The APEC is located in the transition zone between the monsoon region in eastern China and the arid and semi-arid regions in northwestern China (Guo et al., 2013). The APEC includes parts of Inner Mongolia, Liaoning, Jilin, Hebei, Shanxi, Shaanxi, and Ningxia Provinces (36°30′–46°42′N, 106°16′–124°51′E) (Figure 1B). Most of the APEC is semi-arid. The annual precipitation distribution is uneven and is concentrated in summer and autumn. The average annual precipitation is 300–450 mm and the annual average temperature is 2.4–11.5°C. Strong winds are frequent and mainly occur in spring (March–April–May). The annual average wind speed ranges from 1.3 to 3.9 m s⁻¹, with an average maximum wind speed of 16–24 m s⁻¹. The land uses are diverse and the sandy land (including the Mu Us Sandy Land, Hunshan Dake Sandy Land, and Horqin Sandy Land), grassland, and farmland have staggered distributions in the APEC. In addition, agricultural and animal husbandry activities are still intensive in some region with a fragile ecology, and the human activities and severe wind erosion continuously decrease the local soil productivity in these areas.

In this study, we collected meteorological data, vegetation data, land uses/cover data, Digital Elevation Model (DEM) data, and soil data to calculate the wind erosion in the APEC (Table 1). The data from all of the meteorological stations from 2000 to 2012 were obtained from the China Meteorological Science Data Sharing Service Network (http://data.cma.cn). These meteorological data included wind speed, temperature, precipitation, sunshine duration, and other variables. Two wind databases are available, one dataset contains four wind records per day (UTC+8 2:00, UTC+8 8:00, UTC+8 14:00, and UTC+8 20:00), and the other contains two wind records (average and maximum). The hourly wind speed data were generated from the four wind records per day using the linear interpolation method (Li et al., 2013) and from the two wind records per day using WINDGEN (Skidmore and Tatarko, 1990). The Normalized Difference Vegetation Index (NDVI) data were derived from Moderate Resolution Imaging Spectroradiometer (MODIS) data products obtained from the United States Geological Survey’s website (https://www.usgs.gov), and the MOD13A2, MOD11A2, and MOD09A1 products were mainly used. The 1 km resolution land use/cover data of 2010 was obtained from the Resource and Environment Data Center, Chinese Academy of Sciences (http://www.resdc.cn). The DEM data (China 1 km Low Range Model Data Set) were provided by the Science Data Centre for Cold and Arid Areas (http://westdc.westgis.ac.cn). The soil data were mainly obtained from the world soil database (HWSD) and were provided by the Scientific Data Center (http://westdc.westgis.ac.cn) for Cold and Arid Regions. These data were used to calculate the surface soil factors in the different regions. Farmland, grassland and sands (or desert) were selected to evaluate regional potential wind erosion.

In the APEC, there are 130 meteorological observing stations (reference and basic stations), of which 47 are reference climatological stations (reference stations). The four wind speeds per day (UTC+8 2:00, UTC+8 8:00, UTC+8 14:00, and UTC+8 20:00) or the daily wind statistics (average and maximum) per day are generally available in China. Based on the characteristics of the available wind data, four scenarios regarding the selection of the type of wind data were used to model the regional potential wind erosion (Table 2). In Scenario 1, 130 meteorological observing stations (reference and basic stations) with daily average and maximum wind speed data were used. In Scenario 2, 130 meteorological observing stations (reference and basic stations) with four wind records per day were used. In Scenario 3, 47 reference climatological stations (reference stations) with daily average and maximum wind speed data were used. In Scenario 4, 47 reference climatological stations (reference stations) with four wind records per day were used. The “Standards for Classification and Gradation of Soil Erosion” were used to classify the potential wind erosion hazard (weak, slight, moderate, severe, very severe, and catastrophic) (MWRPRC, 2007).

The parameters testing, the base values, the input values, and the sensitivity values are presented in Table 3. Overall, wind speed is the most sensitive parameters of the two models. Generally, wind erosion intensity is remarkably affected by threshold wind speed. The wind speed data with high spatial and temporal resolution is thus important for regional wind erosion simulation. Besides wind speed, for RWEQ, snow cover and vegetation cover can effectively control wind erosion, and the climatic factors have apparent seasonality. Several soil properties, such as the clay content and soil moisture can also reduce the wind erosion. However, the least sensitivity was observed for the inputs of temperature, precipitation days and sunshine time. For IWEMS, the wind speed is also the most sensitive inputs for wind erosion modeling.

The average potential wind erosion values from 2000 to 2012 for the different scenarios are presented in Figure 2A. For the RWEQ, the highest average potential wind erosion values were 27.33, 22.12, 19.34, and 15.73 t hm⁻² a⁻¹ for scenarios 1, 2, 3, and 4, respectively. For the IWEMS, the average annual potential wind erosion values were 98.54, 84.73, 69.46, and 61.77 t hm⁻² a⁻¹ for scenarios 1, 2, 3, and 4, respectively. As is shown in Figure 2B, most of the severe wind erosion occurred in spring and the wind erosion was the lowest in summer. Table 4 presents the potential wind erosion with different erosion hazards from 2000 to 2012 for the different scenarios. For the RWEQ, the percentage of the area with weak and slight erosion hazards accounted about 80% of the total area. The percentages of the areas with weak erosion hazards were 27.52, 41.10, 39.43, and 47.82% for scenarios 1, 2, 3, and 4, respectively, and the percentages of the areas with slight erosion hazards were 50.75, 39.21, 45.36, and 38.97%, respectively. For the IWEMS, the areas with weak and slight erosion hazards were remarkably smaller than those for the RWEQ, and the weak and slight erosion hazard areas occupied the large proportion (∼50%) of the total area. In contrast, the percentages of the areas with severe, very severe, and catastrophic erosion hazards were high for all of the scenarios, i.e., 41.37, 32.82, 33.07, and 25.83% for scenarios 1, 2, 3, and 4, respectively. The areas with severe, very severe, and catastrophic erosion hazards decreased and the areas with weak and slight erosion hazards increased from scenario 1 to scenario 4. In summary, weak and slight erosion hazards were predominant in the APEC. Moreover, the potential wind erosion estimated using the IWEMS model is higher than that estimated using the RWEQ, and the four scenarios produce different results.

Figure 3 shows the spatial patterns of the potential wind erosion estimated using the RWEQ and IWEMS for the different scenarios. For the RWEQ, the area of very severe and catastrophic erosion hazards markedly decreased from scenarios 1 to 4 in the Horqin Sandy Land and the Hunshan Dake Sandy Land, and no catastrophic erosion hazards were scattered in the Hunshan Dake Sandy Land for scenario 4. Moderate erosion hazards dominated the Mu Us Sandy Land for all of the scenarios, and severe erosion hazards occurred in small regions for scenarios 1 and 2. For the IWEMS, the Horqin Sandy Land, the Hunshan Dake Sandy Land, and the Mu Us Sandy Land exhibited very severe and catastrophic erosion hazards for all of the scenarios. The very severe and catastrophic erosion hazards decreased remarkably from scenarios 1 to 4 in the Horqin Sandy Land and the Hunshan Dake Sandy Land. In summary, the geographic distributions of the potential wind erosion were similar, and the wind erosion hotspots were mainly located in the Horqin Sandy Land, the Hunshan Dake Sandy Land, and the Mu Us Sandy Land.

The results show that there were seasonal variations and annual fluctuations in the potential wind erosion. Figure 2B shows the average seasonal potential wind erosion in the APEC. Wind erosion occurred during all of the seasons, and the seasonal variation of the potential wind erosion was as follows for all of the scenarios: spring > winter > autumn > summer. The potential wind erosion exhibited obvious annual fluctuations from 2000 to 2012 (Figure 4). The trends of the inter-annual variations in the average potential wind erosion were similar between 2000 and 2012 for all of the scenarios. The average potential wind erosion was the highest in 2001 for all of the scenarios; and it was the lowest in 2011, except for scenarios 3 and 4 for the IWEMS, in which they occurred in 2005 and 2008, respectively. Overall, the potential wind erosion significantly decreased from 2000 to 2012.

Several methods often employed in soil wind erosion evaluation, particularly the common radioisotope ¹³⁷Cs monitoring method have high degree of precision for wind erosion monitoring in field surface experiments and observations in different model (Gharibreza et al., 2020). The verification of the models was investigated based on collected measured wind erosion data. Here, the seven sites and two regional areas of measured wind erosion were obtained using radioisotope ¹³⁷Cs method from the literatures (Tables 5, 6). The locations of the observation are presented in Figure 1C. We compared our results in all scenarios with previous findings by using radioisotope ¹³⁷Cs method. From Table 5, the simulated results were generally in agreement with the wind erosion sites reported in the literature in two models. For all the sites, the results of simulated wind erosion are usually over the measured wind erosion in IWEMS, except for Datong county and Dalad Banner in scenario 2 and scenario 4, respectively. However, RWEQ modeling results generally larger than previous measured values using ¹³⁷Cs in scenario 1 and scenario 2, meanwhile, simulated results less than the measured in scenario 3 and scenario 4. We also compared our results in regional range with previous study of radioisotope 137Cs in Kangbao and Fengning, Hebei of china (Table 6). On the whole, simulated results with IWEMS larger than RWEQ, particularly in Kangbao county. And maximum values of the simulated over the maximum from observed values.

Herein, the average potential wind erosion was found to be significantly lower for scenarios 1 to 4 for both models. For the same basic spatial meteorological data (scenarios 1 and 2 and scenarios 3 and 4), the average potential wind erosion estimated from the daily average and maximum wind speeds was higher than that calculated using four wind speeds per day (Figure 2). This is due to the fact that the methods of generating the hourly wind speed were different. WINDGEN was used to reproduce the hourly wind speed based on the daily average and maximum wind speeds, whereas the four wind speed measurements per day were interpolated using the linear interpolation method to obtain the hourly wind speed Wagner et al., 1992. Table 7 presents the average wind speed and the percentage of erosive wind (>5 m s⁻¹ wind speed at a height of 2 m). Clearly, the average wind speed and the percentage of erosive wind generated using WINDGEN are greater than those generated using the linear interpolation method. High temporal resolution wind speeds are generally required to generate a precise wind erosion model (van Donk et al., 2008; Guo et al., 2012; Panebianco and Buschiazzo, 2013). Since detailed wind data may be unavailable for some regions, several temporal wind data interpolation methods have been developed (Donatelli et al., 2009). It is suggested that the users need to access the prediction errors of the different temporal interpolation methods when modeling regional wind erosion.

For the same temporal meteorological data (scenarios 1 and 3 and scenarios 2 and 4), the average regional wind erosion obtained from the 130 meteorological stations was higher than that obtained from the 47 meteorological stations (Figure 2). Point meteorological data are generally interpolated into regional raster data for regional geographical and environmental modeling (Mauger et al., 2013; Li and Heap, 2014). Mathematically, the factors affecting the performances of spatial interpolation methods include the sampling design, the spatial distribution of the samples, the nature and quality of the data, the correlation between the primary and secondary variables, and the interactions among the factors (Shepard, 1968; Li and Heap, 2014). Here the kriging method was used to upscale the wind data. As the number of meteorological stations and the measured accuracy increase, the performance of the kriging interpolation tends to increase (MacEachren and Davidson, 1987; Luo et al., 2008; Martin et al., 2016). In addition, the locations and distribution of the meteorological stations within the study area also affect the accuracy of the kriging interpolation (MacEachren and Davidson, 1987; Keskin et al., 2015; Ozturk and Kilic, 2016).

Climate change is an important contributor to wind and water erosion. It can accelerate or decelerate the rates of wind and water erosion and can further alter the soil’s health, productivity, and surface cover (Gao et al., 2002; Sharratt et al., 2015; Webb et al., 2020). The influences of historical climate changes on wind erosion have been extensively studied. The variations in the historical wind speed, precipitation, and temperature have generally been summarized as the main factors affecting temporal wind erosion trends (e.g., Haiyan Zhang et al., 2018; Edwards et al., 2019; Lin et al., 2020). Recently, the effects of future climate change scenarios on wind erosion have also been explored. For example, Sharratt et al. (2015) reported that the soil and PM₁₀ losses via wind erosion will be 25–84% lower under the mid-21st century climate (2035–2064) on the Columbian Plateau, USA, compared with the baseline scenario (1970–1999). Li et al. (2020) reported that projected climate changes (2006–2099) will decrease the regional wind erosion by 10.71–33.74% in Central Asia. These studies generally used different types of wind data (e.g., daily or hourly data) and various wind erosion models (e.g., the RWEQ and WEPS). Mathematically, the magnitude of the wind erosion variations due to climate change may be similar to the effects of using different types of wind data (e.g., Munson et al., 2011; Zhang et al., 2019). The discrepancy in the wind erosion modeling caused by using different types of wind data or different wind erosion models may interfere with characterizing the effect of climate change on wind erosion. Therefore, determining a method of selecting a type of wind data in terms of the temporal resolution (daily or hourly data) and spatial network (basic or reference stations) is vital to quantifying the effects of historical or future climate change on regional wind erosion.

Moreover, many studies involving correlations with wind erosion, such as the impacts of landscape changes (land use/cover) on wind erosion (e.g., Chunlai Zhang et al., 2018; Chi et al., 2019), ecosystem services evaluation of reducing soil erosion (e.g., Hao et al., 2017; Zhao et al., 2017), soil CO₂ sequestration and emissions due to wind erosion (e.g., Webb et al., 2012; Chappell et al., 2019), and wind erosion-induced surface soil nutrient loss (e.g., nitrogen and phosphorus) (e.g., Du et al., 2019; Song et al., 2019), have generally neglected the effect of the type of wind data on precise wind erosion modeling.

Soil loss due to wind erosion at the field or regional scales can be simulated by wind erosion models. However, the models’ predictions are widely underestimated or overestimated (Pi et al., 2017). For instance, Buschiazzo and Zobeck (2008) found that the RWEQ and WEPS underestimated the soil mass transport by 45 and 40%, respectively, in a bare agriculture field. Several studies have demonstrated the underestimation of the maximum sediment transport (Qmax) and field soil loss (SL) and the overestimation of the critical field length (S) when using the RWEQ model at multiple sites (Zobeck et al., 2001; VanPelt et al., 2004).

To improve the accuracy of wind erosion models, calibration of the sensitive inputs of the models is necessary. At the field scale, measured wind erosion data can be relatively easily determined (Jarrah et al., 2020). Several studies have concluded that the prediction accuracy of wind erosion models (e.g., the RWEQ and WEPS) can be remarkably improved by calibrating some of the key parameters (Visser et al., 2005; Fryrear et al., 2008; Youssef et al., 2012; Xing et al., 2018). At the regional scale, the calibration of wind erosion models requires more standard long-term observation data (Jarrah et al., 2020). Du et al. (2018) calibrated the IWEMS model to enhance its performance using data from 452 passive sand traps from 2009 to 2011. Song et al. (2019) predicted the soil organic carbon and nutrient losses resulting from aeolian dust emissions after calibrating the IWEMS using data from 293 field experimental sites from 2014 to 2015. The RWEQ was upscaled to the regional scale and calibrated using measured wind erosion data, and then, it was further used to access the wind erosion risk in the Yellow River watershed, China (Du et al., 2015a; Du et al., 2015b). Several other studies have focused on validating wind erosion models. The soil loss through wind erosion determined using the ¹³⁷Cs method was used to validate the upscaled RWEQ (e.g., Chunlai Zhang et al., 2018; Chi et al., 2019; Wu et al., 2021). Published wind erosion data have also been used to verify wind erosion models (e.g., Wang et al., 2020; Zhou et al., 2020). Nevertheless, many studies have directly assessed regional wind erosion, but no calibration or verification of the wind erosion models were conducted (e.g., Feng and Sharratt, 2005; Guo et al., 2013; Borrelli et al., 2017; Fenta et al., 2020; Lin et al., 2020). Our study indicates that the type of wind data has a significant influence on the potential wind erosion estimation obtained using the RWEQ and IWEMS. Therefore, it is necessary to systematically calibrate and validate these models to achieve precise wind erosion modeling.