Spatiotemporal variability of the potential wind erosion risk in Southern Africa between 2005 and 2019

Regional assessments of the wind erosion risk are rare and vary due to the methods used and the available data to be included. The adaptation of existing methods has the advantage that the results can be compared directly. We adopted an already successfully applied methodology (ILSWE—applied in East Africa), to investigate the spatiotemporal variability of the wind erosion risk between 2005 and 2019 in Southern Africa. The approach integrates climatic variables, a vegetation index, and soil properties to describe the potential impact of wind erosion at the landscape scale. The annual and seasonal variability is determined by the vegetation cover, whereas droughts and strong El Niño events had only regional effects. We estimated that 8.3% of the study area experiences a moderate to elevated wind erosion risk over the 15‐year period with annual and inter‐annual fluctuations showing a slight upward trend. In general, the desert and drylands in the west have the highest proportion of risk areas, the moist forests in the east are characterized by a very low risk of wind erosion, while the grasslands, shrublands, and croplands in the interior most likely react to changes of climatic conditions. The validation process is based on a comparison with the estimated frequency of dust storms derived from the aerosol optical depth and angstrom exponent and revealed an overall accuracy of 65%. The results of this study identify regions and yearly periods prone to wind erosion to prioritize for further analysis and conservation policies for mitigation and adaptation strategies.


| INTRODUCTION
In the 21st century, soil erosion is globally one of the most serious natural hazards, which challenges the environment, agricultural productivity, and thus food security Montgomery, 2007). In arid and semi-arid regions wind erosion accounts for 46% of total land degradation (Zheng, 2009). Wind erosion is also a threat to Southern Africa favored by dry climates and seasonal droughts, affecting natural and anthropogenic-influenced sensitive biomes (Huang et al., 2016;Zhao et al., 2021). Against a backdrop of increasing population in the area and regional food production already reaching its natural limits, wind erosion-induced declines in soil fertility will put further pressure on crop production and ecosystem services and drive up economic costs (Department of Environmental Affairs, 2014; Kirui & Mirzabaev, 2014;Kotze & Rose, 2015). Additional anthropogenic drivers such as deforestation, overgrazing, monocultures, large fields, and generally poor soil management can accelerate the process of wind erosion (UNEP, 2015;Webb et al., 2020). The vulnerability of soils to erosion by wind is affected by certain primary soil characteristics like grain size distribution, lime and carbon content, affecting secondary soil characteristics such as aggregate stability, soil surface crusting and roughness, or moisture levels (Funk et al., 2008;Shao, 2008;Wiesmeier et al., 2012). Additional dynamic factors are vegetation cover (VC) and climatic variability (Shao, 2008).
Studies have shown that emitted dust contains the finest and most valuable parts of soil, such as the silt and clay fraction and soil organic carbon (Funk et al., 2008;Ravi et al., 2011;Sterk et al., 1996).
Soil organic carbon content of eroded dust can spike up to 17-times from its original source (Nerger et al., 2017). With wind erosion occurring in regions with a low net primary production, the loss of soil organic carbon can be considered an irrecoverable loss to soil quality (Chappell et al., 2016;Yan et al., 2005). Global dust emissions are estimated at 2000 Mt, of which 1500 Mt are deposited on land and 500 Mt in the oceans (Shao et al., 2011). While positive off-site effects are ocean and land fertilization, the loss of nutrients and organic matter increases the vulnerability of the source areas. If erosion rates exceed the rate of soil formation, it can push ecosystems to a state of decay (D'Odorico et al., 2013;Obalum et al., 2012;Zabel et al., 2014). The future prospect of climate change puts further pressure on regions prone to wind erosion, leading to a decrease in VC and biomass production (Munson et al., 2011;Ravi et al., 2010).
In the 1930s, Dust Bowl events in the Great Plains region of the United States triggered the development of empirical wind erosion models (Tatarko et al., 2013). Continuous efforts led to the wind erosion equation (WEQ), the first empirical model, based on wind tunnel experiments and field measurements, integrating different factors, for assessing annual soil loss (Woodruff & Siddoway, 1965). The revised WEQ (RWEQ) replaced WEQ as a tool to estimate erosion (Fryrear et al., 2001) and has been extensively tested and found to be in good agreement with on-field measurements (Buschiazzo & Zobeck, 2008).
One of the more recent wind erosion models is a pan-European assessment, utilizing the most influential parameters that affect the potential of wind erosion on landscapes (Borrelli et al., 2016). This approach, the index of land susceptibility to wind erosion (ILSWE), has been adopted and modified by Fenta et al. (2020) for the East African region, with a high resolution of 100 m cell size and a validated accuracy of around 70%.
Research related to soil degradation in Southern Africa has been mainly focused on water erosion at the plot or watershed scale (Le Roux et al., 2008, 2007Weinzierl et al., 2016). While there are studies on dust emissions in the South African Free State, information on a regional scale is rather lacking (Eckardt et al., 2020;Wiggs & Holmes, 2011). In general, focusing on one area alone does not provide a comprehensive view of the process of wind erosion in the Southern African region. This situation calls for a qualitative approach to map out areas most susceptible to wind erosion to narrow the knowledge gap and generate a basis for further studies.
In this study, the aim is to outline the general land susceptibility to wind erosion in the region of Southern Africa with the approved modeling approach ILSWE by integration of high-resolution remote sensing data and spatial analysis tools in a geographic information system (GIS). The specific goals are as follows: 1. To identify and highlight risk areas.
2. To analyze the annual spatial patterns, over a period of 15 years, to estimate a prospective trend.
3. To relate results to major types of land cover and predominant terrestrial ecoregions in the global warming context.
The results can give an impact on future erosion control management policies and land use scenarios. It will also provide environmental researchers with a basis for small-scale studies on the effects of wind erosion in the Southern African region.

| STUDY AREA
The investigated area covers the southern part of the African continent and includes seven countries: Namibia, Botswana, Zimbabwe, Mozambique, Eswatini, South Africa, and Lesotho comprising 3.9 million km 2 (Figure 1). About 11% or 433,619 km 2 of the area is under agricultural use. Southern Africa, as a terrestrial ecoregion, is dominated by savannas occupying 45% of its area, followed by desert and xeric shrublands with 35% coverage, respectively. While the western dry regions are poor in vegetation, the Cape and southern coastal areas have rich flora, ranging from small shrubs to Mediterranean woodlands. Moist broadleaf forests and mangroves characterize the humid east ( Figure 1) (Burgess et al., 2004). The western regions are dominated by sandy soils, while the eastern regions are characterized by loamy soils (Figure 2).
The mean annual precipitation ranges between 5 mm on the west coast up to 1950 mm in the eastern regions with high inter-annual variability (Fauchereau et al., 2003) (Figure 3). The mean annual temperature is around 17 C, ranging from more than 40 C in the desert regions of Namibia and Botswana to À20 C in the high-altitude regions around Lesotho and Zimbabwe (Hijmans et al., 2005).
Observed trends in annual average near-surface temperature in Southern Africa, based on CRUTEM4v data, show a significant positive anomaly in comparison to the long-term average of 1961-1990, with the 2000s being the warmest decade (Engelbrecht et al., 2015;. This trend is more prevalent in minimum temperature, resulting in a limited diurnal temperature range (New et al., 2006). The El Niño-Southern Oscillation also affects Southern Africa and is generally associated with lower-than-normal rainfall, while strong events coincide with high drought severity (Philippon et al., 2012).
The region-wide average wind speed is 3.5-6.5 m s À1 reaching its highest in the mountainous country of Lesotho and the southern coastal area of South Africa and its lowest at the coast of Mozambique ( Figure 3). The general wind direction is east, while westerly winds dominate on the southern coast in the summer months (Rienecker et al., 2011).

| Index of land susceptibility to wind erosion
The complexity of the wind erosion processes demands integration of a variety of physical parameters, which impede an adequate upscaling of plot-sized models (Funk & Reuter, 2006). Plot-sized models can have a demand for extensive data to meet requirements, which poses difficulties even on a local scale. It is necessary to reduce the parameters to the essentials, while adhering to well-established conditions for wind erosion, for a regional scale assessment (Böhner & Köthe, 2003). Therefore, macroscale-sized models are limited by the availability of spatially and temporally uniform datasets and call for modification and incorporation of varying data sources, measurement intervals, and periodicity. The approach in this study, the ILSWE (Borrelli et al., 2016) and its adaptation by Fenta et al. (2020) use static and dynamic erosion parameters. It inherits the basic three rules that wind erosion occurs when wind speed exceeds a threshold, the soil surface is susceptible and there is little to no vegetation (Tsoar, 1994). The factors used in the GIS analysis are divided into five categories to characterize the factorial and combined sensitivity to wind erosion: climatic erosivity (CE), wind-erodible fraction (EF), soil crust (SC), VC, and surface roughness (SR). To create a uniform relationship, the contributing factors were divided into a value range from 0 to 1 using the fuzzy logic approach. Fuzzy logic was introduced by Zadeh (1965) as a more realistic depiction of a range of metrics, in contrast to classic Boolean logic, distinguishing between false and true (McBratney & Odeh, 1997). The analysis was carried out, using ArcGIS (ArcMap 10.8.1, ESRI Inc., CA, USA), describing the established direct relationships between each individual erosivity factor to the rate of F I G U R E 1 (a) Main biomes of Southern Africa comprising countries of the study area. Localized by numbers: (1) Namibia, (2) Botswana, (3) Zimbabwe, (4) Mozambique, (5) Eswatini, (6) South Africa, and (7) Lesotho (modified after Olson et al., 2001) and (b) land cover classification of the study area (ESA CCI).

Inverse logarithmic for SR.
Thus, a sensitivity assessment layer was produced for every factor separately to compute a summary map by multiplying each factor after Fenta et al. (2020): In the final step, the result was classified into five categories from very low to very high sensitivity by applying the Jenks-Caspall-algorithm. This classification method seeks to reduce the variance within classes and maximizes the variance between classes (Jenks & Caspall, 1971). The ILSWE was calculated for the 15-year period of 2005 to 2019, as well as each individual year. Furthermore, a seasonal index was created to highlight the most vulnerable periods throughout the year. Finally, an aerosol optical depth-based validation method was applied. The methodological structure is presented in Figure 4.

| Data sources
The datasets used for the GIS analysis and its validation are derived from the following sources (Table 1): 1. Wind speed, precipitation, and potential evapotranspiration are derived from TerraClimate, a monthly temporal climatic dataset

| Climatic erosivity
The CE factor determines the average rate at which surface soil particles move relative to moisture content and average wind speed (Skidmore, 1986). We used a proposed model from Chepil et al.
(1962) as a revised version by FAO (1979). It is best suited for arid regions, as the CE factor depends only on the wind speed when the rainfall amount approaches zero, while it is zero when the rainfall is equal to the potential evapotranspiration (Skidmore, 1986): Where: u i is the average monthly wind speed (m s À1 ) at 2 m height, PET i is the monthly potential evapotranspiration (mm), P i is the monthly precipitation (mm), and d i represents the total number of days in month i.

| Wind-erodible fraction
The EF describes the susceptibility of a bare soil surface to windgenerated shear force due to its soil properties (Colazo & Buschiazzo, 2010). To determine the erodibility of soils, Chepil (1941) sieved samples into erodible and non-erodible fractions. Subsequent wind tunnel experiments showed that particles larger than 0.84 mm were stable under test conditions and considered non-erodible. The EF of soil is characterized by the portion of erodible aggregates of the top 25 mm of topsoil, with organic carbon and clay contents representing the main factors toward increasing aggregate stability by binding single particles (Fryrear, 1980;Skidmore & Layton, 1992). The calculation of the EF factor includes basic physical and chemical soil properties: EF ¼ 29:09 þ 0:31 Sa þ 0:17 Si þ 0:33 Sa Cl À 2:59 OM À 0:95 CaCO 3 100 ð3Þ Where: Sa is the sand content with a cutoff range from 5.5% to 93.6%, Si is the silt content with a cutoff range from 0.5% to 69.5%, Cl is the clay content with a cutoff range from 1.2% to 53%, OM is the organic matter content with a cutoff range from 0.18% to 4.79%, and CaCO 3 as the calcium carbonate content with an upper cutoff at 25.2% (Fryrear et al., 1994). The equation has not been verified for values outside of these ranges, thus the cutoffs were used if the datasets exceeded the validation limits (Fryrear et al., 1998).

| Soil crust
As most studies agree on the reductive effect of SCs on wind erosion (Belnap, 2003;Eldridge & Leys, 2003;Zhang et al., 2006), the SC factor is included in ILSWE, based on the regressed abrasion coefficient on clay and organic matter content (Fryrear et al., 1998): Where: Cl is the clay content with a cutoff range from 5% to 39.3% and OM is the organic matter content with a cutoff range from 0.32% to 4.74%. The limits of Equation (4) derive from wind tunnel tests on the resistance of soil aggregates and crusts to windblown sand (Hagen et al., 1992).

| Vegetation cover
The VC factor, as the ratio of vertically projected vegetation to the full extent of the area to be examined, can be expressed with the use of

| Surface roughness
Surface roughness refers to natural and man-made vertical irregularities of the earth's surface that decrease wind speed close to the ground in an inverse logarithmic relation (Sozzi et al., 1998). Therefore, SR elements affect near-surface aerodynamics and thus contribute to the reduction of aeolian transport. Since SR maps are not commonly available, it is possible to use land cover classes for an estimation. We calculated the SR factor, by comparing annual 300 m gridded land cover maps with suggested roughness length values from Hansen (1993) (Table S11) and other sources (Floors et al., 2018;Markert et al., 2019).

| Validation by recognition of dust storms
Since long-term ground measurements of wind erosion data are not available, we used the SeaWiFS product's global daily AOD and AE to identify persistent dust sources derived from an estimate of the fre-  (Table S1).
As expected, elevated risk areas are only located in the desert and the Mediterranean shrubland biomes, in the case of the latter, caused by the high wind speeds near the coast. While there are areas in the savanna that are affected by wind erosion, the majority of its extent has a very low susceptibility and risk areas are situated around the border zones to more arid regions. The modelling results show no significant wind erosion risk in the montane grasslands and flooded grassland areas, with mangroves and lakes exclusively representing the low-risk classes, hence the latter was excluded from the supporting tables (Table S2).
In comparison with land cover classes, the modeling outcome displays that the 15-year mean annual wind erosion risk of cropland is very low to low. The reasoning is the long-term averaging time, which averages out times after harvest or fallow periods. Similarly, the shrubland and grassland classification display only a very low to low F I G U R E 5 Susceptibility to wind erosion in Southern Africa as 15-year average. [Colour figure can be viewed at wileyonlinelibrary.com] susceptibility to wind erosion, with grassland indicating a slightly elevated medium risk. On the other hand, the bare and sparse Vegetation classes represent around 95% (9.1 million ha) of the high to very highrisk areas, covering about 7.8% of the investigated region (Table S12).
In context, only 0.3% of shrubland is affected by high susceptibility to wind erosion, representing an area of 0.4 million hectares (Table S6).  (Table S2) (Table S6).

| Seasonal variability of the wind erosion risk
The inter-annual modeling of land susceptibility to wind erosion has been divided into three seasons, ranging from January to April ( (Tables S7-S9).

| DISCUSSION
Our approach follows the validated model ILSWE that has been modified by Fenta et al. (2020) for application in East Africa, a region climatically similar to our study area. We adopted the validation methodology and compared the spatial and temporal derived wind erosion risk with the frequency of dust storms observed by satellites in that area. We got an overall accuracy of 65% for our study area (Table 2, Figure 8). The very low (80%) class has the highest validation accuracy, followed by the very high (40%) and low (38%) risk classification. Natural and anthropogenic fires occurred during the period of In general, studies on wind erosion in Southern Africa are scarce.
The increase of studies on erosion and dust generation in the region in recent years shows that the problem is being recognized and getting scientific attention. The processes of wind erosion vary considerably in their spatial and temporal distribution so large-scale studies can only give an approximate picture of the processes (Funk et al., 2004;Siegmund et al., 2018). While the spatial resolution can still be chosen relatively high even for larger areas in a GIS, the data availability sets limitations for the temporal resolution. Our wind ero- While the influence of the climatic factor is mainly apparent on the southern coast, a region with less tropical influence and therefore characterized by greater climatic variance (Grieser et al., 2006) Our results for the seasonal wind erosion risk assessment are more differentiated than our annual outcomes (Table S1). The general trend shows a slight downward trend of medium to elevated risk classes, with Sea1 being more in line with the annual trend, Sea2 and with a lesser degree Sea3 representing the general trend, while having a larger extent of elevated risk classes respectively. The seasonal modeling outcome moderately represents the rainy, dry, and greening seasons, while the VC has a stronger influence on the results than the CE factor ( Figures S6-S8 and S10-S12). Figure 9 shows the linear relationship of the CE factor toward the ILSWE results, while the VC factor has an overall higher impact with three distinct point clouds, representing the seasonal modeling outcome. Consequently, our results show that the period with the highest risk of wind erosion for agriculture is from January to August and the lowest protective effect of grassland is apparent from May to August. With rather low R 2 values for the annual and seasonal risk class trends, apart from the dry season between May and August, a future trend prediction on the wind erosion risk is associated with a high level of uncertainty (Table S10). A reason could be the variability and limitations of the fractional vegetation index. While climate data is readily available on a monthly basis, it is challenging to generate consistent, continuous vegetation data. In regions in and around the tropics clouding will produce noise and faulty surface reflectance (Hmimina et al., 2013). For our study region, the use of a 120-day composite vegetation index was necessary to provide at least one value per pixel, restricting the results to a seasonal index. Additionally, the fractional vegetation represents healthy leaf green, without fully considering the protective effect of dried and leafless vegetation (Chappell et al., 2018). This particularly affects the savanna landscapes with a natural dry cycle (Chidumayo, 2001 Past and recent studies show evidence, that droughts have become more intense and widespread in the region of Southern Africa (Fauchereau et al., 2003;Masih et al., 2014). Precipitation, or rather its absence, is the main driver of drought occurrence, however the inclusion of heat waves in drought projection is getting increasing attention, which will become more relevant in the context of climate change (Abiodun et al., 2012;Vicente-Serrano et al., 2011). Temperature is only indirectly represented in this study by potential evapotranspiration but has an impact on wind erosion too. Higher temperatures and low rainfall can result in limited vegetation growth, dry out soils and inhibit aggregation of particles, which increases the vulnerability of susceptible soils to wind erosion (Sharratt et al., 2013).
There is evidence of a drastic increase in surface temperature on the African continent, on the order of twice the global temperature increase (Engelbrecht et al., 2015;Jones et al., 2012). Combined with an observed decrease in rainfall and a lengthening of the dry season since the late 60s, future drought scenarios anticipate significant pressures on the primarily rain-fed agricultural systems (Fauchereau et al., 2003;MacKellar et al., 2014;Niang et al., 2014). Our results show the highest share of medium and elevated risk classes in the years of 2007, 2013, 2015, and 2019, years when droughts and heatwaves afflicted large parts of Southern Africa (Nnopuechi, 2021).
Apart from 2013, these were years with moderate to strong El Niño events, negatively affecting rainfall over the western part of Southern Africa and furthering high temperatures over long periods (Blamey et al., 2018;Pomposi et al., 2018). In 2017, the subdivided outcome for the Mediterranean shrubland zone shows the highest share of risk classes, the year with the most devastating drought in the last 100 years around the Cape of Africa (Pascale et al., 2020). Thegrasslands and cropland classification, being the most vulnerable to climatic influences, show the largest extent of medium to elevated risk areas in drought years (Table S6). These results are consistent with reported crop failures and the frequency of high fire danger days, resulting in bushfires in grasslands (Andela & van der Werf, 2014;Guimarães Nobre et al., 2019;Verschuur et al., 2021).
Given the overall trend of warming and decreasing average annual rainfall, wind erosion could become a more serious threat in the future. With limited resources and increasing population growth, pressures on agricultural systems may intensify, necessitating further conservation practices like the implementation of shelterbelts and agroforestry systems (Sheppard et al., 2020).

| CONCLUSIONS
The joint assessment of wind erosion for natural and anthropogenicinfluenced areas makes it difficult to evaluate the effects on ecosystems. The very high risk in deserts associated with large amounts of erosion has to be evaluated differently than the often much lower dust emissions emitted from cropland, but leading to irreversible soil degradation. Agricultural areas are generally not threatened to a great degree in our results, but the extent of moderate and elevated wind erosion risk substantially rose in years of droughts. In addition, the temporal resolution of our modeling results is not sufficient to fully evaluate the wind erosion risk of highly dynamic agricultural systems and therefore demands further investigation. The dynamics of the annual and seasonal wind erosion risk were mainly influenced by the VC, with the CE factor having a regional impact. Under the consideration of a general warming trend, with a decrease in annual rainfall in the context of a changing climate, wind erosion could become a more problematic phenomenon in the future, demanding conservation policies for mitigation and adaptation. While our results identify hotspots of wind erosion, effective land management practices for controlling wind erosion are likely to vary locally and may include shelterbelts, revegetation, and similar sustainable measures to increase soil cover, biological diversity, and minimum soil disturbance. Field-based studies are therefore required for high-resolution mapping of target areas to develop individual adaptation strategies.

CONFLICT OF INTEREST
The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are openly available.