How desertification in northern China will change under a rapidly warming climate in the near future (2021–2050)

Arid, semi-arid, and semi-humid regions (drylands) with fragile ecological balance have undergone dramatic climate change in past decades, and how the desertification will change under a continuous warming background still remain uncertain. In this study, the bias-corrected Community Earth System Model outputs from Coupled Model Intercomparison Project phase 5 were dynamically downscaled using the Weather Research and Forecasting (WRF) model, based on which the evolution trend of desertification over northern China (NC) in the past (1972–2000) and the near future period (2021–2050) under the RCP8.5 scenario were analyzed using the dune mobility index, and the impacts of climate change on the intensification or reversal of desertification over NC in the near future were explored. The results show that WRF downscaling can reproduce the desertification changes over NC in the past. The regions with a high risk of desertification are mainly located on the border of the desert and gobi. Under a rapidly warming climate in the near future, desertification will likely reverse in most regions of NC, especially for regions north of 40°N over NC. Potential evapotranspiration changes will exacerbate desertification, while precipitation changes will promote rehabilitation, and wind speed changes show obvious local impacts on desertification. The results in this study imply that, with rising temperatures in the future, the extent of desertification will not always continue, desertification will likely reverse at the front and margin of deserts and gobi, and responses of desertification to climate change have significant spatial differences.


Introduction
As a worldwide environmental problem, desertification is land degradation caused by climate change and human activities, endangering human society and causing significant damage to the living environment and economic development (UNCCD 1994;Wang 2003). The diversity and complexity of desertification processes make it difficult to quantify and distinguish the relative contributions of climate change and human activities, but there is no doubt that climate change plays a critical role in the evolution of desertification (Wang and Hui 2005;Wang et al. 2006Wang et al. , 2009Mirzabaev et al. 2019;Zhang et al. 2020Zhang et al. , 2021. In China, desertification occurs primarily in Northern China (NC), where more than 90% of regions exhibit high aeolian activity and are thus classified as sandy desertification areas (Wang et al. 2009). This desertification will intensify or occur with the increase of dune activity or reactivity of anchored dunes, changes in aeolian sand transport, surface evaporation, and rainfall, which will significantly affect the vegetation cover, leading to changes in dune activity (D'Odorico et al. 2013).
Dune mobility indices can be used to monitor dune activities and reflect the occurrence and reversal of desertification using climate data, and they can also be combined with global climate models (GCMs) output to project future changes (Talbot 1984;Lancaster 1988;Knight et al. 2004;Wang et al. 2009). So far, the Coupled Model Intercomparison Project phase 5 (CMIP5) datasets provide the significant climate change information on a global scale. However, due to the uncertainties in both spatial resolution and accuracy of GCMs, the simulated results do not accurately represent local characteristics, and there are significant biases at the regional scale (Zou and Zhou 2013). To improve simulation accuracy, regional climate models, which have finer resolution and can better describe mesoscale processes and complex topography, were commonly used (Zhang et al. 2008;Hui et al. 2018;Zhaoye et al. 2022). Therefore, the high-resolution future projections of the desertification evolution over NC using a regional climate model for dynamical downscaling are imperative.
Desertification is a slow-changing process, which is a cumulating result under the background of continuously rising temperatures. Coincidentally, a comparatively high greenhouse gas emissions scenario-representative concentration pathway (RCP) 8.5 can provide a background of continuous rapid warming. The RCP8.5 scenario, under which significant climate change should occur, might result in a prominent desertification evolution in the future, along with outstanding economic and social impacts. Studies (Miao et al. 2015;Mirzabaev et al. 2019;Ma et al. 2021) have shown characteristics of desertification at the end of the twenty-first century in RCP8.5. However, how desertification will change in the near future (2021-2050) is still unclear, and understanding the characteristics of future desertification evolution is urgent.
This study aims to address the following issues: (1) How will the desertification change over NC with rapid rising in temperature in the near future? (2) What is the impact of climate change on the changes of desertification in NC in the near future? To address these, the spatiotemporal variations of desertification in the historical period  and the near future period (2021-2050) were analyzed, and the contributions of climate factors to desertification in near future were explored.
The rest of the paper is organized as follows: The data and methodology are introduced in Sect. 2. Section 3 validates the performance of WRF downscaling simulations. Section 4 analyzes the spatiotemporal variations of desertification over NC in the past. Section 5 investigates the evolution of desertification over NC in the near future. Section 6 explores the impacts of future climate change on desertification in NC. Conclusions and discussions are made in Sect. 7.

Dataset
Monthly global gridded near-surface air temperature and precipitation datasets based on high-resolution stations from the University of Delaware version 5.01 (UDel v5.01; Willmott and Matsuura 2001) were used to validate the performance of WRF downscaling in reproducing characteristics of climate over NC, which is on 0.5° × 0.5° spatial resolution and spans from 1972 to 2000. The UDel dataset is derived from several datasets, such as GHCN2 (Global Historical Climate Network), which is available at the National Oceanic and Atmospheric Administration (NOAA) Physical Sciences Laboratory (https:// www. esrl. noaa. gov/ psd/ data/ gridd ed/ data. UDel_ AirT_ Precip. html).
Based on the evaluation results of Ramon et al. (2019), the monthly near-surface wind data from the reanalysis of the fifth-generation European Center for Medium-Range Weather Forecasts (ERA5; Hersbach et al. 2019) was selected to validate WRF downscaling in simulating near-surface wind over NC. The dataset covers the period 1972-2000 and has a spatial resolution of 0.25° × 0.25°.
The historical outputs of the Community Earth System Model (CESM), which was released by the National Center for Atmospheric Research (NCAR) from CMIP5 (CMIP5-CESM; https:// www. wdc-clima te. de/ ui/) with a resolution of 0.94° × 1.25° and span from 1972 to 2000 as the GCM results to evaluate the improvements of downscaling in the simulation of annual mean precipitation, near-surface air temperature.
To facilitate validation of the WRF downscaling performance, near-surface air temperature and precipitation of the UDel v5.01, near-surface wind speed of ERA5, and downscaling results of WRF are interpolated to a 0.25° × 0.25°grid.
Observational desertification area data over NC is obtained from the National Forestry and Grassland Administration (http:// www. fores try. gov. cn/).

Initial and boundary conditions for WRF
In this study, the initial and boundary conditions were obtained from the bias-corrected CMIP5-CESM outputs at a spatial resolution of 0.94° × 1.25° and temporal resolution of 6 h, this dataset was revised by the European Centre for Medium-Range Weather Forecasting-Interim Reanalysis (ERAI) and has a lower error than its original outputs (Bruyère et al. 2015), it has been widely used as the initial and boundary fields for regional climate models. The bias-corrected CESM outputs are produced by combining a 25-year (1981-2005) mean annual cycle from ERAI and a 6-h perturbation term from CESM, and this bias correction method corrects the mean state while retaining synopticscale and climate-scale variability as simulated by CESM (Bruyère et al. 2015). This dataset is available from NCAR's CISL Research Data Archive (http:// rda. ucar. edu/ datas ets/ ds316.0).
The historical outputs and future projections in RCP8.5 from the bias-corrected CESM dataset were used for driving WRF downscaling. RCP8.5 is the so-called "baseline" scenario and does not include any specific climate mitigation target, greenhouse gas emissions and concentrations increase significantly over time in this scenario, leading to a radiative forcing of 8.5 W/m 2 at the end of the twenty-first century (Zou and Zhou 2013). RCP8.5 scenario was selected to construct a rapidly warming climate background.

Experimental design
The WRF model was used to perform downscaling simulations over NC in this study. The WRF model is a fully compressible and non-hydrostatic model widely used in weather forecasting and climate simulation (Skamarock et al. 2019). The bias-corrected CESM dataset was used as the driven data (initial and boundary conditions) for WRF, the historical simulation was integrated for 30 years from 1971 to 2000, and the RCP8.5 scenario simulation was integrated for 50 years from 2006 to 2055; the integrational step is 180 s. The first year was the spin-up time. One domain in a Lambert conformal conic projection was used for dynamic downscaling, with a horizontal resolution of 30 km (Fig. 1). The simulation domain is centered at a point (45°N, 98°E), with 201 × 157 horizontal grid points in the zonal and meridional directions, respectively. Vertical atmospheric coordinates are divided into 27 layers, and the top of the atmosphere is 50 hPa.

Dune mobility index
To analyze the dune activity and desertification over NC, the dune mobility index (M) proposed by Lancaster (1988) was chosen, which has been widely used in desertification studies (e.g., Thomas and Leason 2005;Wang et al. 2009). Depending on the analysis of previous studies (Zhong 1998;Wang et al. 2004Wang et al. , 2009), we used 1000 as the threshold value for determining whether the dunes are active. When the index M < 1000, it means that the dunes are completely fixed by vegetation and there is no risk of desertification; when the index M > 1000, it means that vegetation cannot completely restrict the dunes, and the dunes become active, meanwhile, the larger value of the index M represents the more intense activity of the dune. When M > 1000, the dunes start to become active, and the surrounding areas have the risk of desertification intensifying. Therefore, areas with M > 1000 are defined as potential deserts, including actual deserts.
The dune mobility index (M) is expressed as where P represents precipitation (mm/day), E p represents potential evapotranspiration (mm/day), and W is U 3 , where U represents the monthly average 10-m wind speed (m/s). We use this formula to calculate monthly M values, and annual M is the sum of monthly M values. In addition, potential evapotranspiration is calculated using the modified Penman formula, as followed by Shuttleworth (1993): where R n represents daily net radiation (mm/day), ∆ represents the slope of the saturation vapor pressure curve (Pa/K), γ represents the psychrometric constant (Pa/K), D represents the vapor pressure deficit (Pa), u 2 represents the daily average wind speed at 2-m height (m/s), and λ represents the latent heat of vaporization of water (2.45 × 10 6 J/kg).
The dune mobility activity varies with climate change. On interannual and interdecadal scales, dune erodibility and erosion forces are influenced by precipitation (P), temperature (via potential evapotranspiration, E p ), and the strength and frequency of sand-carrying winds (W) (Bullard et al. 1997;Thomas and Leason 2005). Because desertification is a relatively slow process of land degradation, long-term changes in dune activity imply trends in desertification. In this study, the dune activity level averaged from 1972 to 2000 is considered the normative dune activity capacity, and if the dune activity level in a period is higher than the normative dune activity capacity, intensification of desertification is indicated; conversely, if the dune activity level in a period is lower than the normative dune activity capacity, a reversal of desertification is indicated.
As dune mobility index M is a function of E p , P, and W, their contributions to M can be approximated as follows: where dE p , dP , and dW are mean changes over a given period for E p , P , and W , respectively. The terms M∕ E p , M∕ P, and M∕ W show the sensitivity of individual factors to changes in M, where each variable is the mean climatologic value. The terms ( M∕ E p )dE p ,( M∕ P)dP , and ( M∕ W)dW represent the contributions to changes in M, i.e., the contributions to desertification.

Validation of WRF downscaling over NC
The performance of WRF downscaling is validated by comparing WRF simulations with the CMIP5-CESM outputs, the UDel v5.01, and the ERA5 dataset. As the most basic and important climate elements influencing desertification, precipitation, near-surface air temperature, and wind can be used to assess the performance of WRF in simulating desertification. Since CMIP5-CESM does not provide near-surface wind output, the near-surface wind of WRF downscaling will be verified with the ERA5 reanalysis. The bias, pattern correlation coefficient (PCC), and root mean square error (RMSE) were used to quantify the downscaling performance.
The biases in the CMIP5-CESM and WRF models for simulating the annual mean air temperature and precipitation averaged from 1972 to 2000 compared to the UDel v5.01 dataset are shown in Fig. 2. Except for high temperatures in the Tarim Basin and low temperatures in the Tibetan Plateau (TP) and northern part of Northeast China, the annual air temperature pattern is characterized by a temperature gradient with cold in the northwest and warm in the southeast (Fig. 2a). The outputs of the CMIP5-CESM are comparable to the UDel v5.01 dataset in the northeastern region of NC, but it underestimates temperatures in the rest of regions of NC (Fig. 2b). Compared with the CMIP5-CESM outputs, WRF downscaling reduces the underestimation of air temperature in the southwestern region of NC and TP, the PCC increases from 0.67 to 0.87, and the RMSE decreases by 0.26 °C ( Fig. 2b and c). The observed annual precipitation from UDel v5.01 shows a decrease from the southeast to the northwest over NC (Fig. 2d). The CMIP5-CESM outputs show a clear overestimation of precipitation over NC, especially in the south-central region of NC, whereas the simulations of the WRF reduce such bias and are comparable to the UDel v5.01 dataset in the northwestern region of NC ( Fig. 2e and f), correspondingly, the PCC increases from 0.47 to 0.51 and the RMSE decreases by 0.16 mm/day. Comparison of near-surface wind speeds ( Fig. 2g and h) indicates that, although the WRF simulations overestimate wind speeds in the southern and northeastern regions of NC, the biases at the north of 40°N where desertification mainly occurs are relatively small (< 1 m/s). Overall, the WRF model corrects the simulation bias of the CMIP5-CESM for low-temperature areas and shows a clear improvement in precipitation in most regions of NC. In addition, the dynamical downscaling results describe the finer-scale spatial characterization of air temperature and precipitation.
To further evaluate the downscaling performance of the WRF model downscaling, we compared the biases in the simulated trends of annual mean air temperature, precipitation, and near-surface wind between the WRF downscaling and the CMIP5-CESM. As shown in Fig. 3, the annual mean air temperature over almost the whole NC shows a significant rising trend from the UDel v5.01 dataset, except for the western margin and the central region of NC. The CMIP5-CESM underestimates the warming trend in the northeastern region of NC but overestimates that in the western region of NC and the eastern region of TP ( Fig. 3a and b). The WRF downscaling simulations appear to be improved in simulating the warming trend in the northeastern region of NC (Fig. 3c). For the annual precipitation, except in the central region of NC and western TP, where the UDel v5.01 dataset shows a decreasing trend, significant increasing trends of annual precipitation appear in most regions PCC and RMSE were calculated between the UDel v5.01 dataset (or ERA5 dataset) and the CESM, WRF results of NC (Fig. 3d). The CMIP5-CESM exhibits obvious biases in simulating the trend of annual precipitation over NC, which underestimates the decreasing trend in western TP and increasing trend in the northwestern and southern regions of NC and overestimates the increasing trend in the eastern region of NC (Fig. 3e). Compared with the CMIP5-CESM outputs, the WRF downscaling shows a distinct improvement in simulating the trend of annual mean precipitation, reducing the trend biases in the western and central NC, with the PCC increased by 0.07 and the RMSE decreased by 0.002 (mm/day)/ yr (Fig. 3f). Near-surface winds from the ERA5 show a significantly decreasing trend in central Inner Mongolia and western region of NC. Although the WRF downscaling simulations show an overestimation of wind trends at TP and underestimation over the northeast region of NC, with PCC and RMSE is 0.18 and 0.005 ( Fig. 3g and h), the spatial pattern of near-surface wind trend is generally reproduced by WRF (PCC = 0.18) over NC.
The above analysis shows that the WRF downscaling simulations can reproduce the patterns of climatology and trends of air temperature, precipitation, and near-surface wind over NC in the past, and these climate variables play key roles in desertification, which means that the WRF model has potential for simulating desertification over NC.
Thus, the next analysis will be based on the WRF downscaling simulations.

Spatiotemporal variations of desertification over NC during 1972-2000
In NC, except for the barren land (i.e., desert and gobi; gray shading in Fig. 4), e.g., Junggar Basin, Tarim Basin, Qaidam Basin, and Alashan Plateau, where the extremely arid climate pattern always keeps an intensification or reversal of desertification hardly occurs, large dune mobility index (M) values appear in the border of desert and gobi, western TP and the northeastern regions of NC, indicating that the dunes are easily mobile in these regions where the vegetation cover fraction is low. The distribution patterns of M are very similar in the past 3 decades (Fig. 4a-c), which means the pattern of desertification risk over NC has been almost unchanged, these regions are also sensitive to the climate change and the border of these regions should own high desertification risk.
In addition, the spatial pattern of M is consistent with the distribution of vegetation cover and deserts over NC (Piao et al. 2003;Zhang and Huisingh 2018;Zhang et al. 2020).  1972-1980, 1981-1990, and 1991-2000. d-f Corresponding changes in M compared to the histori-cal  mean. The gray shading represents the barren land. Slanted lines represent the value significant at the p < 0.1 level by Student's t-test To analyze desertification evolution, the spatial distributions of M changes from 1972 to 2000 in each decade are given (Fig. 4d-f). The desertification during the period 1972-1980 shows a reversal in northern Xinjiang (around Gurbantunggut Desert), whereas the desertification in central Inner Mongolia (east side of Badain Jaran Desert) is intensified and expands eastward, the desertification trend in northeast China (periphery of Hulunbuir Desert, Horqin Desert, and Nengjiang Desert) was not obvious (Fig. 4d). The evolutionary trend of desertification during the period 1981-1990 is generally opposite to that in the last decade, and the regions, where desertification was originally intensified, show reversal trends (Fig. 4e), specifically, the desertification in central Inner Mongolia is reversed, while the desertification in northern Xinjiang is slightly intensified. During the period 1991-2000, desertification is reversed from central Inner Mongolia to northern Shanxi (around Mu Us Desert), while desertification still shows an intensified trend in northern Xinjiang (Fig. 4f). Notably, regions with significant changes in desertification are generally located at the edges of deserts and gobi, which are ecologically fragile and have low annual rainfall, and are sensitive to climate change.
To validate the desertification trends in terms of M, the desertification changes between 1991-2000 and 1982-1990 were compared with the changes in NDVI during the same period. As shown in Fig. 5, the reversal of desertification in the eastern regions of NC (such as the margin of Mu Us Desert and the south side of Otindag Desert) is consistent with the increases in NDVI, whereas the intensification of desertification in the middle Inner Mongolia is accompanied by the unsignificant changes in NDVI. In northern Xinjiang, desertification shows an intensifying trend, while the NDVI shows an increasing trend, this might be due to that the dune mobility index M represents the typical mobility of dunes, which is not only related to the cover fraction of the vegetation but also influenced by the climate change (e.g., high wind). The above results further validate that the WRF downscaling combined with dune mobility index M can reproduce the general characteristics of desertification in past decades.

Evolution of desertification over NC in the near future
The recent AR6 report (IPCC 2021) has pointed out that temperature rising will likely persist over the continent (e.g., Asia), whereas changes in precipitation will be uneven in the future, implying that the changes in desertification over NC will have significant spatial differences under the impacts of climate change. How desertification over NC changes in the near future should be explored. As shown in Fig. 6, compared with the spatial pattern of desertification over NC from 1972 to 2000 (Fig. 6a), the overall spatial distribution of M over NC in the future 30 years (2021-2050) does not change significantly, but the state of local dunes changes (Fig. 6b), the mobile dunes in the eastern part of Inner Mongolia (around the Otindag Desert) will be anchored. The changes in M indicate a trend of desertification, and the difference between the historical  multi-year average M and the future (2021-2050) multi-year average M is Generally, desertification in most of the regions north of 40°N over NC will be reversed from 2021 to 2050, whereas the desertification in the southern region of NC will be intensified. These results imply that desertification in regions with relatively mild climatic conditions is more likely to occur or intensify, whereas desertification in regions with relatively poor climatic conditions will be reversed.
To investigate the decadal evolutions of desertification under continuing warming background in the near future, Fig. 7 shows spatial distributions and changes in M from 2021 to 2050 in the high-emission scenario (RCP8.5). Although the spatial patterns of desertification will not change in the next 3 decades, the sand dune status will change in the eastern Inner Mongolia Plateau (Fig. 7a-c). The change in the dune state is undoubtedly greater than in the historical period, suggesting that more intense climate change in the high-mission scenario leads to greater changes in dune motility. Similarly, the greater change in dune motility also leads to a greater magnitude of desertification evolutionary trends. During the period 2021-2030, desertification will intensify in the range from middle Gansu to northern Shanxi (margin of Tengger Desert and Mu Us Desert) and the east side of Nengjiang Desert, desertification will reverse over the mid-eastern Inner Mongolian and northern Xinjiang (Fig. 7d). Desertification changes over NC during the period 2031-2040 have characteristics similar to those during the period 2021-2030, and the trend of desertification reversal is more obvious, especially over the range from the edge of Horqin Desert to the edge of Nengjiang Desert (Fig. 7e), while desertification will intensify over the range from Gansu to Ningxia. During the period 2041-2050, desertification changes over NC have characteristics similar to those in the last decade: reversal of desertification will occur over northern Xinjiang and mid-eastern Inner Mongolia, especially over the margin of Otindag Desert, while intensification of desertification will weaken (Fig. 7f). Totally, under a rapid warming background (RCP8.5 scenario), climate change shows clear local characteristics, leading to reduced desertification from Inner Mongolia to northeast China, while desertification will intensify in the north of Gansu (around Tengger Desert).
To analyze the overall change in desertification over NC, following previous studies (e.g., Wang et al. 2009), potential deserts that represent dunes are not fully immobilized (i.e., areas at risk of desertification) are defined as areas with a sand dune activity index M greater than 1000 in this study. Generally, areas with a large index M value include the potential deserts, deserts, and gobi, while it is difficult to precisely separate the potential deserts and deserts traditionally using the dune mobility index M. Nevertheless, due to the areas where sand dunes are fully active almost unchanged in decades scale, the temporal evolution of an area with M greater than 1000 actually reflects the temporal evolution of the potential desert. Thus, the areas with M greater than 1000 as a percentage of NC areas are calculated to show the desertification changes over NC in the following.
As shown in Fig. 8, the potential desert area percentage shows an increasing trend during 1972-1990 and then shows a decreasing trend during 1991-2000, with the turning point located in the 1990s. Such a temporal trend is consistent with the temporal evolution of desert areas in previous studies (e.g., Wang et al. 2008;Miao et al. 2015), and it is also consistent with the temporal evolution of NDVI over NC (Piao et al. 2003). The monitored desertification area by the National Forestry and Grassland Administration (NFGA) decreased from 2000 to 2015. Due to the period of historical simulation until 2000, it is hard to directly compare the simulation with the monitoring of the NFGA. Nevertheless, comparing the simulated potential desert area percentage between 2000 (about 70%) and 2010 (about 65%), the simulated potential desert (desertification) area decreased since 2000, which is consistent with the trend of monitored desertification areas. In the near future (2021-2050), the potential desert area over NC shows a decreasing trend. In addition, the interannual and interdecadal variability of the potential desert area in the RCP8.5 scenario is larger than that during the historical period , implying that desertification change will be more unstable under the rapid warming background. The dune mobility index M is composed of precipitation, potential evaporation, and wind speed, so the evolutionary trend of desertification is modulated by the variabilities of these climatic variables. To explore the possible impacts of climate change in the future on desertification over NC, the contributions of precipitation, potential evaporation, and wind speed changes to desertification are calculated based on Eq. 3 (details seen in Sect. 2.4). As shown in Fig. 9, the spatial pattern of the dune mobility index M changes (i.e., the tendency of desertification) induced by the changes in potential evaporation (dE p ) (i.e., ( M∕ E p )dE p ) indicates a positive contribution of dE p to the desertification in almost all regions (Fig. 9a), which is because the sensitivity of M to E p change (i.e., ( M∕ E p )) is positive over NC (Fig. 9b), followed by an increase in  . 9 Attributions of desertification over NC to climate change in near future . a Contribution of potential evaporation (E p , mm/day) to M change, b sensitivity of M to E p , and c changes (Δ) in E p in near future compared to the historical mean . d-f and g-i are same as a-c but for precipitation (P, mm/day) and wind speed (W, m/s). The gray shading represents the barren land E p (positive ΔE p ) in most regions of NC (Fig. 9c). Among them, the sensitivity of M to E p changes in most regions of NC have the same magnitude, whereas in the western part of NC, the changes in E p (ΔE p ) are larger than that in the eastern part, which makes the contribution of E p to the desertification in the western region of NC (e.g., northern Xinjiang) is relatively larger than that in the eastern region. The contribution of P changes to desertification (i.e., ( M∕ P)dP ) over NC is opposite to that of E p and is negative in most of the regions (Fig. 9d), which is inseparable from the negative sensitivity of M to P change (i.e., the term M∕ P ) over NC (Fig. 9e) and an increase in P (ΔP) in most of the regions over NC (Fig. 9f), the negative contribution of P mainly appears over the northern Xinjiang (i.e., the margin of Gurbantunggut Desert), the periphery of Horqin Desert and Nengjiang Desert, which is linked to the large changes of P (positive ΔP) over these regions. The above results imply that, along with rapid warming (RCP8.5 scenario), the precipitation over NC will probably increase in the future, which is the main cause of desertification reversal over NC. Because wind speed is susceptible to the heterogeneity of surface conditions, the spatial distribution of the contributions of W to M changes (i.e., ( M∕ W)dW ) exhibit obvious local characteristics, presenting positive contributions over the range from middle Inner Mongolia to Shanxi and northeast corner of Inner Mongolia, and negative contributions in the northern Xinjiang and east side of Badain Jaran Desert (Fig. 9g). The sensitivity of M to W change (i.e., term M∕ W ) has small spatial differences (Fig. 9h). The spatial pattern of W change (ΔW) is similar to the spatial pattern of the contribution of W to M change (Fig. 9i), which means the contributions of W to M changes are mainly caused by the W changes.
From the above analysis, for desertification changes over NC in the near future, changes in potential evaporation will benefit the intensification of desertification, which is related to the rapid temperature rising in the highemission scenario; changes in precipitation will contribute to the reversal of desertification over NC, as precipitation increases in most areas in the future; changes in wind speed have obvious local impacts on the desertification. Meanwhile, the terms M∕ E p , M∕ P , and M∕ W have large values in the surrounding areas of desert and gobi, this means that these areas are climate-sensitive and vulnerable to the impacts of climate change. In other words, the borders of arid, semi-arid, or semi-humid are more susceptible to climate change, resulting in significant changes in desertification intensification or reversal of desertification.

Conclusions and discussions
It is widely thought from the subconscious that with the temperature rising, the desert extent will continue to expand and desertification will intensify. In this study, the evolution of desertification over NC was analyzed in the past 29 years , and projected changes in desertification in the near future (2021-2050) were based on the WRF model downscaling simulations. The possible impacts of climate changes (i.e., evaporation, precipitation, and near-surface wind changes) on desertification were explored. The results show that WRF downscaling provides more reliable regional features of climate compared to the CMIP5-CESM. WRF downscaling reduces the simulation bias of local temperature over NC; the WRF model significantly improves the simulation of precipitation in most regions of NC, especially in regions where the climate is significantly influenced by the underlying topography and surface inhomogeneity, such as inner deserts and basins. WRF downscaling also can capture the spatiotemporal variations characteristics of near-surface wind over NC.
The dune mobility index M calculated by WRF downscaling can reproduce the general characteristics of desertification in past decades. The simulated desertification in the historical period  is close to the observation. The evolution trend of desertification is consistent with the changes in vegetation cover (NDVI).
In the near future, even with rapidly rising temperatures, the trend of desertification will exhibit a regional response. Under the RCP8.5 scenario, desertification will likely reverse in most regions of NC during the period from 2021 to 2050, especially for regions north of 40°N over NC, such as central and eastern Inner Mongolia, northern Xinjiang, whereas decertification in some regions will be likely intensified such as middle-northern Gansu. Notably, the sand dunes in the Mu Us Desert, Nengjiang Desert, and Hulunbuir Desert are unstable; decertification in both the past and future are vulnerable to climate change, and the greater degree of climate change will lead to a greater change in dune activity. As speculated, the magnitude of desertification change is greater and the potential desert area change is more unstable in the high-emission scenario.
Climate change greatly impacts the decertification of NC in the near future. Potential evapotranspiration exacerbates desertification, precipitation enhances the reversal of desertification, and wind speed has a distinctly localized effect on desertification. In addition, the surrounding arid regions, such as deserts and gobi, are extremely vulnerable to climate change, leading to significant changes in desertification intensification or desertification reversal. This study focuses on wind erosion desertification over NC. Changes in the dune mobility index M, which is related to precipitation, evaporation, and wind speed, reflect possible trends in such desertification under the impact of climate change. However, the dune mobility index M exhibits significant variability in climate-sensitive areas such as deserts and sandy regions. As mitigation policies are developed, future desertification trends will need to be thoroughly investigated in future studies using additional possible scenarios, such as different social development pathways in CMIP6. In addition, the dune mobility index M only represents the impact of climate change; nevertheless, human activities also play an important role in desertification, so determining the direct role of human activities in this regard remains a challenge.