LAND surface processes strongly affect both large- and

Transcription

1 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST Development of a Satellite Land Data Assimilation System Coupled With a Mesoscale Model in the Tibetan Plateau Mohamed Rasmy, Toshio Koike, Souhail Boussetta, Hui Lu, and Xin Li Abstract Soil moisture is the central focus of land surface and atmospheric modeling because it controls surface water and energy fluxes and consequently affects land atmosphere interactions. Although global or regional satellite-derived surface soil moisture data sets are readily available, knowledge about assimilating them into numerical weather prediction (NWP) models is limited. The methods of assimilating soil moisture products in NWP models have several limitations, and they cannot be applied in near-real-time applications. As a result, this paper focuses on the development of a system [a Land Data Assimilation System coupled with a mesoscale Atmospheric model (LDAS-A)] that couples satellite land data assimilation with a mesoscale model to physically introduce land surface heterogeneities into the mesoscale model. The LDAS-A consists of a sequential LDAS that directly assimilates the lower frequency passive microwave brightness temperatures, and therefore, its use is feasible for near-real-time NWP applications. The LDAS-A was validated for the Tibetan Plateau using surface, radiosonde, and satellite observations. The simulation results show that the LDAS-A effectively improved the land surface variables (i.e., surface soil moisture and skin temperature) compared with the no-assimilation case and that it has the potential to correct uncertainties resulting from model initialization, model-specific parameters, and model forcing on a wider scale. The improved land surface conditions in the LDAS-A improve the land atmosphere feedback mechanism, and the assimilated results provide better prediction of atmospheric profiles (i.e., potential temperature and specific humidity) than the no-assimilation case when compared with radiosonde soundings. Improvements in solar radiation, in addition to soil moisture, are necessary to introduce realistic land atmosphere interactions into a mesoscale model. Index Terms Atmospheric modeling, data assimilation, land atmosphere interaction, microwave remote sensing, soil moisture. I. INTRODUCTION LAND surface processes strongly affect both large- and small-scale circulations and convection processes and control the dynamics and thermodynamics of the atmosphere Manuscript received May 6, 2010; revised December 2, 2010 and January 19, 2011; accepted January 30, Date of publication March 22, 2011; date of current version July 22, M. Rasmy and T. Koike are with the River and Environmental Engineering Laboratory, Department of Civil Engineering, The University of Tokyo, Tokyo , Japan ( rasmy@hydra.t.u-tokyo.ac.jp; tkoike@ hydra.t.u-tokyo.ac.jp). H. Lu is with the Center for Earth System Science, Tsinghua University, Beijing , China ( luhui@tsinghua.edu.cn). S. Boussetta is with the European Centre for Medium-Range Weather Forecasts, RG2 9AX Reading, U.K. ( Souhail.Boussetta@ecmwf.int). X. Li is with the Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Science, Lanzhou , China ( lixin@lzb.ac.cn). Digital Object Identifier /TGRS through the exchange of moisture and heat fluxes [2], [36]. The importance of land surface processes is widely recognized in weather and climate modeling because uncertainties in land surface process modeling might greatly affect weather and climate forecasts. Land surface models (LSMs) that aim to describe the most important processes governing the moisture, heat, and momentum exchanges between the land surface and planetary boundary of the atmosphere have remarkably improved in recent years and have been successfully operated within numerical weather prediction (NWP) models [1], [15], [42], [54]. Soil moisture, which governs most land surface processes, is a key element in land surface modeling, and it plays a vital role in global water and energy cycles by dominating the partitioning of the surface energy into surface turbulent fluxes that, in turn, drive the land atmosphere interactions. A number of sensitivity studies on NWP have shown that soil moisture can influence short- and medium-range forecasts [12], [19] and that consideration of soil moisture can enhance seasonal predictability [7], [31]. In addition, the effects of soil moisture are most noteworthy in mid-continental regions with transitional climates [29], and soil moisture controls the initialization and dynamics of mesoscale convective systems [3], [57]. Although the necessity to realistically initialize soil moisture as a lower boundary condition in NWP models has been widely emphasized, global or regional soil moisture monitoring is complicated, owing to high spatial and temporal variability. As a result, the available observation network poorly represents the spatial distributions, or data are entirely absent for most parts of the globe [51]. Therefore, model initializations rely on proxy observations that are indirectly linked to the soil moisture information, e.g., soil moisture initializations based on forecast errors of screen-level parameters such as (2 m) temperature and relative humidity [12], [23], [39]. However, this method has serious limitations, e.g., screen-level observations are scarce, the error relationships are more closely linked to surface turbulent fluxes than to actual soil variables, several empirical thresholds are assumed in the analyses, and they cannot be sensibly applied to situations where the soil moistureatmospheric boundary layer feedback is weak (e.g., situations of strong advection, weak solar forcing, and high cloudiness) [10], [12], [39]. The aforementioned shortcomings can possibly be overcome by assimilating remotely sensed more direct nearsurface soil moisture information into the NWP models [11], [24], [53]. Spaceborne remote sensing, especially passive microwave remote sensing [such as that carried out by the Advanced Microwave Scanning Radiometer on NASA s Earth Observing System (AMSR-E)], provides spatially integrated information /$ IEEE

2 2848 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011 on soil moisture over large areas with a frequent coverage that is useful for the initialization of the NWP models. The microwave region is the only part of the electromagnetic spectrum that permits quantitative estimation of soil moisture based on a large contrast between the dielectric constant of dry soil ( 4) and that of water ( 80), and the relationship between microwave emission and soil moisture has been extensively investigated in recent years by many researchers [14], [16], [27], [32], [41]. Moreover, satellite measurements of surface soil moisture are subject to errors, they cannot provide a complete space-time coverage, and they are limited to a few centimeters of soil depth, whereas the NWP models require moisture information for model sublayers. Data assimilation methods merge the satellite surface information within an LSM and propagate it into deeper soil layers, thereby providing spatially and temporally complete information of superior products such as the vertical profiles of soil moisture, temperature, and turbulent heat fluxes that are required for the initialization of weather and climate models and are otherwise unattainable using other methods [50]. Although there have been many coordinated and institutional attempts to produce regional and global soil moisture information [8], [17], [26], [28], [40], [43], [45], there are systematic differences among soil moisture products and between the products and model fields that are attributed to model-specific parameters such as porosity, hydraulic conductivity, wilting point, layer depth, dynamic range defined by the specific evaporation, and runoff formulations and deficiencies in radiative transfer formulation in describing surface roughness, vegetation, and volume scattering effect [13], [30], [47]. Moreover, knowledge relating to the assimilation of satellitederived soil moisture observations into the NWP models is still limited, and a few studies have been reported. More recently, a nudging scheme combined with a cumulative distribution function (CDF) [30] was introduced to better utilize the satellitederived soil moisture products in the NWP model, and the results showed that the assimilation increases the accuracy of the surface fields and marginally improves the low-level (2 m) air temperature and humidity [11], [53]. However, this method is not applicable in operational near-real-time applications for several reasons: 1) there are difficulties in defining an observation operator that changes spatially and temporarily; 2) the nudging procedures assume that observations are very accurate and require the specification of several parameters (e.g., a nudging coefficient and relaxation time window) that are somewhat arbitrary; and 3) the reanalysis increments are computed once per day and are distributed over a 24-h period, which makes it impossible to correct soil moisture when the NWP model predicts subsequent rainfall events during this 24-h time window [11]. Because of these limitations, the necessity of an advanced data assimilation system has been emphasized for the more effective use of satellite observations in NWP applications [53]. In this paper, we developed a physically based approach [a Land Data Assimilation System coupled with a mesoscale Atmospheric model (LDAS-A)] that can overcome the aforementioned limitations by directly assimilating the lower frequency microwave brightness temperatures within the NWP model rather than using the soil moisture products [24]. This approach removes requirements such as the preprocessing of soil moistures products, CDF matching, and nudging processes because the assimilation is performed online and it uses the calibrated level 1B brightness temperature observation. An initial study that tested the potential of online land data assimilation with a mesoscale model was reported in [3], which implemented a simple modeling framework that employed a variational data assimilation approach with a semiempirical radiative transfer model (RTM) that has a single parameter set for the entire model domain. The present study builds on the previous study [3] by implementing several improvements such as the inclusion of the freezing/thawing land surface processes in the LSM, improvement of the radiative transfer processes in soil (surface and volume scattering), introduction of an assimilation algorithm that is suitable for operational near-real-time applications, and implementation of a standardized and robust interface (coupler), which consists of a superstructure that effectively handles the coupling and exchange of data between individual components of the system (the atmospheric model, LSM, and land data assimilation algorithm) using parallel computing techniques. The LDAS-A uses the physically based microwave RTM as an observation operator, an LSM as a model operator, and a sequential assimilation technique [ensemble Kalman filter (EnKF)] as an assimilation algorithm. The EnKF is capable of resolving the nonlinearity and discontinuity within the model and observation operators, and it updates the land surface states whenever observations are available. The EnKF algorithm applies an ensemble of model states to represent the error statistics of the model estimation and to predict the error statistics continuously in time, and it has proven to be an efficient approach for soil moisture estimation [25], [48], [49]. Furthermore, to quantitatively understand the importance of the land atmosphere feedback mechanism, in addition to ground observations, the present study uses intensive radiosonde soundings and satellite observations for model validation, which is essential in system development and is lacking in recent studies such as [10], [11], [39], and [53]. This paper is organized as follows. In Section II, we first present the development, major components, and methodology of the system framework. The data sets and experimental descriptions are discussed in Section III and Section IV, respectively. The simulation results are analyzed in Section V, and a summary and the conclusion are given in Section VI. II. SYSTEM DEVELOPMENT In this paper, a land atmosphere mesoscale model is coupled to a sequential LDAS that integrates passive microwave brightness temperatures within the model framework. The new system is referred to as the LDAS-A, and its framework is shown in Fig. 1. As shown in the figure, the LDAS-A consists of a mesoscale atmospheric model, an LSM (SiB2) that acts as a land surface driver for the atmospheric model and as a model operator for the assimilation system, a microwave RTM as an observation operator, and an EnKF as an assimilation algorithm. The land atmosphere mesoscale model is initialized with the initial and boundary conditions derived from global analysis. During the land surface integration period, the atmospheric model passes the forcing to the LSM. The LSM runs in the ensemble mode and produces an ensemble of forecasts for the soil moisture and temperature profiles, and heat and moisture fluxes, and then, it feeds back to the atmospheric model to guarantee

3 RASMY et al.: DEVELOPMENT OF A SATELLITE LDAS COUPLED WITH A MESOSCALE MODEL 2849 Fig. 1. Framework of the LDAS coupled with an atmospheric model (LDAS-A). the land atmosphere feedback mechanism. When observations are available from the satellite measurements, improved land surface conditions are computed by the assimilation procedures and are updated in the land atmospheric model to improve atmospheric evolution. The system components are explained in the following sections. A. Atmospheric Model In this paper, a regional- to storm-scale prediction system known as the Advanced Regional Prediction System (ARPS), which was developed at the Center for Analysis and Prediction of Storms, University of Oklahoma, was chosen as the atmospheric driver. The atmospheric prediction component of the ARPS is a 3-D nonhydrostatic compressible model formulated in generalized terrain-following coordinates. The forward components of the ARPS included advanced parameterizations for radiation, turbulence, and cloud microphysics processes. The model is governed by momentum, thermodynamic, continuity, and transport equations of moisture categories and a subgrid scale turbulent kinetic energy (TKE) model. These equations are transformed from the physical domain to the computational domain using terrain-following coordinates and grid stretching in the vertical direction. The ARPS also includes a very simplified two-layer snow-free nonfrozen LSM [42] that needs to be updated to accurately describe the land surface processes. The formulation of the ARPS model and the validation and applications of the results are described in [61] and [62]. B. LDAS The LDAS was developed to improve the surface moisture heterogeneity through the assimilation of lower frequency satellite passive microwave observations such as those at 6.9 and GHz. The system consists of a land surface scheme, an RTM, and an assimilation algorithm. Information on the soil moisture retrieved by satellite is limited to a few centimeters of soil depth, while mesoscale models require moisture information for all model soil sublayers to incorporate a land atmosphere feedback mechanism realistically. The LDAS overcomes this problem by integrating the retrieved surface soil moisture from the satellite observation within a land surface scheme and provides realistic and physically consistent soil moistures profiles for atmospheric models. The components of the LDAS are briefly explained in the following sections. 1) Land Surface Scheme (Model Operator): Considering the physically based formulation and accuracy of estimating soil moisture, surface temperature, and heat fluxes on the surface and within deep soil layers, Simple Biosphere model version 2 (SiB2) was chosen as an alternative to the original ARPS LSM as well as the model operator for the LDAS. SiB2 was primarily designed to be used within atmospheric models, and its parameterization and grid resolution can be considered consistent with the purpose of our study. The SiB2 model includes three soil layers: a surface soil layer of a few centimeters (0 5 cm), which acts as a significant source of direct evaporation when moist; a root zone (5 20 cm), which is the supplier of soil moisture to the roots and which accounts for transpiration; and a deep soil layer, which acts as a source for hydrological base flow and upward recharge of the root zone. The complete formulation of the SiB2 can be found in [54]. For the sake of completeness, a brief summary of the soil moisture governing equations of SiB2 is given in the following. The soil water flow based on the three-layer finite difference method is governed by the following balance equations. The surface layer soil moisture W 1 is given by W 1 t = 1 ( Q 1 Q 12 E ) g. (1) θ s D 1 ρ w The root layer soil moisture W 2 is given by W 2 t = 1 ( Q 12 Q 23 E ) ct. (2) θ s D 2 ρ w

4 2850 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011 The deep layer soil moisture W 3 is given by W 3 t = 1 θ s D 3 (Q 23 Q 3 ). (3) Here, Q 1 is the infiltration of precipitation into the first soil moisture layer (in meters per second), Q 3 is the gravitational drainage from the third layer (in meters per second), Q ij is the water flow from the ith layer to the jth layer, E g is the evaporation rate from the ground surface layer, E ct is the canopy transpiration rate through the stomata (in kilograms per square meter second), D i is the depth of the ith soil layer (in meters), θ s is the volumetric water content at saturation (in cubic meters per cubic second), and ρ w is the density of water (in kilograms per cubic meter). SiB2 does not consider the freezing and thawing processes over time during the freeze/thaw cycles. This water phase transition significantly affects the soil moisture content as well as evaporation, canopy transpiration,and thermal and hydraulic properties of the soil. A modified approximation of the Stefan solution was proposed in [34] and was incorporated in SiB2 to calculate the frost/thaw depth over time and to estimate the soil moisture and temperature profiles during the freeze/thaw cycle. 2) RTM (Observation Operator): A passive microwave sensor does not directly measure soil moisture, but it measures emitted energy or brightness temperature, and the observation operator (RTM) provides the link between soil moisture and brightness temperature. For microwaves, a land surface radiative transfer algorithm is based on the large contrast between the dielectric constants of dry soil ( 4) and water ( 80) and the resulting dielectric properties of the soil water mixture and their effects on natural microwave emission. By neglecting the atmospheric and rainfall effects and by assuming that the reflection at the surface is much less than the radiation from the surface and vegetation layer at lower frequency, the estimated brightness temperature at the satellite level is given in [20]. The following equation gives the total emission and attenuation from the land surface and vegetation canopy: T b = T bs exp( τ c )+[(1 ω c )(1 exp( τ c )T c ]. (4) Here, T b is the brightness temperature at the satellite level (in kelvin), T bs is the brightness temperature at the surface or ground level (in kelvin), ω c is the single-scattering albedo of the canopy, τ c is the vegetation optical thickness, and T c is the canopy temperature (in kelvin). In the absence of vegetation, (4) can be simplified as T b = T bs =(1 R p )T s (5) where (1 R p ) is the surface emissivity, R p is the surface reflectivity, and T s is the surface physical temperature (in kelvin). In the passive microwave remote sensing of the Earth s terrain, the scattering effects due to the rough interface and soil heterogeneity play a dominant role in the determination of the brightness temperature. The calculation of the surface emissivity critically depends on the surface moisture conditions and the degree of roughness. When the land surface is smooth, Fresnel power reflectivity is used to describe the land surface reflectivity. The horizontally (R h ) and vertically (R v ) polarized Fresnel power reflectivities are calculated as cos θ ɛ r sin 2 θ 2 R h = cos θ + (6) ɛ r sin 2 θ ɛ r cos θ ɛ r sin 2 θ 2 R v = ɛ r cos θ +. (7) ɛ r sin 2 θ Here, ɛ r is the dielectric constant of the soil water mixture, which is given in [9] ɛ r = [ 1+(1 ω s )(ɛ α s 1) + ω β ɛ α f ω ω ] 1/α (8) where ω s is the soil porosity, ω is the soil water content, ɛ s is the dry soil dielectric constant, ɛ fω is the dielectric constant of free water, θ is the incident angle, α =0.65, and β is a coefficient that is dependent on the soil texture. β is calculated from the soil texture as [58] β = %sand %clay. (9) The surface roughness is important, and it should be addressed correctly [32], [44]. Many RTMs employ the semiempirical model proposed in [6] and [59] to account for the roughness effect in the RTM. However, the method has a limited range of applicability for a given set of parameters, it is poorly suited to higher frequencies, and the parameters of the semiempirical models have weak physical descriptions and cannot be directly compared with the observation values [55]. It has been shown that the advanced integral equation model (AIEM) with incorporation of a shadowing effect is in good agreement with the observed data, and it is most suitable among the available physically based models, such as the small perturbation model, physical optics model, and geometric optics model, and is applicable over a wide range of frequencies [5], [32]. Therefore, the modified AIEM is considered to represent the surface emission under varying roughness conditions. The reflectivity of a randomly rough surface for p(= vertical or horizontal) polarization Rp e can be determined from the total power of scattering from the boundary in terms of the scattering coefficient, with the inclusion of the shadowing effect [32] Rp e = R p exp [ (2 kσ cos θ) 2] 1 S(θ, θ)+ 4π cos θ 2π π/2 [σ pp (θ, θ j,φ j ) S(θ, θ j )+σ pq (θ, θ j,φ j ) 0 0 S(θ, θ j )] sin θ j dθ j dφ j (10) where q and p refer to the vertical or horizontal polarization states, j is the scattering direction, R is the Fresnel reflectivity, k is the wavenumber, σ is the root-mean-square height, and S is the shadowing function. In nature, soil media are heterogeneous, and the region in the heterogeneous medium that affects volume scattering can be ascertained from the depth of penetration, which is a function of the dielectric constant (soil texture and moisture content) and frequency. For very dry soil, the effects of volume scattering are dominant, and as [56] pointed out, the soil moisture retrieved over a very dry or desert area is greatly overestimated by the available RTM. To overcome this issue, the method proposed

5 RASMY et al.: DEVELOPMENT OF A SATELLITE LDAS COUPLED WITH A MESOSCALE MODEL 2851 Fig. 2. Schematic diagram of the coupler and its link with the system components. in [37] and [38] was adopted in our study. In this case, the soil medium is treated as a multilayer structure to facilitate the simulation of the radiative transfer process that takes place in the soil medium and involves the correlated multiple scattering. The soil is treated as a mixture of many densely packed scatterers and a host medium. The correlated scattering (multiple scattering) effect of the densely packed particles was modeled by introducing the so-called dense-media radiative transfer theory under the quasi crystalline approximation with coherent potential [33]. 3) EnKF (Assimilation Algorithm): The EnKF was introduced in [18] as an alternative to the traditional Kalman filter. It is a sequential estimation procedure for nonlinear dynamical systems, which update the system whenever observations are available. The EnKF represents the distribution of the system state with a collection of state vectors, called an ensemble, and replaces the covariance matrix with the sample covariance computed from the ensemble. The ensemble is operated as if it is a random sample, but the ensemble members are not really independent, and the EnKF ties them together. The EnKF assumes that all probability distributions involved are Gaussian, and therefore, it is computationally inexpensive compared with the particle filter. The EnKF was originally developed for atmospheric data assimilation. In the field of hydrology, [48] and [25] applied the EnKF to soil moisture estimation, and it was found that it performed well against the variational assimilation method. In our model development, we adopted the EnKF as an assimilation scheme. An overview of the EnKF is given in the following. Consider X =[w 1,w 2,w 3 ] T as a state variable and the first guess, where w 1, w 2, and w 3 are the soil moisture contents of the surface, root, and deep soil layers, respectively. This first guess is used to create an ensemble of size (N) by adding pseudorandom noise with known statistics. By dropping the time notation, each member of state variable X i is given by In the forecast step, the forecast state member X f i is determined from the nearest analysis state member Xi a according to X f i = M (Xi a )+u i u i N(0,Q) (12) where M is the model operator and u i is the model error vector of each member, which is obtained from a multivariate Gaussian distribution with zero mean and error covariance matrix Q. In the reanalysis step, the observation data are perturbed by adding a random observation error, and each member of the analyzed state variable Xi a is updated as [25] ( )] Xi a =X f i [(Y + K o + v i ) H X f i v i N(0,R) (13) where K is a Kalman gain matrix, H is the observation operator, Y o is the observation, R is the observation error covariance, and v i is a random error vector of the observation with zero mean and covariance matrix R. 4) Coupler: With the aim to establish a standard system with optional plug-ins and to test the combination of different models and data sources in the future, the coupler was based on object oriented programming (OOP). The OOP technology allows the addition of a new plug-in to the coupler without major modification. The coupler is modular, extensible, and enabled on the parallel computing technology to satisfy the increasingly high-performance computing requirements of the operational NWP models. It has two integrations loops. The first one is time integration, where the atmospheric model is integrated forward in time. When the model reaches the grid integration (second integration loop), which is nested inside the time integration, each grid calls its own combination of the LSM, RTM, and assimilation technique in run time. In this way, the coupler effectively handles access to the model components and data transfer from one model to another. Fig. 2 shows a schematic diagram of the coupler and its link with system components. X i = X + e i e i (i =1, 2...N) N(0,P) (11) where e i is the random error vector of each member and is obtained from a multivariate Gaussian distribution with zero mean and error covariance matrix P and X is the expectation of the first guess X. A. Data Sets III. DATA SETS AND METHODS The following sections briefly explain the data sets used within the newly developed system.

6 2852 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST ) AMSR-E Brightness Temperature Data: AMSR-E is a six-frequency dual-polarized total-power passive microwave radiometer that detects microwave emission from the Earth s surface and atmosphere. It measures horizontally and vertically polarized brightness temperatures separately at 6.925, 10.65, 18.7, 23.8, 36.5, and 89.0 GHz. It maintains a constant Earth incidence angle of 55, covering a swath of 1445 km on the Earth s surface and having a spatial resolution of an individual measurement varying from 5.4 km at 89.0 GHz to 56 km at 6.9 GHz. AMSR-E rotates continuously about an axis parallel to the local spacecraft vertical at 40 r/min. During a period of 1.5 s (one revolution), the spacecraft subsatellite point travels about 10 km. As a result, even though the instantaneous fieldof-view is different for each channel, the active scene measurements are recorded at equal intervals of 10 km (5 km for the 89-GHz channels) along the scan. This paper employs vertical polarization of and GHz to retrieve the soil moisture heterogeneities because the atmosphere can be considered transparent at these frequencies and the vertical polarizations of these frequencies are less sensitive than the horizontal polarizations of the same frequencies to surface roughness and vegetation coverage. Although the instantaneous observations at and GHz have a low spatial resolution ( 40 km), this paper employs the low-resolution data to update the land surface states at finer resolutions (i.e., one footprint of the satellite observation represents several grids in the mesoscale model) to investigate its applicability. The nearest neighborhood interpolation method was used to resample the active scene data to our domain resolution ( 5 km). The calibrated brightness temperature (Level 1B) data were obtained from the Japan Aerospace Exploration Agency. 2) MTSAT-1R Data: MTSAT-1R is a geostationary satellite located at 140 E, and it records images in five wavebands such as those of visible (VIS; μm), infrared channel 1 (IR1; μm), infrared channel 2 (IR2; μm), infrared channel 3 (IR3; μm), and infrared channel 4 (IR4; μm). The image resolutions for the visible and infrared (IR 1 4) channels at nadir are 1 and 4 km, respectively. To investigate the observed cloud activity over the selected domain, split-window IR1 (MTSAT/1R1) observations were obtained since no other direct measurements were available. The MTSAT/1R1 brightness temperature observations are the measurements of the cloud-top temperatures under the assumption that the clouds are blackbodies. In addition, lower brightness temperatures in an infrared image indicate sufficiently thick clouds, with the cloud tops radiating at the atmospheric temperature at higher altitudes. 3) Terra MODIS Land Surface Temperature Data: The Terra MODerate-resolution Imaging Spectroradiometer (MODIS) views the entire Earth s surface every 1 to 2 days and acquires data in 36 spectral bands in wavelengths ranging from 0.4 to 14.4 μm and at varying spatial resolutions ranging from 250 m to 1 km. The Terra/MODIS land surface temperature products provide per-pixel temperature values in a sequence of swath- to grid-based global products. In this paper, the daily level 3 (L3) global products (MOD11C1) configured on a 0.05 latitude/longitude climate modeling grid were selected for model validation. 4) In Situ Data: The automated weather station installed at the Gaize station includes a system for measuring soil moisture and temperature, a precipitation gauge, and a pyranometer (MS-802). It works continuously to collect data at a temporal resolution of 1 h. The soil moisture content is measured with a Trime EZ sensor at depths of 3, 20, and 40 cm, and the soil temperature is measured with a Pt100 resistance thermometer at depths of 0, 5, 10, 20, 40, and 80 cm. In this paper, the data of the soil moisture at 3 cm and the soil temperature at the surface were used to validate the models because the microwave sensors are sensitive to a few centimeters ( 3 cm) of soil depth. In 2008, Chinese and Japanese scientists jointly conducted an atmospheric experiment at the Gaize station in three stages (February 24 March 19, May 12 June 11, and July 6 July 16). During these periods, radiosonde soundings (Vaisala Radiosonde RS92-SGP) were launched every 6 h. The observations made during the period of May 12 June 11 were used to validate the models. This work was supported by a China and Japan intergovernmental cooperation program the JICA-Tibet project [60]. 5) Initial and Boundary Conditions: The initial and lateral boundary conditions are derived from the data of the National Center for Environmental Prediction (NCEP). These NCEP FNL (Final) Operational Global Analysis data are available on 1.0 by 1.0 grids at 6-h intervals. This product is derived from the Global Forecast System that runs four times a day in near real time at the NCEP. To obtain the corresponding initial and boundary conditions required to run the mesoscale model, the analysis results available at 26 mandatory levels from 1000 to 10 mb were used. The variables (e.g., pressure, geopotential height, temperature, relative humidity, and u and v winds) required for each mesoscale model vertical layer were interpolated from the 26 mandatory pressure levels. In addition, NCEP continuously maintains the archived time series of the NCEP/FNL data set to a near-current date but not in real time [ B. Methods 1) As shown in Fig. 1, the coupled land atmosphere mesoscale model (ARPS) was set up using the initial and boundary conditions from the NCEP FNL operational global analysis data. The ARPS model runs for a predefined period (10 min) and passes the forcing data to LSM SiB2. 2) At the beginning of the SiB2 integration time, the ensemble (50 members) of soil moisture profiles is generated by adding pseudorandom noise with prescribed statistics to the first guess soil moisture contents in the surface, root, and deep layers. SiB2 is independently simulated by every ensemble member of the soil moisture profile, model parameter, and atmospheric forcing. At the end of the SiB2 calculation, the mean values of the soil moisture profile, soil temperature profile, surface heat, and momentum fluxes are computed from the ensemble of forecast and are fed back to the ARPS as the lower boundary conditions of the atmospheric model. This procedure continues until the AMSR-E measurements are available. 3) At times when the AMSR-E observations are available, the brightness temperatures at frequencies of 6.9 and GHz are perturbed to produce an ensemble of observations with prescribed statistics. The SiB2-driven

7 RASMY et al.: DEVELOPMENT OF A SATELLITE LDAS COUPLED WITH A MESOSCALE MODEL 2853 soil ensemble of moisture profiles and surface and canopy temperatures is used to obtain the simulated brightness temperatures (at 6.9 and 10.7 GHz) using the forward radiation transfer model. The EnKF calculates the updated soil moisture profile with the ensemble of forecast soil moisture profiles, ensemble of brightness temperatures simulated by the RTM, and ensemble of observed brightness temperatures. The resulting updated land surface conditions such as those of soil moisture, temperature, and heat and moisture fluxes are then used to reinitialize the land atmosphere coupled model. This procedure is repeated whenever AMSR-E observations are available. 4) With the reinitialized land surface conditions, the ARPS model is integrated forward in time for atmospheric prediction until the next AMSR-E observations are available. IV. EXPERIMENTAL DESCRIPTIONS To validate the coupled data assimilation system and to assess its retrieval capabilities, the system was applied to a mesoscale area of the western Tibetan Plateau. The performance of the new system was investigated by considering two cases: 1) the one-way nesting procedure employing the land atmosphere model (ARPS; without AMSR-E data) and 2) the land atmosphere model coupled with a sequential LDAS (LDAS-A). Both models for the land and atmospheric fields were initialized with the NCEP FNL data set, and the boundary conditions derived from the same NCEP FNL data set were supplied every 6 h. The results were compared with the surface Automatic Weather Station measurements, satellite (MTSAT-1R and Terra MODIS) observations, and radiosonde soundings. A. Study Domain and Model Setup The Tibetan Plateau is climatologically unique, owing to its high elevation and vastness. Numerous studies on the plateau have been conducted, and they suggested that the thermal effect of the giant plateau, the large amount of solar radiation absorption, and the dramatic seasonal changes of the surface heat and water fluxes greatly influence the atmospheric dynamics and circulations over Asia [22], [63], [64]. In this paper, a mesoscale area located in the western part of the Plateau, including the Gaize station (84.05 E, 32.3 N), and bounded by the area (82.7 E 85.2 E, 30.7 N 33.2 N) was considered (Fig. 3). This region is especially characterized by a wider flat valley and a mountainous topography with a heterogeneous soil moisture distribution that is favorable in investigating land atmosphere interactions. The land use type is bare land or sparse vegetation without intense human activity, which ensures the applicability of microwave RTM to AMSR-E observations without great complication. The other main reason in selecting this region is the availability of data set for model validation. To capture the small-scale atmospheric features related to the land surface effect, the model domain horizontal resolution was set to The total number of grids in an x y-direction was 70 70, covering a domain area of km 2. For the vertical grid, the atmospheric model used a hyperbolic tangent function to stretch the grid interval from 40 m at the first level and 53 atmospheric layers in total Fig. 3. Mesoscale model domain: topographical map (in meters) including the Gaize station. TABLE I OPTIMIZED RTM PARAMETERS FOR GAIZE ( 18 km above the ground surface). The model is configured with a 1.5-order TKE-based closure scheme for subgrid scale turbulent mixing, the latitude-dependent Coriolis parameters, and the Lin ice microphysics [35] as the microphysical processes. The LSM parameters were set as described in [3], and the geographical data sets, such as those of soil type, land use type, and vegetation parameters, were obtained from ftp://aftp.fsl.noaa.gov/divisions/frd-laps/wrfsi/geog_data. In addition to surface soil moisture, several parameters (e.g., soil texture, porosity, roughness height and correlation length, and vegetation parameters) are required to calculate the brightness temperature at the satellite level. However, the spatially distributed values of these parameters are highly variable and unavailable for each model grid. To obtain the soil parameters (e.g., soil textural compositions and hydraulic conductivity), several studies utilized passive microwave (L- or C-band) brightness temperature observations or near-surface soil moisture information derived from microwave remote sensing. References [4], [46], [52], [65], and [66] proposed a dual-pass assimilation technique in which Pass 1 inversely estimates the optimal value of the parameters at each model grid point using an LSM, a land data assimilation method, and the long-term ( months) AMSR-E observations of the brightness

8 2854 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011 Fig. 4. Spatial distribution of the simulated surface soil moisture contents (in cubic meters per cubic meter) at 2100 UTC on June 6, (a) ARPS (noassimilation case). (b) LDAS-A. (c) Terra MODIS land surface temperature (in kelvin) observed at 0900 UTC on June 7, The contour lines depict the topography. temperatures (C-band) along with the observed forcing data. This method was adopted in the present study to optimize the spatially distributed RTM parameters using two months (May June) of the AMSR-E (6.925 and 10.6 GHz) brightness temperature observations along with the forcing obtained from the land atmosphere model. The vegetation parameters were kept constant in the RTM model calculations because of the land use type in the selected model domain (bare land or sparse vegetation). The optimized parameters for the Gaize station are listed in Table I. V. R ESULTS AND DISCUSSION The new system (LDAS-A) was applied to the mesoscale domain for a period of one month (from May 20, 2008, to June 19, 2008). To test its performance and applicability, the simulation results are compared against the results for the no-assimilation case (ARPS), in situ measurements, and radiosonde and satellite observations. To assess the soil moisture assimilation capabilities of the LDAS-A at regional scales, the spatial distribution of the surface soil moisture content was investigated in the domain scale. For the domain-scale validation of the model results, the spatial distributions of the daytime Terra MODIS land surface temperature products were obtained for clear-sky days (land surface observations are not available for cloudy days). Since the vegetation effects are negligible in the simulated domain, the diurnal range of the MODIS surface temperature can indirectly indicate the soil moisture conditions of the selected footprint of the MODIS observations (i.e., a larger diurnal range of the land surface temperature is associated with a lower surface soil moisture content and vice versa). Fig. 4(a) and (b) shows the spatial distributions of the surface soil moisture content at 2100 UTC on June 6, 2008 (immediately after assimilation in the LDAS-A), simulated by the ARPS and LDAS-A. The

9 RASMY et al.: DEVELOPMENT OF A SATELLITE LDAS COUPLED WITH A MESOSCALE MODEL 2855 Fig. 5. Scatter plots for the simulated surface soil moisture against the Terra MODIS land surface temperature extracted for grid cells with elevations ranging from 4400 to 4900 m. (a) ARPS. (b) LDAS-A. soil moisture content simulated by the ARPS [Fig. 4(a)] was much higher than that simulated by the LDAS-A [Fig. 4(b)] in most of the model cells, particularly in the central to northern regions. Fig. 4(c) shows the spatial distribution of the daytime land surface temperature at 0900 UTC [1500 local time (LT)] on June 7, The figure shows that the spatial distribution of the LDAS-A soil moisture was reasonably comparable to the distribution of the MODIS land surface temperature, and in particular, regions with a lower moisture content corresponded well to regions with higher MODIS land surface temperatures. On the other hand, the ARPS model results did not agree with the MODIS land surface temperature, and they showed wet and dry surface conditions for higher land surface temperatures. To compare the distribution of the model-simulated soil moisture content against the distribution of the MODIS land surface temperature more closely, the soil moisture and corresponding temperature values were extracted from grid cells with elevations ranging from 4400 to 4900 m (regions higher than 4900 m were considered mountainous). Fig. 5(a) and (b) shows the scatter plots of the MODIS land surface temperature against the ARPS and LDAS-A soil moisture contents, respectively. As shown in Fig. 5(b), the LDAS-A provides very low simulated soil moisture contents with a very little variation ( m 3 /m 3 ), and these low moisture contents are densely concentrated to a very high range of land surface temperatures (300 K 320 K). On the other hand, the ARPS model results [Fig. 5(a)] have a wider range of soil moisture contents of m 3 /m 3 for higher MODIS land surface temperatures (almost the same land surface temperatures for wet and dry soil conditions). Therefore, these results indicate that the AMSR-E lower frequency data can be used to correct the errors in the surface soil moisture contents, and the assimilation processes in the LDAS-A model introduce regions of soil moisture contrast into the mesoscale model. To validate the LDAS-A results quantitatively, the model results were compared with the surface and sounding observations recorded at the Gaize station. Fig. 6(a) compares the surface soil moisture simulated by the ARPS and LDAS-A with the soil moisture (at 3 cm) recorded at the Gaize station. The figure shows that, at the time of initialization (0000 UTC on May 20, 2008), the surface soil moisture contents derived from the NCEP/FNL data set in both models were higher than the soil moisture content recorded at the Gaize station. After the initializations, the soil moisture in both models evolved over time under the forcing of the LSM. However, at 2000 UTC on May 20, 2008, when the first AMSR-E observations were available, the procedures of the assimilation in the LDAS-A reduced the biases (a sudden drop in the figure) originating from the NCEP/FNL data set and model forcing of the LSM prior to the assimilation. Similarly, as shown in the figure, the second AMSR-E observations available at 2000 UTC on May 22, 2008, greatly decreased the biases in the surface soil moisture content (the second sudden drop), and as a result, the surface soil moisture content simulated by the LDAS-A realigned with the observation. On the other hand, the biases in the soil moisture contents of the ARPS model were much higher than the biases in the LDAS-A s simulation at the Gaize station on these days. At 1000 UTC on May 23, 2008, the atmospheric prediction components of the ARPS and LDAS-A models produced inaccurate rainfall events, which increased the model-simulated soil moisture contents. The exact timing and amount of the model rainfall events are often incorrect in NWP, and these model rainfall events can destroy the assimilated soil moisture contents and their variations in a very short time over a wide region. In addition, when a model predicts rainfall events, the solar radiation at the land surfaces is also reduced, owing to the model-simulated cloud conditions. The combination of the inaccurate forcing of the LSM and higher soil moisture content (due to rainfall events) in soil layers results in LSM introducing inaccurate land surface forcing of the overlying atmosphere. Therefore, inaccurate rainfall prediction is one of the biggest issues in land data assimilation coupled with mesoscale models. However, the rainfall-induced biases in the simulated soil moisture content were reduced by the assimilation of the AMSR-E observations at 2000 UTC on May 23, 2008, and thus, the LDAS-A surface soil moisture contents were much closer to the measurements at the Gaize grid point. Similarly, the consecutive assimilation of the AMSR-E observations resulted in the LDAS-A soil moisture contents agreeing well with the observed soil moisture variation during a dry or no-rain period (from May 20, 2008, to June 4, 2008), whereas the rainfallinduced biases in the APRS model advanced in time. As a result, the ARPS model overestimated the surface soil moisture content for all days during the dry period. The diurnal variation in the observed soil moisture in Fig. 6(a) indicates three rainfall events that correspond to the periods June 4 5, 2008, June 11 13, 2008, and June 17 18, In all three events, two consecutive morning ( LT) and evening ( LT) peaks of soil moisture increments are clearly noted in the observations

10 2856 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011 Fig. 6. Comparison of the observed and simulated surface variables at the Gaize station from May 20, 2008, to June 19, (a) Surface volumetric soil moisture (in cubic meters per cubic meter) at 3 cm. (b) Surface temperature (in kelvin). (Black circle) In situ measurements. (Gray line) Results simulated by the ARPS (no-assimilation case). (Dark line) Results simulated by the LDAS-A. in Fig. 6(a). The figure shows that both the ARPS and LDAS-A models failed to address the morning soil moisture peaks associated with the morning rainfall events on June 4, 12, and 17, In the case of the LDAS-A, this failure was due to the limited frequency of the AMSR-E observations (one or, at best, two overpasses per day), whereas we only considered the descending ( 2000 UTC or 0200 LT) overpass of the AMSR-E measurements in the assimilation procedures to reinitialize the model soil moisture fields. As a result, the LDAS-A model failed to address the soil moisture increments introduced by the morning rainfall events on these days. However, the morning peaks in soil moisture on June 5 and 13, 2008, were well captured by both the ARPS and LDAS-A models. Furthermore, although the LDAS-A predicted the evening soil moisture peak on June 5, 2008, much better than the ARPS model at 2100 UTC on June 5, 2008, there is a clear and sudden drop in the soil moisture content simulated by the LDAS-A from0.28to0.18m 3 /m 3, resulting from the assimilation of the AMSR-E observations for 2000 UTC on June 5, 2008, whereas at this time, the measurements made at the Gaize grid point show a higher soil moisture content ( 0.25 m 3 /m 3 ).This sudden drop due to the assimilation resulted in the soil moisture simulated in the LDAS-A model to be underestimated until the availability of the next AMSR-E observations at 20 UTC on June 7, To verify the assimilated soil moisture results, which show a sudden drop on June 5, 2008, the AMSR-E brightness temperature observations made closer to the Gaize station were investigated. Fig. 7 shows the variation in the AMSR-E brightness temperature in the 6.9- and 10.6-GHz vertical channels closer to the Gaize station. The figure clearly Fig. 7. AMSR-E brightness temperature vertical channel observations at the Gaize station. (Open circle) 6.9 GHz. (Closed circle) 10.6 GHz. shows that there are sudden drops in brightness temperature on June 5, 12, and 18, These sudden drops correspond to the increases in the soil moisture content due to rainfall events that might have occurred within the pixel that includes the Gaize station. The difference in brightness temperature on June 5, 2008, is less ( 10 K) than the other two drops ( 20 K). However, the observed soil moisture increments in all three instances were similar. Owing to the brightness temperature difference during the first rainfall event being smaller than the differences during the other two rainfall events, the assimilated soil moisture in Fig. 6(a) decreased to less than the observed soil moisture content, whereas the other two rainfall

11 RASMY et al.: DEVELOPMENT OF A SATELLITE LDAS COUPLED WITH A MESOSCALE MODEL 2857 TABLE II MBE AND RMSE FOR THE SURFACE SOIL MOISTURE (3 cm SKIN TEMPERATURE SIMULATED BY THE ARPS AND LDAS-A AT THE GAIZE STATION Fig. 8. Hourly variation of the convective index (Ic) derived from the MTSAT-1R channel 1 brightness temperature (in kelvin) observations at the Gaize station. cases were well captured by AMSR-E, and consequently, the simulated soil moisture content in the assimilation case agreed well with the observation values. This problem could result from the low spatial resolution of the AMSR-E brightness temperature observations at 6.9 and 10.6 GHz, which is of the integral signal of the entire pixel area ( 40 km), where the first rainfall event might correspond to a very localized event concentrated over the Gaize station on a scale smaller than the AMSR-E pixel resolution. To investigate the localized rainfall events, the hourly variations in the MTSAT-1R channel 1 cloud-top temperatures at the Gaize grid point were analyzed. To confirm whether there was a cloud activity over the Gaize station during the model simulation period, the diurnal variation of the convective index (Ic) presented in [21] was analyzed. Ic is defined by Ic = 250 T bb (T bb < 250 K) (14) Ic =0 (T bb 250 K) (15) where 250 K corresponds to the air temperature 9000 m above sea level over Tibet in the summer period. Therefore, Icis an index of deep convective cloud whose top height exceeds 9000 m. As shown in Fig. 8, during later rainy periods (June and 17 18, 2008), the estimated values for Ic at Gaize were much higher (> 25) and remained high for several hours, indicating the existence of a tall and spatially extended cloud system over Gaize. On the other hand, on June 4 5, 2008, the estimated values for Ic were much smaller (< 10), and there were clouds over Gaize but only for a limited period compared with the other two identified rainfall events, indicating a much less cumulus activity in the case of very localized events. Therefore, the MTSAT/1R1 observations confirmed that the soil moisture increments on June 4 5, 2008, resulting from a very localized rainfall event that occurred on a scale much smaller than the AMSR-E pixel resolution. Therefore, when validating the satellite-derived soil moisture with point observations, care must be taken to account for spatial heterogeneity at the subgrid scale (e.g., it may be necessary to upscale the ground observations to the AMSR-E pixel resolution using more observation points) [66]. Moreover, on June 14 and 18, 2008, as shown in Fig. 6(a), after soil moisture increments due to rainfall events and/or assimilation, the LDAS-A failed to dry out or had a lower drying rate than the observation, which indicates a rapid decrease in the surface soil moisture content, owing to higher evaporation and infiltration. This error can be ascribed to many factors such as model forcing of the LSM (e.g., solar radiation at the surface) and model-specific parameters (e.g., hydraulic conductivity, porosity, and variation in the soil depth) that vary on a centimeter scale, where, in the model, a single value represents a model cell ( 25 km 2 ). However, these model-related shortcomings are corrected by the assimilation procedures, and the soil moisture curves realigned with the observation curves [sudden drops in the soil moisture content in Fig. 6(a)] particularly on June 18, Furthermore, the mean bias error (MBE) and root-mean-square error (rmse) for the surface soil moisture and surface temperature derived by the ARPS and LDAS are presented in Table II for the entire simulation period. The table shows that the errors in surface soil moisture simulated by the LDAS-A (MBE = and rmse =0.0422) were much less than those in the ARPS simulation (MBE = and rmse =0.0818). Fig. 6(b) compares the surface temperatures simulated by the ARPS and LDAS-A against the surface temperature observed at the Gaize station. As shown in the figure, the diurnal variations in the surface temperature, especially daytime highs simulated by the LDAS-A, agreed well with the observed diurnal variations, whereas the variations during the dry period are underestimated by the ARPS. The ARPS underestimates the daytime highs because of the overestimation of the soil moisture content [Fig. 6(a)], which inaccurately partitions the available surface energy to give a higher estimate of evaporation and a lower estimate of the daytime surface temperature. Since assimilation significantly improved the soil moisture variations such that they are comparable to observations especially during dry or no-rain periods, the simulation results of the LDAS-A capture the daytime highs well. Furthermore, both models gave reasonably good results in estimating the nighttime lows (except for a few days) compared against the observations that indicate a negligible influence of the assimilation of the soil moisture content on the simulated nighttime lows. Table II shows that the estimated MBE for the LDAS-A is much lower ( 0.28 K) than that for the ARPS model ( 1.38 K). However, the estimated rmse for the LDAS-A is slightly higher ( 0.5 K) than the rmse for the ARPS model. During rainy periods (e.g., June 8, 9, and 17, 2008), lower surface temperatures were observed, owing to less solar radiation on the ground (data not shown), whereas both models estimated a higher surface temperature associated with the overestimation of solar radiation. In addition, the LDAS-A, compared with the ARPS model, underestimated the surface soil moisture for these days, which

12 2858 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011 Fig. 9. Comparison between the observed and simulated potential temperature soundings and solar radiations at the Gaize station from May 20, 2008, to June 4, (a) MBE for the potential temperature soundings. (b) RMSE for the potential temperature. (c) Solar radiation at the surface. resulted in the overestimation of the surface temperature. As a result, the overall rmse for the surface temperature in the LDAS-A was slightly higher than that in the ARPS model. Soil moisture and surface temperature control the heat and moisture transfer within the lower part of the troposphere (planetary boundary layer) and influence the dynamic and thermodynamic properties of the overlying atmosphere. To investigate the effect of improved land surface conditions and, thus, land atmosphere interactions on the overlying atmosphere, the radiosonde soundings were compared with the LDAS-A and ARPS model soundings. Fig. 9(a) and (b) compares the MBE and rmse for the model-simulated potential temperature soundings, respectively. To estimate the error statistics, potential temperature values from land surfaces to a height of 400 hpa (the maximum height of the boundary layer observed during daytime from the radiosonde soundings) are considered to isolate the effect of land surfaces within the boundary layer. The figures show greater error differences between the ARPS and the LDAS-A especially from noon to evening ( UTC) when the surface is heated by strong solar radiation and the land surface memory is translated to the overlaying atmosphere. There are clear improvements in using LDAS-A for some days (e.g., May 23, 24, 28, and 29, 2008, and June 2, 2008) when the observed solar radiation [Fig. 9(c)] agreed reasonably well with the simulated solar radiation and, in particular, the sky was cloud-free according to both models and observations. Since the improvements were clearly noted at 1100 UTC (1700 LT), two cases for 1100 UTC were selected for demonstration. Figs. 10 and 11 compare the observed soundings of the potential temperature and specific humidity with the soundings simulated by the ARPS and LDAS-A for May 23 and 28, 2008, respectively. On these two days, the land surfaces in both models received the same solar radiation from the atmospheric model. However, the soil moisture amount in the LDAS-A was much lower than that in the ARPS model [Fig. 6(a)]. A lower soil moisture in the LDAS-A enhanced land surface heating, which resulted in an increment in the potential temperature profiles with improvements over the ARPS model. In addition, lower soil moisture conditions in the LDAS-A limited the moisture fluxes to the atmosphere within the boundary layer, and therefore, specific humidity profiles simulated by the LDAS-A were better than those simulated by the ARPS, especially for May 23, Improving the surface soil moisture and, thus, heat and moisture fluxes eventually improved the atmospheric profiles of the potential temperature and specific humidity from land surfaces to the maximum pressure height of hpa. Although the profiles of the potential temperature and specific humidity simulated by the models were not exactly the same as the observed profiles, the simulation results of the LDAS-A are

13 RASMY et al.: DEVELOPMENT OF A SATELLITE LDAS COUPLED WITH A MESOSCALE MODEL 2859 Fig. 10. Comparison of the observed soundings with the ARPS and LDAS-A model soundings at 1100 UTC on May 23, (a) Potential temperature (in kelvin). (b) Specific humidity (in grams per kilogram). Fig. 11. Same as Fig. 10 but for 1100 UTC on May 28, better than those of the ARPS model. Therefore, these results suggest that the consideration of land surface heterogeneities has the potential to improve the land atmosphere interactions and atmospheric structures in the NWP models. The model atmospheric structures are not only controlled by the land surface conditions but are also affected by the simulated atmospheric conditions (e.g., moisture fields and cloud conditions), forcing of the LSM (e.g., downward shortwave and longwave radiation and rainfall), environmental forcing (dynamics), and lateral boundary conditions used to derive the mesoscale models. We further investigated the effect of solar radiation (or cloud cover) on the model-simulated soundings. As shown in the figures, although the LDAS-A well simulated the surface soil moisture conditions for some days (e.g., May 27, 2008, May 30, 2008, and June 3, 2008), the MBE and rmse calculated for the potential temperature soundings simulated by the ARPS model were smaller than those for the LDAS-A model. These differences largely result from the combined effect of biases in the simulated soil moisture conditions and solar radiations. The presence of clouds in reality over the Gaize station reduced the amount of solar radiation received at the surface, whereas both models predicted similar and higher values for solar radiation. During the same periods, simulated surface soil moisture status of the models indicated wet conditions for ARPS and dry conditions for LDAS-A [Fig. 6(a)]. A lower soil moisture content together with a strong solar radiation increased land surface heating and, thus, boundary layer atmospheric temperatures, which make the MBE estimated for LDAS-A positive and both MBE and RMSE higher than the errors in the ARPS model on these days. In the ARPS model, the combined effect of higher soil moisture content and strong solar radiation compensated for the biases. As a result, a less surface heating was estimated compared with the case for LDAS-A. Therefore, in addition to the soil moisture content, solar radiation is a critical input to LSMs, and both should be estimated accurately to introduce realistic land atmosphere interactions into coupled models. Moreover, solar radiation can be improved through the assimilation of atmospheric conditions and the cloud distribution using satellite observations, and this is a future direction of the present research. VI. CONCLUSION The land surface processes are an essential and substantial consideration in understanding and predicting the global water and energy budgets. The soil water content is the single most important land surface variable in land atmosphere coupled models, and it controls moisture and heat fluxes that strongly affect large- and small-scale circulations and cloud convective initialization and development. To physically introduce accurate and existing land surface conditions into the NWP models, an LDAS is coupled with a mesoscale atmospheric model.

14 2860 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011 Since the newly developed system (LDAS-A) directly assimilates the lower frequency AMSR-E passive microwave brightness temperatures, the limitations that arise when using the soil moisture products (e.g., preprocessing and nudging methods) and assimilating the screen-level parameters are eliminated. In addition, the LDAS-A assimilates the satellite brightness temperature in a sequential manner. Once the observations are available from the satellite, the LDAS-A can immediately introduce the actual land surface conditions to the land atmosphere models, and thus, the LDAS-A is a feasible approach for nearreal-time applications. The system was evaluated in numerical experiments for the mesoscale domain of the western Tibetan Plateau. The results show that the simulation of the surface soil moisture content was much better than that in the case without assimilation. Moreover, the LDAS-A demonstrated that the AMSR-E lower frequency data have the potential to correct the biases in the surface soil moisture content caused by model-specific parameters and model forcing, especially in terms of precipitation and radiation on different scales. The improvements in the soil moisture content improved the simulated land surface temperature. As a result, the diurnal variation in surface temperature, especially the trend of the daytime highs simulated by the LDAS-A, agreed well with the trend observed at the Gaize station, whereas the no-assimilation case underestimated this trend, owing to the high moisture content simulated in the model. Intensive and unique radiosonde sounding data sets were used to investigate the effect of the improved land surface conditions (i.e., soil moisture and surface temperature) on the simulated land atmosphere interactions. The soundings of the potential temperature and specific humidity were better in the assimilation case than in the no-assimilation case during periods when the simulated solar radiations were comparable to the observations. The improvement in the LDAS-A model could be clearly seen from the surface to 400 hpa, especially on clear-sky days with strong land surface heating. However, the LDAS-A produced larger errors than the ARPS model in simulating the potential temperature soundings when the land surface conditions were drier and the atmospheric model overestimated the solar radiation. These results demonstrate the necessity of improving the solar radiation (or cloud distribution) to accurately introduce realistic land atmosphere interactions in the NWP. The results are very encouraging in terms of reliably in predicting the land surface moisture and heat fluxes and improved land atmosphere interactions in remote regions where no observation networks are available. Although we have presented a clear evidence that the investigated system is capable of improving the land surface variables and the land atmosphere interactions in the NWP models, further improvements in terms of system performance and applicability can be made. 1) In this paper, we have not addressed the improvement of the atmospheric initialization, which is more critical in a numerical weather forecast. Potential improvements in predictability from atmospheric initialization are greater than those from land initialization for short-term forecasts (1 2 weeks) [31]. 2) At the present, the system only assimilates soil moisture. However, the assimilation of the remotely sensed surface temperature and cloud information into the system has the potential to effectively improve the soil moisture and solar radiation estimations, the land surface fluxes, and the land atmosphere interactions. 3) The proposed system is also applicable to available (e.g., Tropical Rainfall Measuring Mission Microwave Imager) and future (e.g., Global Precipitation Measurement Microwave Imager and Global Change Observation Mission) passive microwave sensors that, in turn, will reduce the AMSR-E temporal sampling problem by increasing the frequency of the observations, therefore correcting the model biases more frequently. ACKNOWLEDGMENT The authors would like to thank K. Tamagawa and T. Ohta, Data Archiving Managers of the Coordinated Energy and Water Cycle Observation Project, The University of Tokyo, for their continued support in the model validations; Japan Aerospace Exploration Agency for providing the AMSR-E brightness temperature observations; and Kochi University for providing the MTSAT/1R data set. The surface and radiosonde data were provided by the Japan China JAICA project. This paper was carried out as part of the Research Program on Climate Change Adaptation and verification experiment for AMSR/AMSR-E. REFERENCES [1] G. Balsamo, P. Viterbo, A. Beljaars, B. van den Hurk, M. Hirschi, A. Betts, and K. Scipal, A revised hydrology for the ECMWF model: Verification from field site to terrestrial water storage and impact in the integrated forecast system, J. Hydrometeor., vol. 10, no. 3, pp , [2] A. C. M. Beljaars, P. Viterbo, M. Miller, and A. Betts, The anomalous rainfall over the United States during 1993: Sensitivity to land surface parameterization and soil moisture anomalies, Monthly Weather Rev., vol. 124, no. 3, pp , [3] S. Boussetta, T. Koike, K. Yang, T. Graf, and M. Pathmathevan, Development of a coupled land atmosphere satellite data assimilation system for improved local atmospheric simulations, Remote Sens. Environ., vol. 112, no. 3, pp , Mar [4] D. H. Chang, R. Kothari, and S. Islam, Classification of soil texture using remotely sensed brightness temperature over the Southern Great Plains, IEEE Trans. Geosci. Remote Sens., vol. 41, no. 3, pp , Mar [5] K. Chen, T. D. Wu, L. Tsang, Q. Li, J. C. Shi, and A. K. Fung, Emission of rough surfaces calculated by the integral equation method with comparison to a three dimensional moment method simulation, IEEE Trans. Geosci. Remote Sens., vol. 41, no. 1, pp , Jan [6] B. J. Choudhury and T. J. Schmugge, Effect of surface roughness on the microwave emission from soil, J. Geophys. Res., vol. 84, no. c9, pp , [7] S. Conil, H. Douville, and S. Tyteca, Contribution of realistic soil moisture initial conditions to boreal summer climate predictability, Clim. Dyn., vol. 32, no. 1, pp , Jan [8] W. Crow and X. Zhan, Continental-scale evaluation of remotely sensed soil moisture products, IEEE Geosci. Remote Sens. Lett., vol. 4, no. 3, pp , Jul [9] M. Dobson, F. Ulaby, M. Hallikainen, and M. El-Rayes, Microwave dielectric behavior of wet soil Part II, dielectric mixing models, IEEE Trans. Geosci. Remote Sens., vol. GE-23, no. 1, pp , Jan [10] C. S. Draper, J.-F. Mahfouf, and J. P. Walker, An EKF assimilation of AMSR-E soil moisture into the ISBA land surface scheme, J. Geophys. Res., vol. 114, p. D20104, 2009, DOI: /2008JD [11] M. Drusch, Initializing numerical weather prediction models with satellite-derived surface soil moisture: Data assimilation experiments with ECMWF s integrated forecast system and the TMI soil moisture data set, J. Geophys. Res., vol. 112, p. D03102, [12] M. Drusch and P. Viterbo, Assimilation of screen-level variables in ECMWF s integrated forecast system: A study on the impact on the forecast quality and analyzed soil moisture, Monthly Weather Rev., vol. 135, no. 2, pp , 2007.

15 RASMY et al.: DEVELOPMENT OF A SATELLITE LDAS COUPLED WITH A MESOSCALE MODEL 2861 [13] M. Drusch, E. F. Wood, H. Gao, and A. Thiele, Soil moisture retrieval during the Southern Great Plains Hydrology Experiment 1999: A comparison between experimental remote sensing data and operational products, Water Resour. Res., vol. 40, p. W02 504, 2004, DOI: / 2003WR [14] M. Drusch, E. F. Wood, and T. Jackson, Vegetative and atmospheric corrections for the soil moisture retrieval from passive microwave remote sensing data: Results from the Southern Great Plains Hydrology Experiment 1997, J. Hydrometeor., vol. 2, no. 2, pp , Apr [15] M. B. Ek, K. E. Mitchell, Y. Lin, P. Grunmann, E. Rogers, G. Gayno, J. Tarpley, and V. Koren, Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta, J. Geophys. Res., vol. 108, p. 8851, 2003, DOI: /2002JD [16] D. Entekhabi, H. Nakamura, and E. Njoku, Solving the inverse problem for soil moisture and temperature profiles by sequential assimilation of multifrequency remotely sensed observations, IEEE Trans.Geosci. Remote Sens., vol. 32, no. 2, pp , Mar [17] D. Entekhabi, E. Njoku, P. O Neill, K. Kellogg, W. Crow, W. Edelstein, J. Entin, S. Goodman, T. Jackson, J. Johnson, J. Kimball, J. Piepmeier, R. Koster, N. Martin, K. McDonald, M. Moghaddam, S. Moran, R. Reichle, J. Shi, M. Spencer, S. Thurman, L. Tsang, and J. Van Zyl, The Soil Moisture Active Passive (SMAP) mission, Proc. IEEE,vol.98, no. 5, pp , May [18] G. Evensen, The ensemble Kalman filter: Theoretical formulation and practical implementation, Ocean Dyn., vol. 53, no. C5, pp , [19] E. Fischer, S. Seneviratne, P. Vidale, D. Luthi, and C. Schar, Soil moisture atmosphere interactions during the 2003 European summer heat wave, J. Clim., vol. 20, no. 20, pp , Oct [20] H. Fujii and T. Koike, Development of a TRMM/TMI algorithm for precipitation in the Tibetan Plateau by considering effects of land surface emissivity, J. Meteorol. Soc. Jpn., vol. 79, no. 1B, pp , [21] H. Fujinami and T. Yasunari, The seasonal and intraseasonal variability of diurnal cloud activity over the Tibetan Plateau, J. Meteorol. Soc. Jpn., vol. 79, no. 6, pp , [22] Y. Gao, M. Tang, S. Luo, Z. Shen, and C. Li, Some aspects of recent research on the Qinghai Xizang Plateau meteorology, Bull. Amer. Meteor. Soc., vol. 62, no. 1, pp , Jan [23] H. Hess, Assimilation of screen-level observations by variational soil moisture analysis, Meteorol. Atmos. Phys.,vol.77,no.1 4,pp , Jun [24] T. Holmes, M. Drusch, J.-P. Wigneron, and R. de Jeu, A global simulation of microwave emission: Error structures based on output from ECMWF s operational integrated forecast system, IEEE Trans. Geosci. Remote Sens., vol. 46, no. 3, pp , Mar [25] C. Huang, X. Li, L. Lu, and J. Gu, Experiments of one-dimensional soil moisture assimilation system based on ensemble Kalman filter, Remote Sens. Environ., vol. 112, no. 3, pp , Mar [26] C. Huang, X. Li, and J. Gu, Estimation of regional soil moisture by assimilating multi-sensor passive microwave remote sensing observations based on ensemble Kalman filter, in Proc. IEEE IGARSS, 2008, vol. 3, pp. III-1052 III [27] T. Jackson et al., Soil moisture mapping at regional scales using microwave radiometry: The Southern Great Plains Hydrology Experiment, IEEE Trans. Geosci. Remote Sens., vol. 37, pp , [28] A. Jones, C. Combs, T. Lakhankar, S. Longmore, T. Vonder Haar, G. McWilliams, M. Mungiole, and G. Mason, NPOESS soil moisture satellite data assimilation: Progress using WindSat data, in Proc. IEEE IGARSS, 2007, pp [29] R. D. Koster et al., Regions of strong coupling between soil moisture and precipitation, Science, vol. 305, no. 5687, pp , Aug [30] R. D. Koster, Z. Guo, R. Yang, P. A. Dirmeyer, K. Mitchell, and M. Puma, On the nature of soil moisture in land surface models, J. Clim., vol. 22, no. 16, pp , Aug [31] R. Koster, M. Suarez, P. Liu, U. Jambor, A. Berg, M. Kistler, R. Reichle, M. Rodell, and J. Famiglietti, Realistic initialization of land surface states: Impacts on subseasonal forecast skill, J. Hydrometeor., vol. 5, pp , Dec [32] D. N. Kuria, T. Koike, H. Lu, T. Tsutsui, and T. Graf, Field-supported verification and improvement of a passive microwave surface emission model for rough, bare, and wet soil surfaces by incorporating shadowing effects, IEEE Trans. Geosci. Remote Sens.,vol.45,no.5,pp , May [33] Q. Li, L. Tsang, J. Shi, and C. Chan, Application of physics-based two-grid method and sparse matrix canonical grid method for numerical simulations of emissivities of soils with rough surfaces at microwave frequencies, IEEE Trans. Geosci. Remote Sens.,vol.38,no.4,pp , Jul [34] X. Li and T. Koike, Frozen soil parameterization in SiB2 and its validation with GAME-Tibet observations, Cold Regions Sci. Technol., vol. 3, no. 1 3, pp , Mar [35] Y. Lin, R. Farley, and H. Orville, Bulk parameterization of the snow field in a cloud model, J. Clim. Appl. Meteorol., vol. 22,no. 6, pp , Jun [36] Y. Liou and A. England, A land surface process/radio brightness model with coupled heat and moisture transport for freezing soils, IEEE Trans. Geosci. Remote Sens., vol. 36, no. 2, pp , Mar [37] H. Lu, T. Koike et al., Improving the AMSR-E soil moisture algorithm of The University of Tokyo through field experiments and parameters optimization, in Proc. IEEE IGARSS, 2008, vol. 2, pp [38] H. Lu, T. Koike, H. Fujii, and K. Tamagawa, A radiative transfer model and an algorithm for soil moisture including very dry conditions, in Proc. IGARSS, Jul. 2005, vol. 2, pp [39] J.-F. Mahfouf, K. Bergaoui, C. Draper, F. Bouyssel, F. Taillefer, and L. Taseva, A comparison of two off-line soil analysis schemes for assimilation of screen level observations, J. Geophys. Res., vol.114,p.d08105, 2009, DOI: /2008JD [40] I. Mladenova, V. Lakshmi, J. Walker, R. Panciera, W. Wagner, and M. Doubkova, Validation of the ASAR global monitoring mode soil moisture product using the NAFE 05 data set, IEEE Trans. Geosci. Remote Sens., vol. 48, no. 6, pp , Jun [41] E. Njoku and D. Entekhabi, Passive microwave sensing of soil moisture, J. Hydrol., vol. 184, no. 1/2, pp , Oct [42] J. Noilhan and S. Planton, A simple parameterization of land surface processes for meteorological models, Monthly Weather Rev., vol. 117, no. 3, pp , Mar [43] R. Panciera, J. Walker, J. Kalma, E. Kim, J. Hacker, O. Merlin, M. Berger, and N. Skou, The NAFE 05/cosmos data set: Toward SMOS soil moisture retrieval, downscaling, and assimilation, IEEE Trans. Geosci. Remote Sens., vol. 46, no. 3, pp , Mar [44] R. Panciera, J. Walker, and O. Merlin, Improved understanding of soil surface roughness parameterization for L-band passive microwave soil moisture retrieval, IEEE Geosci. Remote Sens. Lett., vol. 6, no. 4, pp , Oct [45] C. Pathe, W. Wagner, D. Sabel, M. Doubkova, and J. Basara, Using ENVISAT ASAR global mode data for surface soil moisture retrieval over Oklahoma, USA, IEEE Trans. Geosci. Remote Sens., vol. 47, no. 2, pp , Feb [46] V. R. N. Pauwels, A. Balenzano, G. Satalino, H. Skriver, N. E. C. Verhoest, and F. Mattia, Optimization of soil hydraulic model parameters using synthetic aperture radar data: An integrated multidisciplinary approach, IEEE Trans. Geosci. Remote Sens., vol. 47,no.2, pp , Feb [47] R. A. M. de Jeu, W. Wagner, T. Holmes, and A. J. Dolman, Global soil moisture patterns observed by space borne microwave radiometers and scatterometers, Surveys Geophys., vol. 29, no. 4/5, pp , Oct [48] R. H. Reichle and D. B. McLaughlin, Hydrologic data assimilation with the ensemble Kalman filter, Monthly Weather Rev., vol. 130, no. 1, pp , Jan [49] R. H. Reichle, J. P. Walker, and R. D. Koster, Extended versus ensemble filtering for land data assimilation, J. Hydrometeor., vol. 35, no. 6, pp , Dec [50] R. Reichle, W. Crow, R. Koster, H. Sharif, and S. Mahanama, Contribution of soil moisture retrievals to land data assimilation products, Geophys. Res. Lett., vol. 35, p. L01404, 2008, DOI: /2007GL [51] A. Robock, Y. V. Konstantin, G. Srinivasan, J. Entin, S. Hollinger, and A. Namkhai, The Global Soil Moisture Data Bank, Bull. Amer. Meteorol. Soc., vol. 81, no. 6, pp , [52] J. A. Santanello, Jr., C. D. Peters-Lidard, M. Garcia, D. M. Mocko, M. A. Tischler, M. S. Moran, and D. P. Thoma, Using remotely-sensed estimates of soil moisture to infer soil texture and hydraulic properties across a semi-arid watershed, Remote Sens. Environ., vol. 110, no. 1, pp , [53] K. Scipal, M. Drusch, and W. Wagner, Assimilation of a ERS scatterometer derived soil moisture index in the ECMWF numerical weather prediction system, Adv. Water Resour., vol. 31, no. 8, pp , Aug

16 2862 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 49, NO. 8, AUGUST 2011 [54] P. J. Sellers, S. Los, C. Tucker, C. Justice, D. Dazlich, and G. Collatz, A revised land surface parameterization (SiB2) for atmospheric GCMs, Part II: The generation of global fields of terrestrial biophysical parameters from satellite data, J. Clim., vol. 9, pp , [55] J. C. Shi, L. Jiang, L. Zhang, K. Chen, J. P. Wigneron, and A. Chanzy, A parameterized multi frequency-polarization surface emission model, IEEE Trans. Geosci. Remote Sens., vol. 43, no. 12, pp , Dec [56] A. Shibata, K. Imaoka, and T. Koike, AMSR/AMSR-E level 2 and 3 algorithm developments and data validation plans of NASDA, IEEE Trans. Geosci. Remote Sens., vol. 41, no. 2, pp , Feb [57] C. Taylor, Intraseasonal land atmosphere coupling in the West African Monsoon, J. Clim., vol. 21, no. 24, pp , Dec [58] F. Ulaby, R. K. Moore, and A. Fung, Eds., Microwave Remote Sensing: Active and Passive Volume Scattering and Emission Theory, vol. III. Dedham, MA: Artech House, [59] J. R. Wang and B. J. Choudhury, Remote sensing of soil moisture content over bare fields at 1.4 GHz frequency, J. Geophys. Res., vol. 86, no. C6, pp , [60] X. Xiangde, Z. Renhe, T. Koike, L. Chungu, and S. Xiaohui, A new integrated observational system over the Tibetan Plateau, Bull. Amer. Meteorol. Soc., vol. 89, no. 10, pp , [61] M. Xue, K. K. Droegemeier, and V. Wong, The Advanced Regional Prediction System (ARPS) A multi-scale nonhydrostatic atmospheric simulation and prediction model. Part I: Model dynamics and verification, Meteor. Atmos. Phys., vol. 75, pp , [62] M. Xue, K. K. Droegemeier, V. Wong, A. Shapiro, K. Brewster, F. Carr, D. Weber, Y. Liu, and D. Wang, The Advanced Regional Prediction System (ARPS) A multi-scale nonhydrostatic atmospheric simulation and prediction tool. Part II: Model physics and applications, Meteor. Atmos. Phys., vol. 76, pp , [63] M. Yanai, C. Li, and Z. Song, Seasonal heating of the Tibetan Plateau and its effects on the evolution of the Asian summer monsoon, J. Meteorol. Soc. Jpn., vol. 70, pp , Dec [64] K. Yang, T. Koike, H. Fujii, T. Tamura, X. Xu, L. Bian, and M. Zhou, The daytime evolution of the atmospheric boundary layer and convection over the Tibetan Plateau: Observations and simulations, J. Meteorol. Soc. Jpn., vol. 82, no. 6, pp , [65] K. Yang, T. Watanabe, and T. E. A. Koike, An auto-calibration system to assimilate AMSRE data into a land surface model for estimating soil moisture and surface energy budget, J. Meteorol. Soc. Jpn., vol. 85A, pp , [66] K. Yang, T. Koike, I. Kaihotsu, and J. Qin, Validation of a dual-pass microwave land data assimilation system for estimating surface soil moisture in semiarid regions, J. Hydrometeor., vol. 10, no. 3, pp , Jun Toshio Koike received the B.Eng., M.Eng., and D.Eng. degrees from The University of Tokyo, Tokyo, Japan, in 1980, 1982, and 1985, respectively. He was a Research Associate with The University of Tokyo in 1985 and was appointed as an Assistant Professor in He was also appointed as an Associate Professor with the Nagaoka University of Technology, Nagaoka, Japan, in He has been a Professor with the River and Environmental Engineering Laboratory, Department of Civil Engineering, The University of Tokyo, since He is the Lead Scientist of the Coordinated Enhanced Observing Period project. His research interests are in hydrology, water resources, satellite remote sensing, climate change, and Asian monsoons. Souhail Boussetta received the B.Eng. and M.Eng degrees from The University of Tunis, Tunis, Tunisia, in 1991 and 1993, respectively, and the Ph.D. degree from The University of Tokyo, Tokyo, Japan, in He was a Research Associate with The University of Tokyo in 2006 and was appointed as an Assistant Professor for EDITORIA at The University of Tokyo in He is currently a Scientist with the European Centre for Medium-Range Weather Forecast, Reading, U.K. His research interests include land surface parameterization, land data assimilation, and satellite remote sensing applied to water resources. Hui Lu received the B.Eng. and M.Eng. degrees from Tsinghua University, Beijing, China, in 2000 and 2003, respectively, and the D.Eng. degree from The University of Tokyo, Tokyo, Japan, in He is currently with the Center for Earth System Science, Tsinghua University, where he was appointed as an Associate Professor in His research interests are radiative transfer model development through field experiments and numerical simulations, passive microwave remote sensing of land surface parameters, and data assimilation. Mohamed Rasmy received the B.Sc.Eng. (Hons.) degree from the Faculty of Engineering, University of Peradeniya, Peradeniya, Sri Lanka, in 2003 and the M.Sc. and Ph.D. degrees from The University of Tokyo, Tokyo, Japan, in 2006 and 2010, respectively. He is currently a Research Fellow with the River and Environmental Engineering Laboratory, Department of Civil Engineering, The University of Tokyo. His research interests are water resources, microwave remote sensing, land and atmospheric data assimilations, and numerical weather prediction. Xin Li received the B.Sc degree from Nanjing University, Nanjing, China, in 1992 and the Ph.D. degree from the Chinese Academy of Sciences (CAS), Lanzhou, China, in He is currently a Professor with the Cold and Arid Regions Environmental and Engineering Research Institute, CAS. His primary research interests include land data assimilation, application of remote sensing and GIS in hydrology and cryosphere science, and integrated watershed modeling.