical relationship between the MPDI and the vegetation optical depth is simulated with the radiative equation (Eq.10) before solving for this issue using a more unique approach (Fig. 2).
For bare soil, Fig. 2a shows the MPDI increases nonlinearly with increasing soil moisture, while for vegetated soil MPDI has a more complex relationship to soil moisture.The two effects of soil moisture and vegetation water content on the MPDI are synergistic in general.MPDI decreases when vegetation water content increases and/or soil moisture decreases.When vegetation water content increases the MPDI sensitivity to soil moisture decreases rapidly.Additionally, when MPDI is less than 0.01 or the vegetation optical depth is approximately greater than 0.8, the information Introduction

Conclusions References
Tables Figures

Back Close
Full of soil moisture is attenuated by high vegetation biomass and therefore can not be derived.Additionally, Fig. 2b depicts the relationship between the MPDI and the vegetation optical depth for a range of surface roughness (k=12).The two effects of surface roughness and vegetation water content on the MPDI are also interactive.It can be seen that sensitivity of the MPDI to soil moisture decreases nonlinearly very rapidly with surface roughness.This nonlinear attenuation effect of surface roughness is very similar to that of vegetation water content.Therefore, some investigators combined the two effects into one parameter to conduct temporal dynamics of vegetation water content (Njoku and Chan, 2006) and surface soil moisture (Jin and Yan, 2007).
As there are minimal human activities in the study area, the roughness is practically less disturbed for the AMSR-E footprint size.Once the roughness is estimated carefully, it may remain fairly invariable with time.First, we assumed the roughness is unchanging during an annual cycle.We further supposed the dries soil to have ∼5.5% volumetric soil moisture.The later assumption was probably somewhat higher for desert regions while lower for other regions.Note that the penetration depth at 6.9 GHz is only approximately 1.0-2.0cm (Jackson and Schmugge, 1989).There are no soil moisture measurements in this depth, although, it is self-evident that the soil moisture at this depth is in general lower than that at deeper depth (e.g. 5 cm).Njoku et al. (2006) speculated the value of volumetric soil moisture (VSM) in dry soil condition in the desert region of Al-Hadidah is round 5.0%.
For bare surface, the minimal VSM corresponds to the minimal MPDI (MPDI min ) from the AMSR-E observed over an annual cycle.The threshold for distinguishing between vegetated and bare soil surfaces using the MPDI is about 0.04 at 6.9 GHz (Wang et al., 2006).We assumed the values of the VSM for the driest bare soil in this study area are slight higher than that in region of Al-Hadidah, thus 5.5% was a reasonable value of VSM in dries surface soil condition.This baseline minimum value was used ultimately to compute h at each footprint for bare soil (MPDI min >0.04).As a consequence, total footprints for bare soil account for 62.3% in this study area.The others involved vegetated soil, frozen soil, snow cover, glacier etc. Figure 3

Conclusions References
Tables Figures

Back Close
Full Ignoring a very few abnormal negative values, the areas corresponding to the MPDI min not greater than 0.01 are just the glaciers lying on the mountains of Tianshan, Altay and Kunlun mountain each.The areas corresponding to the MPDI min ranging from 0.01 to 0.02 are usually characterized by perennial snow covers on mountains or some dense forests.As for the areas corresponding to the MPDI min ranging from 0.02 to 0.04, they indicate the ever presence of vegetation.Note throughout this paper that the areas ranging from 0.02 to 0.04 only indicate that there were covered by vegetations all the time or just at times.That is to say, there were less vegetations or bare soils covering the same areas at some or even long time for a given annual cycle.
Figure 4 shows the map of the h parameter in 2005 over the study area derived by Eq. ( 12).All the h patterns for other years are almost the same as this one.Note that the lower values of h correspond to the areas of dry, flat, and comparably homogeneous soil conditions, especially the Taklimakan Desert and the Gurbantunggut Desert.
The maximum value of h in bare soil areas is 0.73 and higher values of h mostly exists in the transition zones between deserts and mountains (purple color) where are characterized by great gradients of DEM.Considering the continuity of the h distribution and that the non-bare areas are mostly related to steep mountains, we predisposed empirically the value of the h parameter as 0.6.

Implementation
The previously described methodology has been applied to the historical dataset of nighttime AMSR-E brightness temperatures for the study area.A 7-year time series of estimated daily soil moisture dataset was obtained.Considering the greater stability of nighttime surface temperatures, only nighttime data were used in the study.Additionally, days with snow cover or frozen soil were also eliminated from the analysis, and Introduction

Conclusions References
Tables Figures

Back Close
Full Screen / Esc

Printer-friendly Version
Interactive Discussion therefore we just have selected the growing season (April to October).An iterative procedure was developed for the forward radiative transfer model that iteratively computed the simulated MPDI for VSM ranging from 5.5% to 45%, in 0.1% increments.The single-scattering albedo of the vegetation was set as 0.0 for both H and V polarization.At each iteration step, for 6.9 GHz, the computed MPDI values were compared to the satellite observations.If the absolute value of the differences between the computed and observed MPDI was less than 0.0015, then the input soil moisture would be saved in array.During the iteration process, the dielectric mixing model developed by Hallikainen et al. (1985) was used to convert VSM into soil dielectric constant.This mixing model requires percent of sand and clay as input, in addition to VSM.The soil texture classification of the surface soil on a 1-km grid for this study area was used here (Webb et al., 1991).For each AMSR-E footprint the soil texture properties were retrieved.

Mean annual soil moisture patterns
Figure 5 depicts the 7-year estimated mean annual soil moisture patterns.These patterns are similar to each other.In general, the mean annual soil moisture is around 9%.The lower values exist in desert regions especially in the Taklimakan Desert, and higher values are present on Northeastern Kunlun Mountain and in the transition zones between deserts and mountains, where are drainage regions.Significantly, the glaciers and/or snow covers distributed on the tops of mountains are automatically computed as no-value regions where the convergence between the calculated and the observed MPDI was no longer achieved because of the extremely low MPDI resulting from depolarization prosperities of glaciers and snow covers.
In order to further explore the dynamic characteristics of soil moisture, the year 2005 was selected for this investigation.For arid areas, actually, annual maximum soil moisture pattern reflects the soil moisture information resulting from a maximum (or accumulative) precipitation in a (or a series of) rainfall event(s).From Fig. 6, there are three obvious characteristics: 1) The desert regions (both the Taklimakan Desert and the Gurbantunggut Desert) are characterized by higher soil moisture values.For the Taklimakan Desert, the maximum soil moisture range is from 25% to 30% and for the Gurbantunggut Desert, the maximum soil moisture is 25-40%.Another area of higher soil moisture values shows on Northeastern Kunlun Mountain with values ranging of 30-40%.
2) The eastern part of the study area is characterized by low values with ranging of 6-8%, meaning little chance and few rainfall events took place in this region.3) On the transition zones between mountains and deserts are apparently high values, probably resulting from topographic effects on large precipitations.

Validations
We validate surface soil moisture products from the AMSR-E because the model for estimating soil moisture by utilizing formulations and with assumed or parameterized parameters have not been or are even impossible to be thoroughly developed and verified.Many of these procedures have been qualitative in nature.These limitations must result in uncertainties on the accuracy of the estimated surface soil moisture.However, there are no reliable and spatially averaged ground-measured surface moisture measurements for validation purposes, especially at satellite scale.In addition, although ground measurements are precise, each of them represent at only a specific point and hence can not be used to calculate the mean that the large satellite field of view (FOV) can accomplish, unless they are densely measured.Comparison with precipitation fields is often a validation option available.Although this method can not validate the estimated soil moisture quantitatively, it can at least provide qualitative trend information.To do this, we selected 6 meteorological stations for point validation.Introduction

Conclusions References
Tables Figures

Back Close
Full For each of the locations, matches between estimated soil moistures and antecedent measured daily precipitations were made.Furthermore, we selected typical a series of rainfall events in 2005 for regional validation.Finally, we compared the AMSR-E estimates of VSM and ground observations for two reprehensive sites, the Aletai site and Shache site, on the 8, 18 and 28th days over 2002-2004. 4.3 In general, the temporal patterns of precipitation and soil moisture for both bare and vegetated soil condition are characteristically coincident.The first three locations represent bare soil condition, particularly Tazhong just lies in the center of the Taklimakan Desert.The others represent vegetated soil condition.It can be seen that Tazhong is mostly associated with extremely low soil moisture condition (5.5% VSM) except for the rainy days when soil moisture increases up sensitively and afterwards decreases to original lever quickly.This must be a result of very high evaporative demand in this region.Qinghe and Shache show the characteristic behavior of soil moisture where it also increases up responding to rainfall event evidently.In addition, high dispersions of soil moisture are observed for these two locations.Highly variable surface temperature for the two locations could be responsible for this reason.On the other hand, Yutian, Hetian, and Mosuowan show lower dispersions compared to locations of bare soil, although soil moisture also responses to rainfall event evidently.However, after rainfall event, soil moisture gradually decreases.This lag is largely attributed to the retention capacity of soli moisture of vegetated soil.To broaden validating the soil moisture dataset, regional validation was conducted using precipitation observations over the study area.There are comparatively sparse meteorological stations in the huge area, although, a series of typical rainfall events was found.The events occurred on 4-6 August 2005 (DOY of 216-218), which were characterized in the IR images by a long belt of frontal cloud crossing the study area from north to south.High estimated soil moisture conditions were captured during the rainfall period.The frontal system traveled the Taklimakan Desert from west to east, producing light to moderate rainfall.Dry conditions (10% VSM) prevailed at the beginning of the rainfall events over the Taklimakan Desert.During the rainfall period, the Taklimakan Desert shows a series of rainfall traces with three sequential center regions of increased soil moisture conditions.The maximum values of estimated soil moisture reaches 22.0%, 29.2% and 37.8% for these three center regions, respectively.Note that the soil moisture conditions in these regions became rapidly dry after these meteorological events.

Comparison with in situ observations
Validation of the soil moisture retrieval algorithm here is based on comparisons of the AMSR-E estimates of VSM and ground observations for two sites, the Aletai site and Shache site, on the 8, 18 and 28th days over 2002-2004.Figure 9 shows comparisons of the measured soil moisture with the estimated soil moisture for the two sites.
Generally, the estimated soil moisture has temporal scale comparable to observations, but has small amplitude of variation.This small amplitude can be attributed to low sensitivity of soil moisture at FOV scale compared to that at point scale.In spite of some dispersion in data, higher correlation coefficients (R=0.649,0.604 for the two sites, respectively) and RMSE (3.5, 5.4%) are shown in Table 1.we can conclude that the algorithm is quite able to estimate the correct value of VSM. Figure 10 shows plots of estimated verse measured soil moisture for the two sites.Generally, the algorithm Introduction

Conclusions References
Tables Figures

Back Close
Full overestimates the soil moisture amount on the dry condition (below 14%) and, on the contrary, underestimates the soil moisture amount on the wet condition (above 14%), with positive average differences on the order of 2%.These results clearly demonstrate that the estimated VSM from the AMSR-E is consistent with the ground observations, which strongly supports the developed retrieval algorithm.

Conclusion and discussions
A model for estimating surface soil moisture over the arid area of Xinjiang, Northwest China, from the AMSR-E brightness temperatures at 6.9 GHz was developed.This model was based on first-order radiative transfer model calculation and applied to conduct 7-year period (2002)(2003)(2004)(2005)(2006)(2007)(2008) of soil moisture dataset.Finally, we compared the estimated soil moisture with the measured precipitation.Results clearly demonstrated the performance of the model for soil moisture estimates in the study area.The temporal patterns of rainfall and soil moisture for both bare and vegetated soil condition were characteristically coincident and soil moisture also responsed to rainfall event evidently, especially for bare soil condition.Furthermore, we captured a series of rainfall traces over the Taklimakan Desert.To quantitatively analyse the accuracy of the product derived from the AMSR-E, comparisons of the estimated soil moisture with the ground observations were also made for two representative sites.The results indicated the estimated soil moisture followed well these ground observations at temporal scale.Moreover, higher correlation coefficients (R=0.649,0.604) were obtained, indicating there has a good agreement between them and the soil moisture product derived from the AMSR-E is realistic and acceptable.
Two important model parameters were calibrated a priori.One was vegetation optical depth which was calibrated using an excellent numerical solution which made it possible to separate the effects of surface roughness and vegetation, especially over bare soil regions.The other was surface roughness which was fine-tuned with elab-Introduction

Conclusions References
Tables Figures

Back Close
Full orate assumptions for bare soil and vegetated soil, respectively.The later is worth further discussing as follows.Surface roughness parameterization is needed prior to surface soil moisture estimates, particularly in complex topographic region.We have supposed the soil moisture corresponds 5.5% for the dries bare soil.Additionally, for vegetated soil, we empirically assigned the parameter h with a uniform value 0.6.These assumptions allow possible and even reliable h parameterization, which has not been possible until now in this study area.However, these guesses would be not right for some footprints.On the other hand, the previous assumption of a value of zero for the single scattering for various vegetations is also unrealistic.In practice, to some extent, the higher assigned h value can offset the effect of underestimation of the single scattering, and then improve the quality of soil moisture estimates.
In addition, the assumed minimum value of 5.5% VSM, for example, must be some higher for some desert soils.∼3% may be a realistic guess.However, this input could lead to a negative h value.Volume scatting can account for this issue.This uncertainty in the h parameterization will propagate into the retrieval algorithm as part of the soil moisture biases, although, the present study shows progress in the characterization of the surface roughness, and proves potential for surface moisture estimates.
Moreover, the model is mathematically well behaved, so convergence is normally fast and the glaciers and/or snow covers distributed on the tops of mountains can be computed as no-value regions automatically.The estimated soil moisture dataset now has been made available for climate analysis and forecast modeling for the study area.

1. 3
Polarization index algorithm Owe et al. (2007) has shown the potential of this algorithm for developing a historical climatology of continuous satellite-derived global land surface soil moisture.The Microwave Polarization Difference Index (MPDI) can effectively eliminate or minimize the Screen / Esc Printer-friendly Version Interactive Discussion of the MPDI min for 2005, clearly distinguishing the bare soils from other surface types.

Fig. 1 .
Fig. 1.Map of the study area combined with digital elevation model (DEM) and distribution of the meteorological obviation network.

Fig. 1 .Fig. 2 .
Fig. 1.Map of the study area combined with digital elevation model (DEM) and distribution of the meteorological obviation network.

Fig. 5 .Fig. 7 .
Fig. 5. Patterns of mean annual surface soil moisture estimated from the AMSR-E at night orbit for years from 2002 to 2008.

Fig. 7 .
Fig. 7. Comparisons of estimated soil moisture and measured precipitation at six reprehensive locations.Left column are representative to bare soil condition (from top to bottom, corresponding to locations Tazhong, Qinghe, Shache, and right column to vegetated soil condition for Yutian, Hetian, and Mosuowan.
• C for the north and the south of Xinjiang, respectively.The rainfall is concentrated mainly in the summer season, with annual precipitation of about 146 mm.
.1 Temporal variability of estimated soil moisture Six stations were selected as the representatives for different surface types (vegetated surface, desert surface, bare arid surface) and climatic characteristics (the north and the south climate, separately).Figure7shows time series of estimated VSM and antecedent observed precipitation in 2005 for six representative locations -Tazhong • E, 36.85 • N), Hetian (79.93 • E, 36.85 • N), and Mosuowan (86.10 • E, 45.02 • N).

Table 1 .
Statistics of estimated and measured VSM for the Shache site and the Aletai site over2002-2004.