Regional Potential Wind Erosion Simulation Using Different Models in the Agro-Pastoral Ecotone of Northern China

Wind erosion is crucial for assessing regional ecosystem services and sustainable development. The Agro-Pastoral Ecotone of northern China (APEC) is a typical region undergoing wind erosion and soil degradation. In this study, the National Wind Erosion Survey Model of China, the Integrated Wind Erosion Modeling System, and the regional versions of the Revised Wind Erosion Equation and Wind Erosion Prediction System were used to evaluate the regional potential wind erosion of the APEC during 2000 and 2012. The results showed that the potential wind erosion predicted by National Wind Erosion Survey Model of China (NWESMC), Revised Wind Erosion Equation (RWEQ), Wind Erosion Prediction System (WEPS), and Integrated Wind Erosion Modeling System (IWEMS) were significantly related to the observed wind erosion collected from published literature, but the observed data were generally smaller than the predicted values. The average potential wind erosions were 12.58, 25.87, 52.63, and 58.72 t hm−2 a−1 for NWESMC, RWEQ, WEPS, and IWEMS, respectively, while the spatial pattern and temporal trend of annual potential wind erosion were similar for different wind erosion models. Wind speed, soil moisture, and vegetation coverage were the dominant factors affecting regional wind erosion estimation. These results highlight that it is necessary to comprehensively calibrate and validate the selected wind erosion models. A long-term standard wind erosion monitoring network is urgently required. This study can serve as a useful reference for improving wind erosion models.


Introduction
From most arid and semi-arid regions to some high-latitude and high-altitude areas, wind erosion is one of the main processes causing soil loss and dust emission, which further results in soil degradation and reductions in land productivity [1][2][3][4]. Chappell [5] called on the academic community to consider and deal with wind erosion and dust emissions more widely. Onsite direct observation [6][7][8], rare earth elements or radio isotope (such as 137Cs, 7Be) surveys [9,10], and wind erosion modeling [11,12] were generally summarized as the main methods to quantitatively determine wind erosion. Of those, wind erosion models were considered the most efficient method to obtain regional patterns of wind erosion [13]. Many wind erosion models were developed to evaluate wind erosion at different scales for different land use [14,15]. APEC refers to the transitional zone between the monsoon region in eastern China and the arid and semi-arid region in northwest China. Although the detailed boundaries of APEC might differ for different research purposes, the core areas are roughly the same [37]. In this study, the boundary described by Guo [32] was used to define the geographical scope of APEC. The study area is approximately 36 • 30 N-46 • 42 N, 106 • 16 E-124 • 51 E (Figure 1), including Inner Mongolia, Liaoning, Jilin, Hebei, Shanxi, and Shaanxi provinces, covering an area of 543,616 km 2 . The land use types in the study area are various, mainly including arable land, forest land, grass land, and sand land. Most of the research area is semi-arid, with an average annual precipitation of 300-450 mm. The precipitation mainly occurs in summer and autumn. The average wind speed in this area is 2.49 m s −1 , and the maximum wind speed is 16-24 m s −1 . Because of the sensitivity of the ecological environment, desertification has developed rapidly in this area.

Wind Erosion Models 2.2.1. The NWESMC Model
The NWESMC model was developed based on the environment in northern China and wind tunnel experiments. This model was first applied to the evaluation of the Beijing-Tianjin Sand Source Control Engineering Project [30], then corrected according to the national large-scale survey data, and was improved and used in the first water conservancy survey [38]. The model establishes the empirical equations of arable land, grass land

The NWESMC Model
The NWESMC model was developed based on the environment in northern China and wind tunnel experiments. This model was first applied to the evaluation of the Beijing-Tianjin Sand Source Control Engineering Project [30], then corrected according to the national large-scale survey data, and was improved and used in the first water conservancy survey [38]. The model establishes the empirical equations of arable land, grass land (forest land), and sand land in Equations (1)-(3), respectively. The actual soil wind erosion modulus of farmland [30] is as follows in Equation (1) (1) The wind erosion modulus of grassland (forest land) is: The wind erosion modulus of sandy land is: where Qfa is the wind erosion of farmland (t hm −2 a −1 ), Qfgf is the wind erosion of grassland (forest land) (t hm −2 a −1 ), and Qfs is the wind erosion of sandy land (t hm −2 a −1 ). C is the scale revision factor (C = 0.0018), Uj is the j-th wind speed (m s −1 ) higher than the critical erosion wind speed in the hourly wind speed statistics of meteorological stations, Tj is the cumulative time (min) when the wind speed, Uj, occurs in the month when the wind erosion activity occurs, VC is the vegetation coverage (%), Z0 is the surface aerodynamic roughness (cm), W is the topsoil humidity factor (%), A is the wind speed revision coefficient related to the underlying surface properties, a1, b1, and c1 are constant terms, with values of −9.208, 0.018, and 1.955, respectively, The values of a2, b2 and c2 are 2.4869, −0.0014, and −54.947, respectively. The values of a3, b3, and c3 are 6.1689, −0.0743, and −27.9613, respectively.
In NWESMC, potential wind erosion of half a month was computed, and the annual potential wind erosion was calculated by the half-monthly potential wind erosion.
The wind erosion modulus of grassland (forest land) is: The wind erosion modulus of sandy land is: where Q fa is the wind erosion of farmland (t hm −2 a −1 ), Q fgf is the wind erosion of grassland (forest land) (t hm −2 a −1 ), and Q fs is the wind erosion of sandy land (t hm −2 a −1 ). C is the scale revision factor (C = 0.0018), U j is the j-th wind speed (m s −1 ) higher than the critical erosion wind speed in the hourly wind speed statistics of meteorological stations, T j is the cumulative time (min) when the wind speed, U j , occurs in the month when the wind erosion activity occurs, VC is the vegetation coverage (%), Z 0 is the surface aerodynamic roughness (cm), W is the topsoil humidity factor (%), A is the wind speed revision coefficient related to the underlying surface properties, a 1 , b 1 , and c 1 are constant terms, with values of −9.208, 0.018, and 1.955, respectively, The values of a 2 , b 2 and c 2 are 2.4869, −0.0014, and −54.947, respectively. The values of a 3 , b 3 , and c 3 are 6.1689, −0.0743, and −27.9613, respectively.
In NWESMC, potential wind erosion of half a month was computed, and the annual potential wind erosion was calculated by the half-monthly potential wind erosion.

The RWEQ Model
RWEQ is one of the most commonly used empirical models for estimating wind erosion in farmland [39]. Because of the full consideration of climate and surface factors, data are easy to obtain and have been widely used; the RWEQ model has been applied to wind erosion areas in China, and the estimation results have been generally verified [40,41]. The basic governing equation of RWEQ is as follows in Equation (4) [39]: where Q x is the sediment flux at the block length × (distance from the upwind direction) (kg m −1 ), Q max is the maximum sediment transport capacity of the wind force (kg m −1 ), and s is the key block length (m): where WF is a weather factor (kg m −1 ), EF is a soil erodible component (dimensionless), SCF is soil crust factor (dimensionless), K is the soil roughness factor (dimensionless), COG is the combined crop factor, including growing vegetation and withering vegetation (dimensionless). The weather factor (WF) can be calculated using the following equation: where U 2 is the wind speed (m s −1 ) at a height of 2 m, which can be converted from wind speed observed at standard anemometer heights using the 1/7 power expression method [11]. Ut is the threshold wind speed (m s −1 ) at a height of 2 m; Guo suggested that the threshold wind speed for arable land in north-central parts of China is 5 m s −1 [42]. N is the observation times of wind speed, N d is the number of days in the period (usually 15 days), g is gravitational acceleration (m s −2 ), SW is a soil wetness factor (dimensionless), and SD is a snow cover factor (dimensionless). The soil erodibility factor (EF) and the soil crust factor (SCF) can be calculated as follows: where Sa is the soil sand content (%), Si is the soil silt content (%), Sa/Cl is the ratio of soil sand and clay content, OM is the organic matter content (%), and CaCO 3 is the calcium carbonate content (%). Because of the small interannual variation of soil texture and organic matter content, it is assumed that the soil erodibility and crust factors will not change with time.
The combined crop factor (COG) is determined by the flat residues (SLR f ), standing residues (SLR s ) and crop canopy (SLR c ) factors: In the APEC region, crop residues are generally used for heating and fuel and most cropland contains no crop residue [32]. Therefore, the wind erosion is governed by vegetation cover, and the COG is determined by SLR c [27,32,43]. SLR c is calculated as in Equation (11): where cc is the fraction of soil surface covered with crop canopy. The cc is obtained from the normalized difference vegetation index (NDVI) [44]: where NDVI is the normalized difference vegetation index, NDVI s is the value of bare soil pixels, and NDVI v is the value of vegetated pixels. In RWEQ, the potential wind erosion of half a month was calculated, and the annual potential wind erosion was determined from the half-monthly potential wind erosion.

The WEPS Model
WEPS-a physics-based model-can simulate weather, surface conditions, field management, and wind erosion in time steps less than daily (e.g., hourly) [45,46]. The WEPS manual stipulates that, when the maximum wind speed at a height of 10 m exceeds 8 m s −1 , the wind erosion submodule will run [47]. The WEPS has been successfully extended to non-agricultural disturbed lands for simulating regional potential wind erosion in Western U.S. and Northern China [48][49][50][51]. The calculation of each step and the required basic equations are as follows: where Q is the emission transport capacity (kg m −1 s −1 ), u * is the friction velocity (m s −1 ), and u *t is the static threshold friction velocity (m s −1 ). It reveals that the driving force of sand transport is when the friction velocity of u * is greater than the static threshold friction velocity of u *t . Friction velocity at the sub-region is calculated in two steps. First, the friction velocity at the weather station, where wind speeds are measured, is calculated using the loglaw profile: where u *f is friction velocity at the weather station (m s −1 ), u is wind speed at the weather station (m s −1 ), z is anemometer height at the weather station (mm) (wind speeds were adjusted to 10 m height in the WEPS database), z 0f is aerodynamic roughness at the weather station, which is assumed to be 25 mm in WEPS. Second, when there is no vegetation, the calculation method of friction velocity is as follows: where z 0 is the local aerodynamic roughness as in [38]. The static threshold friction velocity takes into account the surface soil texture, flat biomass and surface wetness, and the calculation formula is as follows: u * t = WUB * ts + WUC * ts + WUCW * ts (16) where WUB *ts is the static threshold friction velocity of bare surface (m s −1 ); the minimum static threshold friction velocity for field surfaces was generally set to be 0.35 m s −1 [47]; WUC *ts is the change in static threshold friction velocity caused by flat biomass cover (m s −1 ), and WUCW *ts is the increase in static threshold friction velocity from surface wetness (m s −1 ). Because of the large area of APEC, some soil texture data are difficult to obtain, resulting in a large error of WUB *ts . Previous studies [52] replaced u *t in the SWEEP model with u *t as in Lu and Shao [53] and have been well verified. Therefore, this study replaced WUB *ts with u *t on smooth and dry surfaces in the IWEMS model. The formula for WUB *ts is as follows: where δ p is the particle-to-air density ratio, d is the particle diameter (m), g is the acceleration of gravity, 9.8 m s −2 , ρ is the air density (kg m −3 ), β 1 is 0.0123, and β 2 is 3 × 10 −4 kg s −2 .
The formula for WUC *ts is as follows: where SFC cv is the fraction change in soil surface area protected from emission, SF cv is the soil fraction covered by clod/crust and rock so it does not emit, and BFF cv is the biomass fraction of flat cover. The formula for WUCW *ts is as follows: where HROwc is the surface soil water content (kg kg −1 ), and HR15wc is the surface soil water content at 1.5 MPa (kg kg −1 ). In WEPS, the daily potential wind erosion was evaluated, and the annual potential wind erosion was calculated from the daily potential wind erosion.

The IWEMS Model
IWEMS is a model developed based on arid and semi-arid regions in Australia to predict wind erosion processes at regional and national scales [54]. Recently, the IWEMS were widely used to evaluate regional wind erosion and dust emission across the northern China [33,55,56]. The streamwise saltation flux Q(ds) (kg m −1 s −1 ) for soil of uniform particle size ds can be estimated using Owen's model [57]: where A e is the fraction of erodible area and Co is the Owen coefficient. In theory, Co is not a constant but dependent on ω t (ds)/u * , equal to 0.25 + ω t (ds)/3u * in Owen's original formulation. The typical value of Co is around one, but with a considerable scatter.
Threshold friction velocity is estimated by: where u *t (ds,λ,θ) is the threshold friction velocity of sand particles with diameter ds in the presence of vegetation and soil moisture (m s −1 ), λ is the frontal area of the roughness element (m 2 ), f λ (λ) is a function that modifies the threshold friction velocity to reflect the roughness elements, θ is the volumetric soil moisture (m 3 m −3 ), f w (θ) is a function that corrects threshold friction velocity for soil moisture, and u *t (ds) is the threshold friction velocity under the ideal condition that the surface is covered by loose sand particles of uniform and spherical shape. The threshold friction velocity under ideal conditions, u *t (ds) can be expressed by an equation proposed by Shao [54]: where δ p is the particle-to-air density ratio, ds is the particle diameter (m), g is the acceleration of gravity, 9.8 m s −2, ρ is the air density (kg m −3 ), β 1 is 0.0123 and β 2 is 3 × 10 −4 kg s −2 .
The formula for f λ (λ) is as follows: where m r is a tuning parameter with a value less than one, which accounts for nonuniformities in the surface stress distribution, δ r is the ratio of basal to frontal area (δ r = η/λ) of the roughness elements, and β r = C p /C s is the ratio of the pressure-drag coefficient to the friction-drag coefficient.
The formula for f w is as follows: where θ r is air-dry soil moisture (m 3 m −3 ), and A and b are dimensionless parameters.
In IWEMS, daily potential wind erosion was evaluated, and the annual potential wind erosion was determined from the daily potential wind erosion.

Data Preparation
The meteorological data used in this research were obtained from the national station data from 2000 to 2012 provided by the China Meteorological Data Service Center (CMDC) (http://data.cma.cn, accessed on 10 July 2020), of which the meteorological data mainly include wind speed, wind direction, temperature, precipitation, and sunshine hours. NDVI data were obtained from MODIS data products provided by USGS website (https://www. usgs.gov, accessed on 12 July 2020), of which MOD13A2, MOD11A2, and MOD09A1 were used. The soil properties data were mainly determined from the Chinese soil data set of the World Soil Database (HWSD) provided by the Cold and Arid Regions Scientific Data Center (http://westdc.westgis.ac.cn, accessed on 12 July 2020). Digital elevation data used the China 1-km resolution digital elevation model data set provided by the Cold and Arid Regions Scientific Data Center (http://westdc.westgis.ac.cn, accessed on 12 July 2020) Table 1. The land use data of 1-km resolution for 2000, 2005, and 2010 were provided by the Resources and Environment Data Cloud Platform (http://www.resdc.cn, accessed on 15 July 2020). Farmland, grassland, and sands (or desert) were selected to estimate potential wind erosion. Furthermore, we estimated potential wind erosion for a square 1 ha (100 m × 100 m) field [32]. Aerosol optical thickness (AOD) data are provided by the TGP group, Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences (http://www.tgp.ac.cn/, accessed on 15 July 2020) [58]. In addition, the flow chart of this study is presented in Figure 2. In this research, the "Classification Standard of Wind Erosion (SL190-2007)" was used to classify the potential wind erosion hazard (weak, slight, moderate, severe, very severe, or catastrophic) [59].

Potential Wind Erosion for Different Hazards
The potential wind erosion from 2000 to 2012 for the four models with different erosion hazards is shown in Table 2. In NWESMC, more than 80% of the land suffered weak or slight wind erosion in the region, and no catastrophic wind erosion occurred. When compared with NWESMC, the area of "slight" hazard class increased significantly, but the "weak" hazard agricultural decreased considerably in RWEQUATION The percentages of potential wind erosion hazards were similar for WEPS and IWEMS, with about 60% of the land undergoing weak or slight wind erosion and more than 20% of the land suffering very severe and catastrophic wind erosion. In summary, IWEMS generally yielded the highest potential erosion, while NWESMC estimated the lowest potential wind erosion. The average potential wind erosion was 12.58, 25.87, 52.63, and 58.72 t hm −2 a −1 for NWESMC, RWEQ, WEPS, and IWEMS, respectively.

Potential Wind Erosion for Different Hazards
The potential wind erosion from 2000 to 2012 for the four models with different erosion hazards is shown in Table 2. In NWESMC, more than 80% of the land suffered weak or slight wind erosion in the region, and no catastrophic wind erosion occurred. When compared with NWESMC, the area of "slight" hazard class increased significantly, but the "weak" hazard agricultural decreased considerably in RWEQUATION The percentages of potential wind erosion hazards were similar for WEPS and IWEMS, with about 60% of the land undergoing weak or slight wind erosion and more than 20% of the land suffering

Spatial Variation of Potential Wind Erosion
The geographic distributions of the potential wind erosion were somewhat similar, but the magnitudes of potential wind erosion were different for different models (Figure 3). For NWESMC, very severe wind erosion scattered in the Horqin Sands and the Hunshan Dake Sands with a sporadic distribution pattern. Moderate wind erosion occurred in the Mu Us Sands. For RWEQ, the highest potential erosion with catastrophic hazard occurred in the southeast of the APEC, near and within the Horqin Sands. The other high erosion region was mainly scattered in the Hunshan Dake Sands. The regional patterns of wind erosion hazard for WEPS and IWEMS were markedly different from those of NWESMC and RWEQ, the catastrophic and very severe hazard wind erosion regions appeared in the Hunshan Dake Sands and Mu Us Sands, while the Horqin Sands still suffered catastrophic wind erosion.

Spatial Variation of Potential Wind Erosion
The geographic distributions of the potential wind erosion were somewhat similar, but the magnitudes of potential wind erosion were different for different models ( Figure  3). For NWESMC, very severe wind erosion scattered in the Horqin Sands and the Hunshan Dake Sands with a sporadic distribution pattern. Moderate wind erosion occurred in the Mu Us Sands. For RWEQ, the highest potential erosion with catastrophic hazard occurred in the southeast of the APEC, near and within the Horqin Sands. The other high erosion region was mainly scattered in the Hunshan Dake Sands. The regional patterns of wind erosion hazard for WEPS and IWEMS were markedly different from those of NWESMC and RWEQ, the catastrophic and very severe hazard wind erosion regions appeared in the Hunshan Dake Sands and Mu Us Sands, while the Horqin Sands still suffered catastrophic wind erosion.

Temporal Distribution of Potential Wind Erosion
Although the potential wind erosion showed obvious annual fluctuation, it de-

Potential Wind Erosion under Different Land Use
Most territory of the APEC was covered by farmland, grassland, and sand land [32]. The average predicted potential wind erosion under different land use varied remarkably ( Figure 5). The average potential wind erosion values of grassland, farmland, and sand land calculated by NWESMC were 1.37, 15.85, and 34.3 t hm −2 a −1 , respectively. The average potential wind erosion values of grassland, farmland, and sand land calculated by RWEQ, WEPS, and IWEMS showed similar successive increases. The ratios between average potential wind erosion of grassland and sand land for NWESMC, WEPS, IWEMS, and RWEQ were 0.04, 0.12, 0.15, and 0.21, respectively.

Potential Wind Erosion under Different Land Use
Most territory of the APEC was covered by farmland, grassland, and sand land [32]. The average predicted potential wind erosion under different land use varied remarkably ( Figure 5). The average potential wind erosion values of grassland, farmland, and sand land calculated by NWESMC were 1.37, 15.85, and 34.3 t hm −2 a −1 , respectively. The average potential wind erosion values of grassland, farmland, and sand land calculated by RWEQ, WEPS, and IWEMS showed similar successive increases. The ratios between average potential wind erosion of grassland and sand land for NWESMC, WEPS, IWEMS, and RWEQ were 0.04, 0.12, 0.15, and 0.21, respectively.

Models' Verification and Applicability
Quantitative regional measured values of soil loss by wind erosion are not available for the APEC. However, some studies describing field soil loss by wind erosion were performed using different methods. To evaluate the performance of the four wind erosion models, 31 observed wind erosion datasets obtained by different observation methods were collected from published literature ( Table 3). The 31 locations of these observed data scatter in the main land use of the APEC (Figure 6). To investigate the relationship between observed wind erosion data and values obtained using the four models, linear regression analysis and Sutcliffe efficiency coefficient (NSC) analysis were conducted (Figure 7). The coefficients of determination (R 2 ), NSC, and the magnitudes of the slopes obtained through the SPSS and used to evaluate how well these models predict wind erosion [60,61]. The R 2 values suggest that predicted wind erosion was linearly related to measured wind erosion. The magnitudes of the slopes indicate that predicted wind erosion was generally smaller than observed wind erosion (Figure 7). The values of the R 2 (p < 0.05) and NSC demonstrated that the performance of RWEQ, WEPS, and IWEMS were relatively satisfactory, while performance of the NWESMC was relatively poor. Furthermore, the annual average potential wind erosions are 12.82, 26.97, 54.23, and 61.14 t hm −2 a −1 for NWESMC, RWEQ, WEPS, and IWEMS, which are similar to or slightly higher than the previous studies [62,63]. Figure 8 presented the spatial distribution and interannual variation the MODIS satellite AOD data from 2003 to 2010. High AOD generally scattered in the Mu Us Sands, Horqin Sands, and the Hunshan Dake Sands, which were similar to the

Models' Verification and Applicability
Quantitative regional measured values of soil loss by wind erosion are not available for the APEC. However, some studies describing field soil loss by wind erosion were performed using different methods. To evaluate the performance of the four wind erosion models, 31 observed wind erosion datasets obtained by different observation methods were collected from published literature ( Table 3). The 31 locations of these observed data scatter in the main land use of the APEC (Figure 6). To investigate the relationship between observed wind erosion data and values obtained using the four models, linear regression analysis and Sutcliffe efficiency coefficient (NSC) analysis were conducted (Figure 7). The coefficients of determination (R 2 ), NSC, and the magnitudes of the slopes obtained through the SPSS and used to evaluate how well these models predict wind erosion [60,61]. The R 2 values suggest that predicted wind erosion was linearly related to measured wind erosion. The magnitudes of the slopes indicate that predicted wind erosion was generally smaller than observed wind erosion (Figure 7). The values of the R 2 (p < 0.05) and NSC demonstrated that the performance of RWEQ, WEPS, and IWEMS were relatively satisfactory, while performance of the NWESMC was relatively poor. Furthermore, the annual average potential wind erosions are 12.82, 26.97, 54.23, and 61.14 t hm −2 a −1 for NWESMC, RWEQ, WEPS, and IWEMS, which are similar to or slightly higher than the previous studies [62,63]. Figure 8 presented the spatial distribution and interannual variation the MODIS satellite AOD data from 2003 to 2010. High AOD generally scattered in the Mu Us Sands, Horqin Sands, and the Hunshan Dake Sands, which were similar to the spatial distribution of severe wind erosion (Figure 3). In addition, the interannual variations of AOD is also consistent with the trends of annual potential wind erosion obtained from the four models ( Figure 4).    (Table 3).     (Table 3).  (Table 3).
For the model structure, the NWESMC and RWEQ belonged to the empirical model, and the WEPS and IWEMS were mechanistic models [34]. These models were widely used for regional wind erosion evaluation in the arid and semi-arid China to date [64]. In this study, the version of the NWESMC was improved according to the national largescale survey data in the first soil and water conservation survey [38]. Abundant studies demonstrated that the RWEQ is capable of modeling daily, monthly, and annual (potential) wind erosion across field, regional, and global scales after some adjusting [32,64,65]. Here, we used the up-scaling method of RWEQ proposed by Guo [32]. Determining the friction threshold wind velocity (u *t ) is the key step when using the WEPS [48,49,51]. The regional version of the WEPS was used to calculate daily u *t and further model potential wind erosion in the APEC. The IWEMS can incorporate geographic information systems and remote sensing data to estimate regional wind erosion. Du adjusted the IWEMS for extending it to northern China based on observed wind erosion data [55]. It revealed that the IWEMS can predict regional wind erosion and dust emission. The revised IWEMS was used to model potential wind erosion in the APEC. The above analysis indicated that the revised or improved models could evaluate the temporal trends and spatial patterns of potential wind erosion in the APEC.
Only small areas showed significant correlation between potential wind erosion and annual average precipitation or temperature (Table 4). However, the areas with significant positive correlation (R > 0.74, p < 0.05) between potential wind erosion and annual average wind speed were 40.68%, 42.37%, 27.01%, and 31.53% of the total territory of the APEC for NWESMC, RWEQ, WEPS, and IWEMS, respectively. The regions with significant negative correlation (R < −0.74, p < 0.05) between potential wind erosion and annual average soil moisture or vegetation coverage were smaller than those with a significant positive correlation but showed similar trends. For NWESMC, the areas with significant positive (negative) correlation between potential wind erosion and wind speed, soil moisture and vegetation coverage decreased, with the same trend observed for RWEQ, WEPS, and IWEMS. These results indicated that the magnitudes of sensitivity of wind speed, soil moisture, and vegetation coverage to regional wind erosion modeling successively reduced.
Wind speed is generally considered as the primary driving factor initiating wind erosion while high vegetation coverage and wet surface soil can significantly lower soil sus-
Only small areas showed significant correlation between potential wind erosion and annual average precipitation or temperature (Table 4). However, the areas with significant positive correlation (R > 0.74, p < 0.05) between potential wind erosion and annual average wind speed were 40.68%, 42.37%, 27.01%, and 31.53% of the total territory of the APEC for NWESMC, RWEQ, WEPS, and IWEMS, respectively. The regions with significant negative correlation (R < −0.74, p < 0.05) between potential wind erosion and annual average soil moisture or vegetation coverage were smaller than those with a significant positive correlation but showed similar trends. For NWESMC, the areas with significant positive (negative) correlation between potential wind erosion and wind speed, soil moisture and vegetation coverage decreased, with the same trend observed for RWEQ, WEPS, and IWEMS. These results indicated that the magnitudes of sensitivity of wind speed, soil moisture, and vegetation coverage to regional wind erosion modeling successively reduced.
Wind speed is generally considered as the primary driving factor initiating wind erosion while high vegetation coverage and wet surface soil can significantly lower soil susceptibility to wind erosion [43,84]. The sensitivity of parameters of WEPS, NWESMC, RWEQ, and IWEMS were determined by the previous studies about [30,84,85]. These results concluded that wind speed was the most sensitive input parameter and vegetation and soil moisture could curb the wind erosion. Vegetation with different land-use or soil moisture was also important for WEPS regional wind erosion modeling [51]. Shao revealed that wind speed (friction wind velocity) and vegetation coverage soil moisture were very sensitive factors when modeling wind erosion with IWEMS [12]. Studies of regional wind erosion modeling by RWEQ indicated that wind speed and the soil crust factor (which occurred after rain) were very sensitive inputs [32,43]. NWESMC used the cumulative time of erosive wind speeds and vegetation coverage to directly calculate soil loss by wind erosion [30]. Consequently, wind speed, vegetation coverage, and soil moisture can be summarized as the main factors affecting simulated regional potential wind erosion. Table 4. Spatial correlation analysis between potential wind erosion and wind speed, soil moisture, vegetation, precipitation, and temperature.

Limitations and Future Perspectives
In fact, the wind erosion predicted by the models was the wind erosion potential or not the real on-site wind erosion in this study. More detail model calibration may be necessary based on a long-term observed wind erosion dataset to obtain more reliable large-scale wind erosion. These wind erosion models with diverse computation structure originated from different countries based on various geographic conditions. The selection of a particular model is generally dependent on the available databases and the specific requirements [15]. In practice, these models have been widely used in regional wind erosion assessment. Some studies calibrated [55,86] or validated [44,55,87] the models for regional wind erosion estimation. However, other studies directly evaluated regional wind erosion without calibration and validation of wind erosion models [26,27,32,48,64,88,89]. Here, we demonstrated that various values of potential wind erosion were yielded by different models using the same available databases. Accordingly, it is necessary to systematically calibrate and validate the selected wind erosion model before extending it to regions with differing geographic conditions. Theoretically, the reliable calibration and validation of a wind erosion model requires a long-term observed wind erosion dataset with a consistent measurement method. In water erosion research, experimental plots have been the standard method since the 1930s [35]. In contrast, aeolian sand samplers have been deployed and extensively used for various different purposes. For example, Chepil generally used a rectangle field to monitor sand transport [90]. Wind erosion circles were popular when developing RWEQ and WEPS [11,45]. Webb proposed a national wind erosion research network based on square experimental plots (100 m × 100 m) [1]. Some studies used sand traps with a uniform grid network [65,91]. The above analysis indicates that a long-term standard wind erosion monitoring network is urgently required. The standard network requirements would include the shape of experimental plots, meteorological data observation procedure, field soil sampling method, sand trap type, and its deployment scheme [13,34,92].

Conclusions
In this study, NWESMC, RWEQ, WEPS, and IWEMS were used to simulate the temporal and spatial pattern of the potential wind erosion in the APEC from 2000 to 2012. The impacts of meteorological data, soil moisture, and vegetation coverage on potential wind erosion were discriminated and quantified. The main conclusions were: 1.
The potential wind erosion values predicted by the four models were correlated with the observed wind erosion collected from published documents, but the correlation coefficients between the predicted and the measured wind erosion data for the four models vary greatly; 2.
The values of average potential wind erosion were different while the spatial pattern of potential wind erosion was similar for different wind erosion models. Most areas of APEC suffered from weak and slight hazards of wind erosion, while severe and catastrophic hazards of wind erosion mainly occurred in the Horqin, Mu Us, and Hunshan Dake sands; 3.
The temporal trends of annual potential wind erosion were similar and the total potential wind erosion decreased significantly from 2000 to 2012; 4.
The average potential wind erosion of grassland, farmland, and sand land calculated by NWESMC, RWEQ, WEPS, and IWEMS showed similar successive increases.

5.
Wind speed, soil moisture, and vegetation coverage were the dominant factors affecting regional wind erosion estimation.
These results further revealed that it is necessary to comprehensively calibrate and validate the selected wind erosion models. A long-term standard wind erosion monitoring network is urgently required.