Soil wind erosion in ecological olive trees in the Tabernas desert (S.E. Spain): a wind tunnel experiment

Wind erosion is a key component of the soil degrada tion processes. The purposes of this study were to find out the influence of material loss fro m wind on soil properties for different soil types and changes in soil properties in olive groves when the y are tilled. The study area is on the property of an olive oil company and is located in the north of th e Tabernas Desert, in Almeria Province, Spain. The 15 climate is semiarid thermomediterranean and it is o ne f the driest areas in Europe. We used a new win d tunnel model over three different soil types (Calci sol, Cambisol and Luvisol) and concentrated on micr oplot losses and deposits detected by a laser scanne r integrated in the tunnel. We also studied the ima ge processing possibilities for examining the particle s attached to collector plates located at the end o f the tunnel to determine their characteristics, and whet her they were applicable to the setup. Samples coll ected 20 in the traps at the end of the tunnel were analyzed . We paid special attention to the influence of org anic carbon, carbonate and clay contents because of thei r special impact on soil crusting and the wind-erod ible fraction. A Principal Components Analysis (PCA) was done to estimate any relationships on generated dust properties and the intensity and behavior of t h se relationships. Component C1 separated data wit h high N and OC contents from samples high in Fine si lt, CO3 and Available K2O content. C2 separated 25 data with high Coarse silt and Clay contents from d ata with high Fine sand. C3 was an indicator of Available P2O5 content . Then analysis of variance (ANOVAs) were done to ana lyze the effect of soil type and sampling height on different properties of capt ed dust. Calculations based on tunnel data showed overestimation of erosion in tilled Cambisol compar ed to the other tilled soil typologies and calculat ion of the fraction of soil erodible by wind done by ot her authors for Spanish soils. As the highest loss wa 30 found in Cambisols, mainly due to the effect on soi l crusting and the wind-erodible fraction aggregati on

carbon, carbonate and clay contents because of their special impact on soil crusting and the wind-erodible fraction. A Principal Components Analysis (PCA) was done to estimate any relationships on generated dust properties and the intensity and behavior of those relationships. Component C1 separated data with high N and OC contents from samples high in Fine silt, CO3 = and Available K2O content. C2 separated 25 data with high Coarse silt and Clay contents from data with high Fine sand. C3 was an indicator of Available P2O5 content. Then analysis of variance (ANOVAs) were done to analyze the effect of soil type and sampling height on different properties of capted dust. Calculations based on tunnel data showed overestimation of erosion in tilled Cambisol compared to the other tilled soil typologies and calculation of the fraction of soil erodible by wind done by other authors for Spanish soils. As the highest loss was 30 found in Cambisols, mainly due to the effect on soil crusting and the wind-erodible fraction aggregation of CaCO3, a Stevia cover crop was planted between the rows in this soil type and this favored retention of particles in vegetation.

Introduction
Soil is the key component of the Earth System as it controls the hydrological, erosional, biological and geochemical cycles, and also contributes with services, goods and resources to the humankind (Keesstra et al., 2012;Brevik et al., 2015;Smith et al., 2015). Soil degradation is related to soil compaction, loss of 10 vegetation and organic matter and increase of soil erosion, either by water or wind (Novara et al., 2011;Prosdocimi et al., 2016;Arjmand Sajjadi & Mahmoodabadi, 2016). To research those processes of land degradation will contribute to restore and rehabilitate them and to understand the soil genesis and related process on soil degradation and formation. Wind erosion is a world wide environmental concern (Houyou et al., 2014;Martínez-Graña et al., 2015) but some regions of the world are more affected due to their 15 climatic conditions (Cerdâ et al., 2010;Borrelli et al., 2016). In semiarid regions, where the distribution and intensity of precipitation are irregular, wind moves enormous amounts of soil, with the consequential ecological imbalance. Several authors (Liu et al., 2003;López et al., 2000;Li et al., 2004;Gao et al., 2015) have studied the relationships of wind erosion, wind speed, soil typology and vegetation, which affect the quality of soil by modifying the organic carbon content. 20 Leys et al. (2002) evaluated the wind erosion range based on the effect of dry aggregation levels and percentage of clay. Zobeck et al. (2013) observed that dry mechanical and aggregate stability diminished and erodible material increased as soil organic matter content decreased. Beniston et al. (2015) discuss P losses driven by the transport of the mineral fraction, under different tillage management. Kaiser et al. (2014) mention the stocks of the stable C and N pools were not affected by the tillage intensity but were 25 positively correlated with the stocks of the clay-size fraction, indicating a strong influence of site-specific mineral characteristics on the size of these pools. Benlhabib et al. (2014) analyzed dryland Mediterranean cultivation systems, discussing and recommending sustainable cultivation technologies which showed a significantly positive effect on crop productivity, yield stability and environmental sustainability. Hevia et al. (2007) found that no-till showed more large aggregates and fewer fine aggregates than traditional tillage. This was also indicated by Gao et al. (2015) and Wang et al. (2015) over soil-conserving tillage in Northern China. Gomesa et al. (2003) 5 observed that soil erodibility by wind under traditional tillage was lower than in conservation tillage, since only a limited amount of material was available to wind erosion due mainly to crusting of the soil surface. Colazo & Buschiazzo (2010 confirmed that cultivation increased the erodible fraction of soil (EF) and reduced dry aggregate stability (DAS) in medium-textured soils by causing weakening of the soil structure due to loss of organic carbon (OC) and breakup of aggregates. In fine-textured soils, the 10 formation of large resistant aggregates by tilling causes EF and DAS to be more alike than under no-till conditions. According to Rawlins et al. (2015), soils with more stable aggregates have larger disaggregation reduction values. These authors mention soil quality, measured by critical soil physical properties, may decline if the organic carbon concentration is less than a critical threshold. Hagen et al. (2010) observed that tilling ridges are effective for trapping aggregates transported by saltation, but do 15 not usually reduce erosion rates in soil where aggregates transported in suspension predominate. Feras et al. (2008) demonstrated in a wind tunnel study that sediment traps efficiency depended mainly on particle size and wind speed. Traps placed at different heights can measure vertical sediment flow (Basaran et al., 2011).
Expansion and intensification of olive tree cultivation in Andalusia, especially in the late 18 th century, 20 accelerated erosion as a result. The introduction of the use of cover crops in the region due to application of the standards derived from the EU Common Agricultural Policy, involves the need for additional management investment (Gómez et al., 2014). In fact, cultivation of Stevia down the center between rows of olive trees is under study in our experimental area.
Vegetation can diminish soil loss from wind, because it reduces the wind speed and soil erodibility, 25 and traps more eroded material (Touré et al., 2011;Leenders et al., 2011;Lozano el al., 2013;Asensio et al., 2015). Udo & Takewaka (2007), in their wind tunnel experiments, arrived at the conclusion that in addition to density, the height and flexibility of vegetation are essential in determining the effectiveness in lowering mass transport by wind. Youssef et al. (2012) suggest that the pattern of vegetation in rows parallel to the predominant wind direction lowers total mass transport.
Our objectives here were to: (1) study the influence of material loss on soil properties, (2) compare the differences in soil loss due to the soil type, and (3) observe the changes in generated dust properties when olive groves are tilled. other ecological crops, such as Stevia, almonds and grapes. We concentrated on four-year-old picual olive trees drip irrigated. Soils are mainly Calcisols (CL), Cambisols (CM) and Luvisols (LV). Texture is silty clay loam to loamy with 37 to 48% gravel fragments and a weak, coarse subangular blocky to strong, medium angular blocky structure.

Data acquisition and experimental design
Soil parameters of the different soil types are analyzed, before making applications of artificial wind in the tunnel, in order to get starting data. To analyze the soil volume lost from wind erosion and its effect on the surface microtopography, we tested both crusted and recently tilled soil. The crusted soils were strongly protected from wind erosion, while right after tilling, soils were highly susceptible to it. After tilling, soils tended to recover the physical surface crust within 10 to 12 days, reacquiring extra protection 5 from the wind.
The simulations were done in May 2013 in three plots for each soil type with olive trees, with 7 x 5 m tree spacing, which were only tilled once a year and where the aisles between the rows were very close to the predominant wind direction (there was about a 5º offset in Calcisols). The slopes and lengths of fields of the three experimental plots were 2% and 143 m on CL, 0% and 95 m on CM, and 0% and 152 m 10 on LV.
Our reference for weather records was the Tabernas Meteorological Station (located about 2 km away from the study area), one of the network of automatic stations belonging to the Institute of Agrarian,

Wind tunnel
To monitor wind intensity, as well as direction and shear, we worked with a wind tunnel with laminar and 20 turbulent flow similar to real wind conditions, in which the material transported was collected in traps for study. Our tunnel has two parts, as shown in Figure  Then comes the tunnel itself (Part 2) which consists of three compartments (0.8x0.8x0.8 m each one) into a telescopic structure: -The first compartment has a metal sheet completely covering the ground to keep the wind from affecting this area.
-The second compartment is the study area itself (0.8x0.8 m), where wind erosion is actually quantified. This area is equipped with a PCE-424 hot wire anemometer with 0.1 m·s -1 resolution, with which wind speed is monitored. There is also a NextEngine Desktop 3D laser scanner, which 5 is used to find the volume of eroded soil and alterations in the microrelief of the soil. The scanner has an indispensable lifting system which acts as a support structure and enables it to be set at the desired height.
-A liquid latex (Latepren ® Rx-505) coating was applied to the soil surface in the third compartment to fix particles so they would not move around during blowing and mix in with the particles in the 10 study zone. This way the natural roughness of the ground was maintained. The latex coating is spread with a bulb shape from outside the third tunnel compartment to avoid return of external particles due to edge turbulence.
At the end of the tunnel, the particles become attached to some vertical adhesive plates placed at different heights (0, 15, 40 and 70 cm), and are later analyzed using an industrial machine vision camera. 15 There are also particle traps (Fryrear BSNE, adapted for a fixed wind direction) located at the same height as the adhesive plates (Asensio et al., 2015), retain the dust that is later analyzed to quantify the loss. 0 cm one was partially buried, to locate the window at surface level.
The duration of the wind tunnel experiment was ten minutes by using a wind speed of 7.6 m s -1 at 70 cm height, following criteria of Fister & Ries (2009). The wind speed value corresponds to the maximum 20 daily average wind speed (last 15 years) registered at 2 m height by Tabernas Station. Flow simulation software, according to 15 cm tunnel rounding radius, shows wind profile homogeneity from 70 cm height to the central part of ground study area, at cross section.
In each case, the ground was scanned twice, before and after simulation with the wind tunnel. Scans were done at a height of 44 cm using a NextEngine 3D laser scanner under conditions of natural dryness. 25 This scanner has shown its applicability in acquiring microreliefs of agricultural soils (Aguilar et al., 2009) in high-precision field work (High Definition mode and MACRO) considering a sample size large enough to represent the plot in great detail. This provides a 120 cm 2 scan area with a 400 ppi capture density and nominal precision of 0.127 mm. Based on two point clouds found for each plot (before and after wind simulation), two digital terrain models with a 0.1 x 0.1 cm resolution were generated. The eroded soil volume was estimated as the difference in volume between DTMs. Once the soil volume was known, we estimated the amount of soil lost using the bulk density of each soil.
After making applications of artificial wind in the tunnel, we analyze the soil volume lost from wind 5 erosion and its effect on the surface microtopography, both crusted and recently tilled soil. The results of the scans done in the wind tunnel only take the loss model (no deposit model) into account. Deposits would have to be considered along with the loss model for the erosion balance to be more moderate.
To find out how wind erosion modifies surface microtopography, the point cloud from each scan was used to calculate random roughness (RR) in each sample, before and after simulation. RR is defined as 10 the standard deviation from the points within the plot after eliminating the slope effect. But in natural areas with a complex topography or on hillsides with high variability (changes in both flat curvature and profile), the elimination of slope does not eliminate the effects of changes in height caused by roughness factors, such as mounds, curvature or higher-order variations in surface, so the RR index tends to overestimate surface roughness in experimental plots. Therefore, the local RR index (RRL) estimation 15 method was applied (Rodríguez-Caballero et al., 2012) using Equation (1) Where Nw is the number of points in window w, Zw is the height of each point after eliminating the slope 20 effect and μw is the mean height in window w.
The adhesive plates were analyzed using a machine vision camera (JAI-CM080). This monochrome progressive scan camera with a 1024 x 768 pixel resolution is connected to a computer. During image processing with the "ImageJ" program, the number of particles present in each image can be counted, and 25 the mean size of particles, presence of aggregates or a color histogram of the image can be found.
The possibilities for image processing for detailed examination of particles adhering to the plates were studied. In a first approach, to determine the characteristics and applicability to the setup, a series of images of reference samples was taken to demonstrate system capabilities for: -Colorimetry studies: Quantitative study of the color of adhered particles, changing the color model and using the H component in the HSV model. 5 -Measurement on image of adhered particles -Particle count -Roughness analysis of isolated particles

Sampling and analytical determinations 10
Soil samples were taken from the upper 3 cm. We therefore concentrated on recently tilled soils, from which three repetitions of each soil type were evaluated. When using aggregate stability to assess soil erodibility, samples are usually collected from the plough layer, while soil erosion occurs at the soil surface. Hence, the potential changes in erodibility caused by crusting are ignored (Algayer et al., 2014). 15 Using the sub-crust would have led to greatly over-estimated erodibility.
Ground and collected samples were dried, crushed, and passed through a 2-mm sieve to eliminate large fragments. For both ground and trap samples, particle size distribution was assessed by dry sieving and the Robinson pipette method after eliminating organic matter with H2O2 (30%) and dispersion by agitation with sodium hexametaphosphate (10%). The sand fraction was separated by wet sieving, dried in an oven 20 and later fractionated by dry sieving. The organic carbon content (OC) was determined using the Walkley-Black wet digestion method. Total N (N) was calculated from NH3 volumetry after Kjeldahl digestion.
Available soil phosphate (P2O5) was calculated by photocolorimetry. Available soil potassium (K2O) was calculated by flame photometry. To determine bulk density (BD), 100-cm 3 cylinders were used to refer to sample dry weight by cylinder volume. 25

Statistical analysis
Data on soil characteristics acquired were examined for any changes or differences. A Principal Components Analysis was done to estimate any relationships on generated dust properties and the intensity and behavior of those relationships. Then an analysis of variance (ANOVA) was done to analyze 5 the effect of soil type on OC and CO3 = contents and another ANOVA was done to analyze the effect of height on clay content. The variables under study did not require transformation to satisfy the requirements of residual normality or variance homogeneity, so when the ANOVAs showed the effect analyzed to be significant, pairwise comparisons were assessed using the least significant difference test.
The level of significance was 0.05 in all tests. All statistical analyses were carried out with SPSS 20. 10

Results
At the study area, mean soil characteristics before artificial wind recorded for Calcisols, Cambisols and Luvisols (CL0, CM0 and LV0) are shown in Table 1. Three replicates were taken of each soil type. 15 Surface stoniness of these soils is high and the average gravel for the different typologies is 37% in CL, 48% in CM and 43% in LV.
In this study, we were concentrating on losses and deposits localized in a micro plot, which is what the laser scanner can detect. The results for the three soil typologies in soil blown with the same wind intensity generated artificially in the tunnel are shown in Table 2. 20 As an example of the results found by scanning, the digital terrain models and erosion maps for a sample from the Cambisol plot are shown in Figure 2. The variations in random roughness are conditioned by the balance of material lost and deposited.
Some of the results of the image processing plate study are shown below (Figure 3). The results, still being validated, can give us an idea that Calcisols have fewer granulometric fractions susceptible to wind 25 erosion. The three typologies have a clear contrast in color conditioned by their clay content and the presence of iron oxides, which makes them darker. In Calcisols, the aggregating effect of CaCO3 may be seen even with the naked eye. Figure 3 shows large-sized particles, plant residue and, in the color analysis, two groups of materials with different colorimetry.
The samples collected in the traps at the end of the tunnel were also analyzed. A Principal Components Analysis (PCA) of the results was performed using the R correlation matrix. The variables included in the PCA were Fine sand, Very Fine sand, Coarse silt, Fine silt, Clay, OC, N, Available P2O5, Available 5 K2O and CO3 = . First, eigenvalues were calculated and used to decide how many components should be extracted to adequately explain the data. Only the first three roots were worth exploring further, because the rest of them were less than 1. The first three roots account for 81.091% of the variance.
The factors extracted were rotated by Varimax to simplify the interpretation of the Pearson's correlation coefficients of the principal component scores and the raw data. Figure 4 (a) shows the 10 correlation coefficients on the plane of Components 1 and 2. From this figure, it is obvious that the % Fine silt, % CO3 = and Available K2O variables are clearly negatively associated with Component C1 while % N and % OC are associated positively with C1. Thus C1 separates data with high N and OC contents from samples high in Fine silt, CO3 = and Available K2O content. Component C2 is negatively associated with Fine sand and positively associated with % Coarse silt and % Clay. C2 separates data with high 15 Coarse silt and Clay contents from data with high Fine sand. Figure 4 (b) shows the correlation coefficients on the plane of Components 1 and 3, where it may be observed that Component 3 is positively associated with Available P2O5. Therefore C3 is an indicator of Available P2O5 content. Figure 5 show how samples cluster around the three main components. Figure 5 (a) shows that C1 is 20 strongly related to the soil group, that is, C1 separates the three groups of soil, and so the contents of the variables associated with C1 differentiate groups of soil. Thus from Figure 5  Therefore, the lower heights are associated with higher content in Fine sand while higher heights are associated with more content in Coarse silt and Clay. Figure 5 (b) shows that C3 separates soil Groups CL and LV from CM, suggesting that Available P2O5 content is the major difference.
Special attention should be paid PCA results for OC, CO3 = and Clay contents because of their special impact on soil crusting and the wind-erodible fraction. Variables OC and CO3 = were associated with Component C1, which is strongly related to soil type. So we tested the effects of soil type on OC and 5 CO3 = contents with an analysis of variance (ANOVA). Furthermore, the Clay variable was associated with Component C2, which was related to the height, and therefore, we tested the effect of height on Clay content. We did not need to apply transformations to satisfy the requirements of residual normality and variance homogeneity.
The ANOVA table for collected dust samples from different Soil Types (Table 3) applied to % OC 10 and % CO3 = shows significant differences for soil provenance in two variables. When its effect was found to be significant in the ANOVAs, and as none was considered the reference soil, we performed pairwise comparisons of the three soils. These were assessed using the least significant difference test (LSD, P≤0.05). Table 4 show that samples with the highest OC content were from CM, but with no significant difference from LV. From CL had significantly lower OC than the other two soil types where this variable 15 was similar. CO3 = contents were significantly different in the three soil types and CL had the highest % CO3 = while LV had the lowest.
The ANOVA table for height (Table 3) applied to % Clay shows significant differences too, but this time we did not do pairwise comparisons of all the four heights, but analyzed for a significant difference in heights of 0-15 cm and 40-70 cm. From Table 5 As t is 7.0912 (t=4.0725/0.5743) with a p-value of P=4.811713e-08, the difference between the heights 25 of 0-15 and 40-70 is highly significant.
We were also interested in the comparison between heights of 0 and 15 cm and between 40 and 70 cm.
As seen in Table 5, the difference between 0 and 15 cm is clearly not significant, whereas the difference between 40 and 70 is significant, and the highest content of Clay was at a height of 70 cm.

Discussion 5
Soil erosion is currently a major cause of land degradation and relevant at the European and global level.
Therefore more studies are needed to improve current models and integrate plot and regional studies to assist decision making (Panagos et al., 2012;Panagos et al., 2014;Borrelli et al., 2015).
As suggested by Lozano et al. (2013) and Asensio et al. (2015), bulk density is doubly influenced, on 10 one hand it tends to be reduced by the effect of organic enrichment, but on the other, increased by the accumulation of fine materials. This has great influence in physical soil crusting. Organic matter often combines with fine soil particles and Zhao et al. (2009) found a correlation coefficient between clay and organic matter content of 0.95.

The availability of phosphates to plants is influenced by factors such as neighboring plants and 15
seasonal dynamics in this semiarid environment. Zhang et al. (2014) made reference to damages by windblown sand and Zhao et al. (2007) showed that the effect of small deposits of sand on soil properties and performance of vegetation were not significant, although the soil temperature tended to rise with increasing thickness of deposits, which could affect the decomposition of organic matter.
According to PCA, Component 1 separates data with high OC contents from high CO3 = content and 20 Component 3 is an indicator of Available P2O5 content, being able to establish differences on soil Groups. C1 separates the three groups of soil, and samples belonging to the LV group show high content in OC.
C2 is related to the sampling height and show how higher heights are associated with more content in Clay. ANOVA shows significant differences for CM and LV in front of CL in OC content. We also find significant differences in Clay content for heights of 40-70 cm. 25 It is well known that velocity threshold increases as the particle size is increases. We did not take into account aggregate size for dust capted samples because these were crushed and passed through a 2-mm sieve for chemical analysing. The soil fraction erodible by wind (EF) is a key parameter for estimating soil susceptibility to wind erosion. Fryrear et al. (1994) proposed a multiple regression equation for calculating EF which considers the organic matter, sand, silt, clay and calcium carbonate contents as predictive variables. In fact, it has been included in prediction models such as the current Revised Wind Erosion Equation, RWEQ. Calculation of the EF in Spanish soils is problematic due to their high content in CaCO3 (López et al., 2007), so the equation proposed is: EF = 4.77 + 7.43 sand/clay + 27.6/organic 5 matter. Average EF calculated this way for our soils shows a slightly higher result in Cambisols (31%, compared to 27% in Calcisols and Luvisols).
Making a comparison for crusted and tilled soil types on the average soil loss, the tilling loss increase from CL more than six times in CM and more than fourteen times in LV. But this is taking into account only a loss model, without the deposition one. 10 Image analysis is a useful tool enabling submillimetric particles to be counted, and to analyze their size, shape and color. This could lead to creation of a database of soils with objectively measurable visual characteristics in the mid-term.
As the highest loss was found in Cambisols, a Stevia cover crop has been planted between the rows in this soil type (Figure 1). Stevia cultivation can offer up to three crops a year, which is done manually for 15 leaf that after drying is crushed.This crop favored retention of particles in vegetation, on the property subject of study.

20
Tilled soils in olive groves show a direct relationship between the differences in OC and clay content after blowing. Cambisols are more eroded than Calcisols and Luvisols, mainly due to the effect on soil crusting and the wind-erodible fraction aggregation of CaCO3 in CL and Clay in LV.
The wind tunnel led to overestimation of differences in soil type loss compared to other EF evaluation methods. We could suggest that higher-precision data have been found with this new wind tunnel than 25 found with other any tunnel designed to date, due to high resolution of the devices used, such as laser scanner and particle imaging.
Where wind erosion is higher, it is recommended that cover crops be planted between the rows of olive trees. Table 1. Initial soil characteristics for Calcisols (CL0), Cambisols (CM0) and Luvisols (LV0) tested.