Coupling numerical simulation with remotely sensed information for the study of frozen soil dynamics

HuiRan Gao,WanChang Zhang

1.Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China

2.University of Chinese Academy of Sciences,Beijing 100049,China

ABSTRACT The acquisition of spatial-temporal information of frozen soil is fundamental for the study of frozen soil dynamics and its feedback to climate change in cold regions. With advancement of remote sensing and better understanding of frozen soil dynamics,discrimination of freeze and thaw status of surface soil based on passive microwave remote sensing and numeri‐cal simulation of frozen soil processes under water and heat transfer principles provides valuable means for regional and global frozen soil dynamic monitoring and systematic spatial-temporal responses to global change.However,as an impor‐tant data source of frozen soil processes,remotely sensed information has not yet been fully utilized in the numerical simu‐lation of frozen soil processes.Although great progress has been made in remote sensing and frozen soil physics, yet few frozen soil research has been done on the application of remotely sensed information in association with the numerical model for frozen soil process studies. In the present study, a distributed numerical model for frozen soil dynamic studies based on coupled water-heat transferring theory in association with remotely sensed frozen soil datasets was developed.In order to reduce the uncertainty of the simulation, the remotely sensed frozen soil information was used to monitor and modify relevant parameters in the process of model simulation.The remotely sensed information and numerically simulat‐ed spatial-temporal frozen soil processes were validated by in-situ field observations in cold regions near the town of Naqu on the East-Central Tibetan Plateau.The results suggest that the overall accuracy of the algorithm for discriminating freeze and thaw status of surface soil based on passive microwave remote sensing was more than 95%.These results pro‐vided an accurate initial freeze and thaw status of surface soil for coupling and calibrating the numerical model of this study.The numerically simulated frozen soil processes demonstrated good performance of the distributed numerical mod‐el based on the coupled water-heat transferring theory. The relatively larger uncertainties of the numerical model were found in alternating periods between freezing and thawing of surface soil.The average accuracy increased by about 5%af‐ter integrating remotely sensed information on the surface soil.The simulation accuracy was significantly improved,espe‐cially in transition periods between freezing and thawing of the surface soil.

Keywords:frozen soil;water-heat coupled model;passive microwave remote sensing;coupling;frozen soil dynamics

1 Introduction

As an essential component of the cryosphere, fro‐zen soil is one of the key drivers for and core links in the interaction between ground surface and global cli‐mate system. Studies on frozen soil dynamics are ba‐sic for ecology, research in climate change, as well as establishment of future climate scenarios or forecast of future climate (Shi and Cheng, 1991; Qin and Ding,2009; Guglielminet al., 2018). Under the background of global change, the frozen soil has been degrading,leading to the change of ecological diversity, the deg‐radation of ecological functions and many other envi‐ronmental problems (Jinet al., 2000; Jorgensonet al.,2001;Yanget al.,2013;Gaoet al.,2020).

The acquisition of spatial-temporal frozen soil pro‐cesses is the basis for the study of climate, ecological change and hydrological processes in cold regions.Traditional studies experienced significant challenges under increased demands for fast information extrac‐tion and spatial parameterization on frozen soil dy‐namics (Cao and Zhang, 1997; Dinget al., 2000;Yaoet al., 2013). The advancement of remote sensing technology and better understanding of frozen soil physics provide a valuable means for frozen soil dy‐namic studies. Since the 1970s, significant achieve‐ments have been made in the study of seasonal frozen soil spatial-temporal dynamics at regional or global scales by using passive microwave remote sensing technology (Morrisseyet al., 1986; Zhang and Arm‐strong, 2001; Jinet al., 2009). With in-depth study of the mechanism on water-heat transferring dynamics in frozen soil processes, various numerical model for simulation of frozen soil processes have been devel‐oped, among which one-dimensional and most repre‐sentative ones are Soil Heat and Water model (SHAW)proposed by Flurchingeret al. (1989); a distributed Variable Infiltration Capacity model(VIC)for large-scale hydrological process simulation developed by Woodet al. (1992) and the Distributed Water and Heat Cou‐pling model coupled with a hydrological process mod‐el established by Chenet al. (2006). Based on the CLM4 model,Guo and Wang(2013)extended and en‐hanced the simulation of hydrological process and snow melting process. The modified model was ap‐plied and verified in the simulation of the Qinghai Ti‐bet Plateau in China. Gaoet al. (2018) improved the processes of glacier melting, snow melting and soil freezing and thawing on the basis of the GBHM mod‐el.This method was applied to evaluate the eco-hydro‐logical effects of frozen soil degradation in the upper reaches of Heihe River Basin and the source area of the Yellow River, China. However, numerical simula‐tions coupled with remotely sensed information on frozen soil processes are still in an exploratory stage,although great efforts have been made in recent de‐cades. However, large uncertainty remains a big chal‐lenge in relevant studies (Yang and Chen, 2011; Chenet al., 2014), among which the spatial-temporal varia‐tions of frozen soil distribution and the associated changes of heat and water parameters were attributed to the biggest resources leading to those uncertainties(Pomeroyet al., 2007; Ouet al., 2016; Gaoet al.,2018). Remotely sensed information with new tech‐nology in association with distributed new generation of numerical models is urgently needed for fulfillment of the requirement for 3-demonsional frozen soil dy‐namic simulations.

In association with remotely sensed surface frozen soil spatial-temporal distribution variations, a distrib‐uted numerical model for 4-demensional frozen soil process simulations under the principle of water and heat transfer mechanism in frozen soil by considering the influence of surface soil freeze and thaw status on the processes of water and heat transferring processes was developed in the present study. The distributed numerical model was developed to simulate character‐istic parameters of frozen soil processes, such as the spatial-temporal distribution variations of the frozen soil, depth of the frozen soil, soil ice content and so on. The present study is aimed to examine further the integration of remotely sensed information on surface frozen soil in the developed distributed numerical model to reduce uncertainty of the numerical model and improve the accuracy of frozen soil process simu‐lations. We also explore the rationality and feasibili‐ty of the distributed numerical model in association with remotely sensed information for regional and global frozen soil dynamic studies and its possible spatial-temporal responses to global change.

2 Material and methods

2.1 Methodology

The present study mainly includes three research contents. 1) Discriminating the surface soil freeze and thaw status with the improved Dual-index Algorithm(DIA) by using passive microwave remote sensing technology; 2) Establishing a distributed numerical model for frozen soil process simulation based on wa‐ter and heat transferring mechanism;3)Integrating re‐motely sensed information on surface frozen soil in the developed distributed numerical model. The spe‐cific methods and workflow for the purpose of the study is presented in Figure 1, and briefly introduced as follows.

As a widely used algorithm for discriminating sur‐face soil freeze and thaw status, Dual-index Algo‐rithm (DIA) was selected and improved in this study.The AMSR-E passive microwave remote sensing brightness temperature products were used, and for detail of the algorithm, refer to Gaoet al. (2018). In association with remotely sensed surface frozen soil spatial-temporal distribution variations, a distributed numerical model for 3-demensional frozen soil pro‐cess simulations under the principle of water and heat transfer mechanism in frozen soil by consider‐ing the influence of surface soil freeze and thaw sta‐tus on the processes of water and heat transferring processes was developed. This model was developed with support information of DEM, soil content and texture map,vegetation and meteorological data as in‐puts in the study area. Finally, initial distribution of surface soil freeze and thaw status derived from pas‐sive microwave remote sensing was input into the numerical model by using parameterized scheme with other support information to drive the simula‐tion. Spatial-temporal variations provided by remote sensing in the process of simulation were used to guide the model simulation process to reduce model uncertainties.

Figure 1 Overall framework of present study

2.2 Materials

2.2.1 Study area

Soil temperature and moisture conditions directly affect the distribution of frozen soil, hydrological pro‐cesses,and evolution of the ecological environment in cold regions. The monitoring and study of frozen soil in plateaus has drawn extensive attention in recent de‐cades. As the highest altitude and cold plateau in the world, permafrost and seasonal frozen soil extensive‐ly developed in the Qinghai-Tibet Plateau where nu‐merous studies on frozen soil have been carried out in this typical cryosphere region (Jiaoet al., 2016). In the present study, an area near the town of Naqu on the East-Central Tibetan Plateau with an area of about 2.5×106km2above an average elevation of 4,000 m a.s.l.was selected as an experiment site(Figure 2).A multiscale and intensive Soil Moisture/Temperature Moni‐toring Network (SMTMN) has been set up and man‐aged by the Chinese Academy of Sciences in this ar‐ea, which provided valuable in-situ observation data on frozen soil processes for various frozen soil studies(Qinet al.,2013;Yanget al.,2013).

2.2.2 Datasets of microwave satellites

AMSR-E is a passive microwave radiometer sys‐tem onboard NASA's (the National Aeronautics and Space Administration)Aqua satellite, which measures horizontally and vertically polarized brightness tem‐peratures at 6.9, 10.7, 18.7, 23.8, 36.5, and 89.0 GHz.AMSR-E soil moisture product was retrieved by us‐ing C band (6.9 GHz) and X band (10.7 GHz) bright‐ness temperature according to the algorithm proposed by Njokuet al. (2003).The retrieval accuracy of AM‐SR-E soil moisture product is less than the root mean square error (RMSE) of 0.06 m3/m3.Attenuation from vegetation increases the retrieval error in soil mois‐ture due to the effect of moisture contained in vegeta‐tion. Fortunately, only sparse vegetation, mainly in‐cluding alpine shrub,alpine meadow and alpine grass‐land cover the study site, which guarantees accuracy of the AMSR-E soil moisture product for this region(Yanget al.,2008).

In this study, AMSR-E daily brightness tempera‐ture product in the Equal-Area Scalable Earth Grid(EASE-Grid) format with a 0.25° spatial resolution(National Snow and Ice Data Center Distributed Ac‐tive Archive Center,http://nsidc.org/data/AE_Land3/versions/2) were utilized. The AMSR-E data products used in the discrimination of freeze and thaw status of surface soil are the 18.7 GHz and 36.5 GHz vertically polarized brightness temperature and soil moisture content.

Figure 2 Geographic location and the SMTMN set up in the study area

2.2.3 Ground-based measurements

In-situ observations obtained from local meteoro‐logical stations and SMTMN in this study mainly in‐clude meteorological factors such as temperature, air pressure, precipitation, wind speed, sunshine duration,soil temperature and soil moisture content, spanning from the August 1, 2010 to July 31, 2011.Also, mete‐orological data were derived from daily observation data of five stations provided by the National Meteo‐rological Information Center (http://data.cma.cn/). Me‐teorological station data were interpolated into grid da‐ta by using the Inverse Distance Weighted(IDW)inter‐polation scheme.

Soil temperature and soil moisture content data were derived from SMTMN which covers 56 stations distributed over an average elevation of 4,500 m a.s.l.with three spatial resolutions(1.0°,0.3°,0.1°gridded)at four soil depths (5, 10, 20 and 40 cm). The 1.0°gridded ground-based observations from 36 stations presented in Figure 2 were selected in this study. In validation of the discriminated freeze and thaw status of surface soil by the improved DIA, only one station was reserved in each AMSR-E passive microwave re‐motely sensed data grid owning to coarse resolution of AMSR-E data. In total, 12 retained stations were chosen.

2.3 Methods

2.3.1 Improved DIA

The DIA proposed by Zuerndorferet al. (1990)took the best use of two brightness temperature pa‐rameters remotely sensed with passive microwave sat‐ellite for discriminating surface soil status. The algo‐rithm has been widely used, and many satisfied re‐sults has been achieved (Zuemdorfer and England,1992; Zhanget al., 2001). However, as pointed out,ample space still remains to improve the original DIA algorithm (Jinet al., 2009; Chaiet al., 2014; Caoet al., 2017). Gaoet al. (2018) improved the algorithm by taking local variance of soil moisture (LVSM) into account. The improved algorithm can be summarized in Equations(1),(2)and(3).

whereTb37vrepresents vertical polarization brightness temperatures at 37 GHz (K), andthe nega‐tive spectral gradient between 19 GHz and 37 GHz(K/GHz);ηis the pixel value of soil moisture at pixel(i,j) at timep;λis a fixed partial time span of the time series used to calculate local variance, which was set to be 25 in this study; and,μλis the average value of soil moisture over the time span.P37,PSGandPSMstands for the thresholds of three indicators ofandLVSM, respectively. Usually, the value ofPSGis set to zero.When the condition for Equa‐tion(1)and Equation(2)are satisfied simultaneously,the ground surface can be identified as either freezing or thawing.

The DIA has different degrees of discrimination errors in the period of soil thawed or frozen (not in‐cluding the period of soil freezing and thawing).How‐ever, discrimination errors in these periods were easi‐er to identify than those in the period of soil freezing and thawing. Therefore, the discrimination error in the period of soil thawed or frozen was corrected in this study. According to existing research, the initial regime of soil freezing is defined as the first day that surface soil remained frozen for five successive days,while the initial regime of soil thawing was defined as the first day that the surface soil remained thawed for five successive days (Jinet al., 2009). The initial day of soil freezing and thawing was obtained in the study period based on the improved DIA algorithm.The soil state is considered as frozen in the period between ini‐tial days of soil freezing and thawing, otherwise, it is considered as thawed.

2.3.2 The distributed numerical model of frozen soil processes

The hydrological processes under the condition of frozen soil is very complicated with its own partic‐ularity (Xuet al., 2010). Freezing-thawing alterna‐tive changes with water-heat transferring are impor‐tant hydrological processes in cold regions through‐out watershed processes of runoff generation, infil‐tration and evapotranspiration (Flerchinger and Sax‐ton,1989;Yang and Chen,2011;Luet al.,2017).Nu‐merous hydrothermal processes, such as meteorolo‐gy,vegetation canopy,snow cover and soil,have been considered in the design and development of the dis‐tributed numerical model of frozen soil processes in this study. The general schematic diagram of the hy‐drothermal process of frozen soil is illustrated in Figure 3.

Figure 3 Hydrothermal processes in physical system of frozen soil

Meteorological conditions are important driving factors in the frozen soil process. Basic meteorologi‐cal parameters, such as temperature, air pressure, pre‐cipitation,humidity,wind speed,evaporation and sun‐shine duration were considered in this regard. Charac‐teristic changes of plants and their canopy were de‐signed as a single layer and single node structure. En‐ergy and water balance equations in the canopy can be represented by Equations(4)and(5)(Flerchingeret al.,2012;Flerchingeret al.,2015).

where,ρa,caandTstand for atmospheric density(kg/m3),atmospheric specific heat capacity (J/(kg?°C)) and can‐opy temperature (°C), respectively.keis canopy heat conductivity (m2/s),Hlis heat transfer from canopy to atmosphere (W/m3).ρvis vapor density (kg/m3), andElis evapotranspiration of canopy(mm).

Snow cover was designed as a single layer and double-node structure.The accumulation and sublima‐tion of snow cover and the conduction process of wa‐ter and heat in the snow layer were parameterized with energy balance equation in snow cover represent‐ed in Equation(6)(Richardet al.,1990).

where,ρsp,wspandkspstand for snow density (kg/m3),volume water content (m3/m3) and heat conductivity of snow (W/(m?°C)), respectively. Theciis specific heat capacity of ice (J/(kg?°C)),ρlis density of liquid water(kg/m3),Rnis net radiation flux under snow cov‐er (W/m2),LfandLsstand for latent heat of melting and sublimation (J/kg), andqvrepresents vapor flux(kg/(s?m2)).

The soil column with different soil layers and depths in different grids as the basic computation unit was designed to account for the effects of soil freez‐ing and thawing process on water and heat transfer‐ring. Energy and water balance equations in the soil column can be described by Equations(7)and(8)(Fuchset al.,1978;Zhou and Li,2012).

where,Csis specific heat capacity of soil (J/(kg?°C)),ksis soil thermal conductivity (W/(m?°C)).Kis soil hydraulic conductivity (m/s),ψis soil water poten‐tial (m), andUis source/sink terms of soil water flux(m3/(m3?s)).

In computation of water heat balance equations,the canopy, snow cover and soil columns were divided into finite layers and each of the layers are represented by a node.The Newton-Raphson iterative method was used to solve the energy and water balance equations for each node(implicit finite difference equation).Be‐cause there are too many formulas in the calculation methods of each unknown variable in the energy and water balance equations, and the corresponding finite difference equations, it is no longer listed in this pa‐per. For detailed equations and formulas, refer to Chenet al. (2006), Flercingeret al. (2015) and Luet al. (2017). The schematic diagram of the numerical model running process is presented in Figure 4.

2.3.3 Integration of remotely sensed information with the developed numerical frozen soil model

Surface soil freezing and thawing status discrimi‐nated by the improved DIA algorithm with passive mi‐crowave remote sensing at initial stage was input into the distributed numerical frozen soil model by using parameterized scheme to guide the simulation foot‐prints and adjust deviations for some specific frozen soil processes.As surface soil temperature directly af‐fects surface soil freezing and thawing status, there‐fore, temperature of the first soil layer was taken as the main corrected variable in the simulations. Ac‐cording to the coupled equations of water-heat trans‐ferring processes,soil temperature of the first soil lay‐er can be corrected with Equation(9)theoretically.

where,Tiis soil temperature of theilayer.Sstands for the states of surface soil, where 1 represents frozen soil and 0 represents thawed soil.The equation means that the numerical model will correct the value of soil surface temperature to 1 or ?1, when the simulated soil surface temperature is higher than 0°C and the re‐mote sensing retrieved surface soil state is frozen,else the simulated surface soil temperature is less than 0 °C and the remote sensing retrieved surface soil state is thawed. The soil layers in the profile below the soil surface will be adjusted accordingly. In this study, the spatial resolution of the simulated grid was 1km×1km.The conventional cubic convolution method was used to resample surface soil freezing and thawing status data to 1km×1km.

Figure 4 Main processes in the running of the numerical model

3 Results

3.1 Soil freeze/thaw states discrimination

The thresholds ofP37,PSGandPSMindicators were set to 256 GHz (K), 0 GHz (K/GHz) and 0.168 m3/m3, respectively, in this study. Daily varia‐tion of surface soil freezing and thawing status in the studied period was obtained by the improved DIA. Simulation verification is summarized in Table 1.The discriminated surface soil status by the im‐proved DIA algorithm was more than 95% on aver‐age, which demonstrated that the improved DIA al‐gorithm was both effective and credible to provide accurate enough results for numerical simulation corrections.

Table 1 Accuracy of the discriminated surface soil status by the improved DIA algorithm over 36 stations

3.2 Frozen soil process simulations

Numerical simulations mainly outputted frozen soil process parameters such as soil temperature, soil water content,soil ice content,soil heat flux and snow cover thickness. It is difficult to show the variations of these variables in the simulated period because of its 4-dimensional spatiotemporal characteristic.There‐fore, the spatiotemporal simulation results of frozen soil processes were plotted from the perspective of re‐gional average as presented in Figure 5 to Figure 8,re‐spectively.In Figure 5,the observed and simulated soil temperatures were plotted by green and red curves for the soil layers in different depths,respectively.Unfortu‐nately, few in-situ observed frozen soil parameters are available for validating the simulated ones although many frozen soil process parameters can be simulated in the model. In the process of numerical simulations,various parameters involved in frozen soil processes were interrelated and influenced with each other. The numerical simulations were considered reliable in gen‐eral since the simulated cone parameter was accurate enough. The correlation coefficient and RMSE be‐tween the simulated soil temperature and the in-situ ob‐served one at different depths was about 0.9 (°C) and 2.8(°C),as presented in Figure 5,respectively.

As presented in Figure 9, two periods (October and April of the next year) in hydrological year of 2011 were selected to display the spatial distribution of the surface frozen soil parameters by taking the upper 5 cm thickness soil layer as an example. From the perspective of spatial distribution, simulation re‐sults of the numerical model have shown a strong spatial heterogeneity and wealth of spatial details.There was no significant heterogeneity in the spatial distribution of soil ice content in October, which was due to the low average soil ice content with a range from 0.00 to 0.03 (m3/m3) in the period of soil freez‐ing and thawing.

The proposed model was validated by using in-si‐tu observed soil temperature from SMTMN. Soil tem‐peratures simulated at different soil depths were con‐sistent with the observed ones at each corresponding depth of 5 cm, 10 cm, 20 cm and 40 cm, respectively.The comparisons between observed and simulated soil temperatures at different depths are presented in Figure 10 by taking two sites (BC02 and MSNQRW)as an example.The statistical results of accuracy veri‐fication suggest that RMSE andR2derived from BC02 and MSNQRW were 3.00, 0.94 and 3.33, 0.89,respectively, which demonstrate rather high accuracy of the developed distributed frozen soil numerical model. However, one shortcoming was also found for unstable model simulations during the period of soil freezing and thawing processes (about November or April of next year, presented on a blue and red back‐ground in Figure 10).The validated results will be dis‐cussed in the next section.

Figure 5 Comparison of soil temperature between observed data and results of the model simulation

Figure 6 Simulation results of soil water content and soil ice content

Figure 7 Simulation results of snow depth and snow density

Figure 8 Simulation results of the soil heat flux and sensible heat flux

Figure 9 Distributed simulation results of the numerical frozen soil model in the soil surface layer

3.3 Verifications of the developed distributed frozen soil numerical model coupled with remotely sensed information

The spatial-temporal information of surface soil status was parameterized with resampling schema for their large-scale characteristics either in freezing or in thawing status.After the developed distributed frozen soil numerical model was calibrated with the ob‐served dataset obtained from SMTMN, the parame‐ters for freezing and thawing surface soil were input into the model coupled with remotely sensed informa‐tion.The footprint of the numerical model was guided by Equation(9). Validation of simulated results for each observatory from SMTMN before and after cor‐rection was applied by using remotely sensed informa‐tion of surface soil as presented in Table 2 and illus‐trated in Figure 11, respectively.R2and RMSE of the numerically simulated frozen soil processes were ana‐lyzed, but only verification results of the simulated soil temperature at 5 cm depth were provided for com‐parison for the sake of limited space.The average val‐ue of verified results at four soil layers are presented at the bottom of Table 2.

It is obvious that after the developed frozen soil process numerical model was corrected with remotely sensed surface soil freezing and thawing information,the overall simulation accuracy of the numerical mod‐el improved dramatically.This is attributed to that the correction of the numerical model was mainly made during surface soil at freezing and thawing alternative periods.

Figure 10 Comparisons between observed and simulated soil temperatures at different depths at BC02(up)and MSNQRW(down)

Figure 11 Accuracy comparisons of the developed numerical model before and after corrections

Table 2 Validation of simulated results for each observation from SMTMN

4 Discussions

As presented in Table 1, the high accuracy of more than 89%for the improved DIA at each observa‐tion from SMTMN, and rather high overall accuracy of 95% proved its superiority in discriminating the surface soil freezing and thawing status.The remotely sensed information with improved DIA algorithm pro‐vided reliable freezing and thawing ground surface soil information for the model simulation and the fol‐lowing further corrections.

The numerical simulations suggests that the re‐gional frozen soil process can be depicted accurately and the spatial components of frozen soil processes can be obtained by the frozen soil process model un‐der water and heat transferring principle. TheR2and RMSE indices of the simulation results were calculat‐ed by using soil temperature data from the ground ob‐servation stations. In soil depths of 5, 10, 20 and 40 cm, the averageR2of simulated soil temperature is 0.89, 0.90, 0.88 and 0.87, respectively, with the aver‐age RMSE of approximately 4.08, 3.88, 3.96 and 4.07,correspondingly. The average accuracy increased by about 5%.This indicates that the model has high sim‐ulation accuracy for the process of frozen soil, espe‐cially for the thermal process.With an increase of soil depth, the simulation accuracy of the model tends to decrease slightly.

Widely scattered patches in the simulated distrib‐uted frozen soil processes presented in Figure 9 re‐veals that soil physical properties determined by soil types affect some soil parameters, hence affect the fi‐nal simulation results.Additionally, surface soil prop‐erties are greatly affected by meteorological condi‐tions, the spatial characteristics of meteorological pa‐rameters are also reflected in the spatial distribution of the simulated frozen soil processes, such as spatial temporal snow cover variations.

To better understand the correction effect of the remotely sensed surface soil freezing and thawing status more clearly, differences of the simulated fro‐zen soil processes in two observatories (BC02 and MSNQRW) from SMTMN before and after the mod‐el were corrected as illustrated in Figure 12. In Fig‐ure 12, the green lines represent the in-situ observed frozen soil process, while the red and blue represent the simulated ones before and after correction was made to the model. It is obvious that the influence of surface soil freezing and thawing status determined by passive microwave remotely sensed information on the simulation of frozen soil processes with the developed model is mainly reflected in the periods of surface soil freezing or thawing status. The great‐er uncertainty in the model simulation also occurred in these periods. Compared with the period of soil freezing process, improvement of the model in the period of soil thawing process was more significant,due to a large simulation error of the model in this period.

Figure 12 Comparison of simulation results before and after corrections at BC02(up)and MSNQRW(down)

Furthermore, Table 2 shows that the accuracy of simulation results of the model has been further im‐proved by coupling the remotely sensed information of surface soil freeze and thaw status. The accuracies of the simulated frozen soil processes for the overall observatories of SMTMN were improved in certain levels, which demonstrated that it was reasonable and feasible to improve the coupled remotely sensed infor‐mation on surface soil status.

5 Conclusions

In the present study, a distributed numerical mod‐el for frozen soil dynamic studies based on coupled water-heat transferring theory in association with the remotely sensed frozen soil datasets was developed.In order to reduce simulation uncertainty, remotely sensed frozen soil information was used to monitor and modify relevant parameters in the process of model simulation. The remotely sensed information and the numerically simulated spatial-temporal frozen soil pro‐cesses were validated by in-situ field observations in cold regions near the town of Naqu on the East-Central Tibetan Plateau.

Our results suggest that the overall accuracy of the modified algorithm for discriminating freeze and thaw status of surface soil based on passive microwave remote sensing was more than 95%, accurate enough for initial freeze and thaw status of surface soil when coupling and calibrating the numerical model of this study. The numerically simulated frozen soil process‐es demonstrated good performance of the distributed numerical model based on the coupled water-heat trans‐ferring theory in general. Relatively larger uncertain‐ties were found in the alternative periods between freezing and thawing of the surface soil.After integrat‐ed with remotely sensed information on surface soil,the simulation accuracy was significantly improved(about 5% overall), especially in the alternative peri‐ods between freezing and thawing of surface soil.

Acknowledgments:

This work was supported by the National Key R&D Program of (Grant No. 2016YFA0602302). Soil tem‐perature and soil moisture datasets utilized in the pres‐ent study was generously provided by Data Assimila‐tion and Modeling Center for Tibetan Multi-spheres,Institute of Tibetan Plateau Research, Chinese Acade‐my of Sciences. We greatly appreciate the National Snow and Ice Data Center for providing the AMSR-E daily brightness temperature products that are publi‐cally accessible through http://nsidc.org/data/AE_Land3/versions/2 to public. We would also wish to thank all the graduate students of WanChang Zhang's group for their helpful comments in our weekly seminars. The revision by Dr.ZhiJie Zhang is acknowledged.