**6. Land degradation geospatial model**

dust to Central Europe. Dust from the Sahara desert is transferred with windblast into the Central Europe several times per year. The solid particles concentration in the dust plume in

from the "Borna" station in Germany, the concentration of solid particles from the Ukrainian

Up to 70 tons of soils per hectare and per hour is blown away during the dust storms. The storm of March, 2007 was not an extraordinary phenomenon for the south of Ukraine. It is known that in the early 1950s, the countermeasures were taken to prevent the wind erosion spread. The wind protective forest strips were planted on more than 440,000 hectares in all the

Until recently, the soil dust natural particles were traditionally considered as harmless for the human health. But new research proved chronic bronchitis happened quite frequently in farmers [17] or increase of respiratory diseases in population residing in the territories of Karakalpak (Uzbekistan) [18], that is. in regions with the similar conditions described for the

As it is obvious, land degradation phenomena is of high priority to be researched and different techniques are to be considered for that. The solution to this problem requires not only a detailed study of the land degradation causes, but also involves identifying the risk of degradation. Satellite imagery has its advantages and widely used for investigation of

**5. Satellite imagery for the purpose of land degradation mapping**

. As it is obvious, this value exceeded the African plume parameters

. According to the monitoring data

the south of Germany is made up on the average 280 μg/m3

**Figure 2.** Dust storm in the region of the Kakhovske water reservoir, EOS/MODIS, 23.03.2007.

dust reached 640 μg/m3

58 Land Degradation and Desertification - a Global Crisis

natural zones of Ukraine [16].

by two times.

south of Ukraine.

The land degradation mapping technique was used on the basis of processing of a two-level model for multispectral satellite imagery of low and medium spatial resolution. First-level model applies several different thematic classifications of source multispectral images, for example vegetation change, soil erosion, etc. Second level gives data fusion of specific thematic classifications of the first level into final thematic map to improve accuracy and reliability owing for information support systems to provide land management.

Vegetation changes are usually detected on the multispectral satellite images by unified wellknown methods [20]. It should be noted that a modified soil-adjusted vegetation index MSAVI *Fv* is more preferable over generic normalized vegetation index NDVI for vegetation mapping in terms of steppe soil erosion in the southern Ukraine. The MSAVI index can be calculated by special equation:

$$F\_{\nu} = \rho\_{in} + \frac{1 - \sqrt{(2\rho\_{nr} + 1)^2 - 8(\rho\_{nr} - \rho\_{nd})}}{2} \tag{1}$$

where *ρnir* and *ρred* are land surface spectral reflectances in the near infrared and red spectral bands, respectively [21].

Water erosion depends on soil type and mineral composition, rainfall, slope steepness and vegetation density. The value of water erosion *zs* (mm/month) can be calculated from the regression relationship of the form [22]

$$z\_s = k\_s Q^2 (\text{tg}\,\alpha)^{167} \exp(-0.07V) \tag{2}$$

where *ks* is a soil erosion factor, *ks* ≅ 0.13 (mm/month)−1 for clay and sandy soils of the study area [23], *Q* is runoff (mm/month), α is a slope angle of the terrain, *V* is the vegetation cover fraction. Runoff is determined by the ratio of precipitation *P* (mm/month) and water retention in soil *R* (mm/month):

$$\underline{Q} = \frac{\left(P - 0.2R\right)^2}{P + 0.8R} \tag{3}$$

where *R* depends on the table hydrological soil index *Cs* as [24]

$$R = 25.4 \left( \frac{1000}{C\_S} - 10 \right) \tag{4}$$

The percentage of vegetation coverage *V* is generally considered to be proportional to scaled NDVI value square within the study area [25] and is easily calculated by multispectral imagery [26]:

$$V = \left(\frac{N\_{\nu} - N\_{\nu 0}}{N\_{\nu 1} - N\_{\nu 0}}\right)^2 \tag{5}$$

where *Nv*<sup>0</sup> is NDVI threshold for open soil, *Nv*<sup>1</sup> is NDVI threshold for full coverage of vegetation,

$$N\_{\nu} = \frac{\rho\_{\text{air}} - \rho\_{\text{rad}}}{\rho\_{\text{air}} - \rho\_{\text{rad}}} \tag{6}$$

Wind erosion is caused by the interaction of the structural soil particles from the ground-level air flow. A simplified model of wind erosion is given by [27]:

$$
\omega\_w \approx 0.059 \left(\omega - \mu\right) d\_s^{-3.67} \tag{7}
$$

where *zw* is the quantity of wind erosion (mm/month), *w* is the near-surface airflow velocity (m/s), *u* is critical air flow velocity (m/s),

$$
u = 3.202 + 0.025d\_s \tag{8}$$

and *ds* is soil structural particles equivalent size (mm). The near-surface airflow velocity at a steady dynamic wind velocity *w*0 is determined mainly by vegetation resistance [28]:

$$\dot{m} = \dot{w}\_0 \exp\left(-0.0139V\right) \tag{9}$$

The total soil erosion is a summation of (2) and (7) values.

2 1.67 (tg ) exp( 0.07 ) *S S z kQ* = a

in soil *R* (mm/month):

60 Land Degradation and Desertification - a Global Crisis

imagery [26]:

vegetation,

where *ks* is a soil erosion factor, *ks* ≅ 0.13 (mm/month)−1 for clay and sandy soils of the study area [23], *Q* is runoff (mm/month), α is a slope angle of the terrain, *V* is the vegetation cover fraction. Runoff is determined by the ratio of precipitation *P* (mm/month) and water retention

> <sup>2</sup> ( 0.2 ) 0.8 *P R <sup>Q</sup> P R*

> > <sup>1000</sup> 25.4 10 *S*

*C* æ ö = - ç ÷

*N N <sup>V</sup> N N* n n

*N*n

air flow. A simplified model of wind erosion is given by [27]:

(m/s), *u* is critical air flow velocity (m/s),

n

r

r

æ ö - <sup>=</sup> ç ÷

The percentage of vegetation coverage *V* is generally considered to be proportional to scaled NDVI value square within the study area [25] and is easily calculated by multispectral

> n

where *Nv*<sup>0</sup> is NDVI threshold for open soil, *Nv*<sup>1</sup> is NDVI threshold for full coverage of

*nir red nir red*

 r

 r

Wind erosion is caused by the interaction of the structural soil particles from the ground-level

where *zw* is the quantity of wind erosion (mm/month), *w* is the near-surface airflow velocity

2 0 1 0

where *R* depends on the table hydrological soil index *Cs* as [24]

*R*



è ø (4)

è ø - (5)


( ) 3.67 0.059 *w S z w ud* - » - (7)

3.202 0.025 *<sup>S</sup> u d* = + (8)

To map land degradation of the study area calibrated multispectral images from medium resolution, Earth observation satellite systems can be used for the period of analysis. All multispectral satellite images must be undertaken with atmospheric correction and then converted to surface reflectance for each spectral band. The MSAVI *Fv* (1) index must be calculated, and its changes must be mapped. At the same time, the total erosion *z* = *zs* + *zw* must be estimated and its changes must be mapped too. Required auxiliary parameters can be extracted directly from the input multispectral satellite imagery (fraction of vegetation cover), digital terrain elevations data (DTED—slopes), soil and climatic data [29, 30] (particle size distribution and hydrological parameters of soil, the average monthly rainfall, wind velocity profiles).

At the first stage of processing, time-series of satellite imagery-based classifications should be built. These classifications represent principally different land degradation indicators that are appeared in vegetation and soil erosion changes. At the second stage, the previously obtained partial first-level classifications should be fused into the resulting classification by data fusion methods [31]. In this study, Bayesian statistical inference can be applied as data fusion model [32]. Values obtained by fusing the partial classifications are subdivided conveniently into few classes. The first half of classes with negative values describes the negative changes of indicators which provide increasing land degradation risk.

The second half reflects the positive trends in land degradation indicators and shows a decrease in the risk of land quality deterioration. The special class must be reserved to map the territory where the evident changes did not occur during the period of analysis.

Thus, a hybrid two-level model for data fusion appears in land degradation risk mapping using remote sensing data and geospatial modelling: A few partial raster classifications are performed at the first level, and then, these classifications are fused into final map [33].

As relating to land degradation mapping, the geospatial model also has two levels. The model's first layer includes the spatial distribution of two main indicators of land degradation, namely trends in vegetation change and soil erosion. The model second layer provides the Bayesian fusion of the first-level data into the final map of land degradation. In detail, the geospatial model data flowchart is described in **Figure 3**.

At the model's first level, the data processing is performed in multiple concurrent threads to extract a temporal trends of land degradation indicators. For simplicity, **Figure 3** shows the multispectral imagery (*a, c*) and DTED (*b, d*) for the initial and final stages only. By the MSAVI (*e, g*), vegetation index maps the vegetation cover fractions (*h, j*) are estimated, and using additionally, the DTED and soil map (*f*) of territory the levels of soil erosion (*i, k*) are determined. At the model's second level, the partial classifications of trends in vegetation cover change (*l*) and soil erosion change (*m*) are fused into the land degradation final map (*n*) of study area.

**Figure 3.** The land degradation mapping dataflow diagram.
