• No results found

Assessment of Soil Variability of South Moravian Region Based on the Satellite Imagery

N/A
N/A
Protected

Academic year: 2020

Share "Assessment of Soil Variability of South Moravian Region Based on the Satellite Imagery"

Copied!
11
0
0

Loading.... (view fulltext now)

Full text

(1)

119

Volume 66 14 Number 1, 2018

https://doi.org/10.11118/actaun201866010119

ASSESSMENT OF SOIL VARIABILITY

OF SOUTH MORAVIAN REGION BASED

ON THE SATELLITE IMAGERY

Jaroslav Novák

1

, Vojtěch Lukas

1

, Fernando Rodriguez‑Moreno

2

, Jan Křen

1

1 Department of Agrosystems and Bioclimatology, Faculty of AgriSciences, Mendel University in Brno, Zemědělská 1, 613 00 Brno, Czech Republic

2Department of Remote Sensing, Global Change Research Centre AS CR, Bělidla 4a, 603 00 Brno, Czech Republic

Abstract

NOVÁK JAROSLAV, LUKAS VOJTĚCH, RODRIGUEZ‑MORENO FERNANDO, KŘEN JAN. 2018. Assessment of Soil Variability of South Moravian Region Based on the Satellite Imagery. Acta

Universitatis Agriculturae et Silviculturae Mendelianae Brunensis, 66(1): 119 – 129.

The aim of this study was to analyse the sets of atmospherically corrected Remote Sensing (RS) data in order to assess the variability of bare arable land between different soil blocks in the selected area and evaluate the suitability of this approach for locally targeted management. The study was carried out on the territory of parts of South Moravian, Vysocina and Zlín regions. The RS data was analyzed by two models developed in the Arc GIS 10.22 environment using the Normalized Differential Vegetation Index (NDVI) and the Principal component analysis (PCA) analyzing bare soil (BS) with more than 50 % and more than 95 % of the block representation. A layer of agricultural land from the LPIS system was used to delimit arable land areas. For correct determinig of BS value for NDVI were carried out the terrestrial measurements in the monitored area by using the GreenSeeker handheld crop sensor and The FieldSpec® HandHeld 2 spectrometer. Based on these measurements and image dates, 0.2 NDVI was selected as the limit value. As a more suitable source for identifying BS, RapidEye appears to be able to identify an average of 26 % of the observed area as bare ground compared to Sentinel 2 data (22.5 % of the observed area) in models with more than 50 % By representing the BS in a block (NDVI 50, PCA 50).

In these versions of the model was more variable soil (VS) indicated by RapidEye. With more than 95 % of the BS in the block (NDVI 95, PCA 95) was found more variable soils by Sentinel 2.

This method of indirectly identifying soil variability can assist in the application of fertilizer or soil treatment in the area of site‑specific management.

Keywords: soil variability, remote sensing, RapidEye, SENTINEL 2, LPIS, PCA, NDVI, coefficient of variation

INTRODUCTION

Mapping of in‑field soil variability is one of the first important steps on the way to applying site‑specific management technologies in agricultural practice. Soil variability can be detected and evaluated through direct (contact) mapping methods, such as soil sampling and their subsequent analysis, or indirectly through digital Remote Sensing (RS) records. Knowing certain features, such as changes in surface reflectance, make these records possible to create a basic idea of soil variability without the need to use the results of other more complex or

costly methods of soil variability mapping (eg. soil sampling).

(2)

a useful technique for mapping the soil surface properties over bare soil pixels (Ouerghemmi et al., 2016). The content of organic carbon (Stevens et al., 2010)clay content (Chabrillat et al., 2002) or soil salinity (Metternicht and Zinck 2003)can be deduced from the reflectance in these bands.

Vegetation indexes enable, based on a linear or linear combination of red (R) and infrared (NIR), to improve the vegetation signals contained in RS data. They primarily change with the vegetation cover, but are also sensitive to other factors included in the data, such as soil variability, atmospheric conditions or sensing angle (Jiaguo Qi et al., 1994). To minimize these factors, a number of modified vegetation indices have been developed, which are dealt with, for example, by Qi et al. (1994a,b). Indices are most commonly derived from the standardized vegetation index (NDVI) defined by Rouse et al. (1974).

In their use, the spectral expression of vegetation is accentuated at the expense of soil. However, they are less sensitive to changes in vegetation (unlike NDVI) at lower LAI values and are also more susceptible to atmospheric anomalies (J. Qi et al., 1994; Jiaguo Qi et al. 1994). If the soil is not covered by vegetation, the NDVI values are the lowest in light soils or when the soil is covered with snow. Higher are for darker soils with higher organic matter content. Higher values of NDVI correspond to higher LAI values. However, the relationship is not purely linear (Tiffin and Ross‑Ibarra 2014). NDVI has a positive correlation with carbon, nitrogen, soil moisture (Sumfleth & Duttmann 2008) organic matter (Liu et al., 2015), pH and organic carbon (Kumar et al., 2016).

Principal component analysis (PCA) was used to highlight of BS variability. Li a Chen (2014) refers to the PCA as one of the methods for highlighting the BS for its further classification. Metternicht a Zinck (2003) mapped the soil salinity using the RS and PCA data.

For the assessment of soil heterogeneity, statistical variability characteristics (variance, variation span, standard deviation and coefficient of variability) can be used. Wollenhaupt et al. (1997) used coefficient of variability (CV) in the study of variability of accessible P, and K, organic matter content and other factors providing higher yield. CV and standard deviation were used, for example by Vivoni et al. (2008),for measuring of the soil moisture obtained from RS data.

Braidek et al. (2007) shows the range of variability of soil properties based on their CV according to Tab. I. Here the variability of soil properties is divided into 4 categories at Low (CV < 15%), Moderate (CV15% – 35%), High (CV35% – 75%), Very high (CV 75% – 150%).

The aim of the paper was to capture, map and evaluate the variability of soil between different soil blocks in the selected territory based on the above findings using the digital records of the Remote Sensing through four variants of the analysis,

differing from the techniques used and the amount of BS areas entering the analysis.

MATERIALS AND METHODS

The study was carried out on the territory of parts of Southmoravien, Vysočina and Zlín regions (see Fig. 1). In this area, the soil is extensively used in agriculture (the view is about 90 %). At the time of shooting (March, April and September), the soil surface was only rarely covered with vegetation. The main types of soil types were represented in the monitored area: Cambisols (43.12 %) and Chernozems (22.94 %).

For greater representativeness, soil blocks of more than or equal to 10 hectares were selected. Their basic statistical characteristics are shown in Tab. II and Fig. 2. The relatively low variability of the soil blocks (Mean area = 31 ha and CV = 7.73 %) is evident from the above values despite the significant left‑hand histogram asymmetry. In the total area of the assessed soil blocks is 75 % in blocks up to 38.8 ha.

Two sets of soil blocks data were prepared with 8 meters (RapidEye) and 10 meters (Sentinel 2) buffer blocks to eliminate marginal effects.

Data sources

The first input were 4 scenes taken in March 2012, April 2012 and September 2012 by RapidEye. In these terms, the assumption of the highest incidence of BS was due to the beginning of the growing season (March and April) and the time after the harvest (September), the crops sown at this time did not affect the characteristics of bare arable land.

RapidEye, a heliosynchronous flight path launched in 2008 for agricultural applications (Tapsall et al. 2010; Brüser et al. 2014), allowing a potential daily review of the images. It carries 5 multispectral sensors that capture Blue (440 to 510 nm); Green (520 to 590 nm); Red (630 to 685 nm); Red‑edge (690 – 730nm) and near infrared (760 – 850 nm)(Adelabu et al. 2013). By the supplier (Gisat, Ins.) radiometric corrections and basic geometric corrections were performed. The scenes were further orthorectified in ENVI with using the Digital Terrain Model of the Czech Republic – 4th

generation and subjected to atmospheric corrections using the QUAC module, which allows atmospheric corrections for multispectral and hyperspectral images containing visible and NIR or SWIR bands. For more information on QUAC, see Bernstein (2012).

(3)

level and then corrected in the SNAP application by the SEN2COR module to the L2A level. Uwe et al. (2013) describe L2A data as an orthocticified digital surface reflectance record based on L1C data, which is data at the “Top‑Of‑Atmosphere” (TOA) level. L2A data, known as Bottom‑Of‑Atmosphere (BOA), contain maps outside the ortho‑corrected bands of atmospheric corrected reflections:

• Aerosol Optical Thickness (AOT), • water column,

• Scene classification,

• The probability of clouds and snow in the scene

(Uwe et al., 2013).

In the analysis, the spatial data from the LPIS (eagri.cz) agricultural land layers were also used, provided by the Ministry of Agriculture of the Czech Republic in the form of a spatial and descriptive representation of the agricultural land blocks. The data are valid for year 2012.

Ground measurements

A ground measurement has been performed to select the correct NDVI bare land level in the monitored area. The data were collected at 43 locations using the Green Seeker handheld crop sensor (Jones et al. 2015 and Field Spec® Handheld 2 spectrometer operating at wavelengths 325 – 10750 nm with an accuracy of ± nm and resolution <3.0 nm @ 700 nm).

At each of the monitored sites, the Green Seeker handheld crop (GS) sensor was subjected to 5 measurements, at which the mean values used for the analyzes were calculated. Subsequently, one Field Spec® Handheld 2 spectrometer was calibrated, which was calibrated at 15 minute intervals using the spectral standard. From the measured data, the NDVI values (Fig. 5) were calculated according to:

• Rouse et al. (1974) for wavelengths (NIR – 780 nm, RED – 670 nm),

I: Variability of soil properties according Braidek et al. (2007)

Coefficient of variation

Low (CV < 15 %) Moderate (CV 15 % – 35 %) High (CV 35 % – 75 %) Very high (CV 75 % – 150 %)

Soil hue and value pH Sand contentClay content Solum thicknessExchangeable Ca, Mg, K

Nitrous oxide flux Electrical conductivity

A horizon Thickness CEC Soil nitrate N

Saturated hydraulic conductivity

% BS Soil‑available P Solute dispersion coefficient

Silt content Ca CO3 equivalent Soil‑available K

Porosity Crop yield

Bulk density Soil organic C

(4)

• The wavelengths used of the Green Seeker handheld crop sensor (NIR – 780 nm, RED – 660 nm) (Trimble 2016).

The curve of the values measured by GS is generalized in some parts in Fig. 3. This is due to the averaging of the measured values and the larger sensing area than to the spectrometer. The larger scanning area is also reflected in the higher mean NDVI obtained from GS. Differences between soil types are not apparent in Fig. 3, but the differences between the measurements are significant (standard deviation of 0.03). On the basis of these facts, after the ground measurement and comparison with the available data, the value of 0.2 NDVI was chosen as the limit for bare soil mapping in the monitored dates.

Model description

To determine variability, two models were created in Arc GIS 10.22. The first one calculates soil variability directly from the NDVI layer soil layer with a BS fraction above 50 % (first situation) and over 95 % (second situation). These situations were chosen to test the suitability of the selected data acquisition dates. The CV (%) was used to evaluate soil variability.

Variability detected from NDVI

From the calculated layer of NDVI with disguised clouds a bitmap was created in which:

• 1 represents values below 0.2 NDVI, i.e. the BS assumed, but also the area, water areas, roads and others.

• The No Data value then represented NDVI values greater than 0.2 which were removed (forests, meadows and other green vegetation).

From the NDVI layer, it was necessary to select areas that actually represent arable land. The other data input was therefore the polygon layer of arable soil from iLpis. This classification was used

to calculate the percentage of arable land within a polygon.

This value served to determine the 50 % or 95 % coverage of the BS block. The selected blocks then enter the zonal statistics above the NDVI layer. The next step was to calculate the variation coefficient (CV) and its standardization.

The Scheme of the NDVI variability detection procedure is given in Fig. 4.

Variability detected from PCA

To highlight the variability of BS, the principal component analysis was used (Li and Chen 2014). The selected blocks entered the zonal statistics above the layer obtained from the PCA, and the CV was again calculated. Red and infrared radiation bands were included in the analysis because of their relative linear relationship in the interpretation of BS reflectance(Baret et al. 1993). The SWIR band was used to highlight the humidity when analyzing data from the Sentinel 2 satellite. The Assessment of soil variability based upon first component of PCA is given in Fig. 6.

Classification clue

The assessment of soil variability according to the CV was done according to Braidek et al. (2007) (see Fig. 1). Area categories of land were selected on the basis of criteria for the evaluation of agrochemical properties of soils (Beránek and Klement 2007) for better description between the blocks of different dimensions. The selected area categories also capture the most blocks in the monitored area (See Fig. 2).

RESULTS AND DISCUSSIONS

The findings of Montandon a Small (2008) have been confirmed, indicating that, in the absence of soil information, the most popular method is to use the lowest NDVI value in the observed scene. II: Basic statistic overview of the analyzed area

Count Sum Average Q25 Median Q75 Min Max Mode Std Vx Skew. Kurto.

8189 253869.1 31.00 14.6 22.5 38.8 10.0 256 10.7 2.33 7.73 2.33 7.73

(5)

However, this method does not take into account the spatial variability of the soil, which is often significantly reduced by selecting the low value. The authors mentioned the impact of soil reflection on the quantification of green vegetation from NDVI. They originated from the normally stated value, where the NDVI for BS is close to zero and usually reaches the lowest positive values of this index, but they found by calculating from 2906 soil samples that the mean value of this index for BS is much higher (NDVI = 0.2) and is highly variable (standard deviation = 0.1).

In the case of terrestrial measurements carried out by us using a spectrometer, the mean NDVI was lower (0.13 ± 0.03). However, the mean value measured by the Green Seeker sensor was higher than the values found by the spectrometer (0.18 ± 0.02) and close to 0.2 NDVI. Higher values from Green Seeker are due to differences in soil properties within a larger area. The detected values, however, far exceed the commonly reported NDVI values for BS.

The threshold value of NDVI (upper limit for arable land) affects soil properties such as color, brightness, carbon content, or humidity. Qi et al. (1994a) describes the influence of wet dark soil on different vegetation indices in a different vegetation cover. Indicates a value of 0.2 NDVI for dark and damp soil at zero vegetation coverage.

Jones et al. (2015) measured through a Green Seeker handheld crop sensor (Trimble Navigation Limited, Sunnyvale, CA) a reflection of 4 soil types and post‑harvest residues from 4 different angles. They found that drier soils have a statistically proven

lower NDVI than wetlands. The lower mean NDVI (0.18) was found for fine kaolithic soils. For sandy and clay‑sandy mixed soils, the mean NDVI was higher (0.25).

The amount of BS selected by individual satellites is captured in the following Tab. II. The area covered by clouds was taken into account when calculating the area under analysis. The development of captured area of arable land is shown in Tab. IV.

RapidEye captures the least BS selected by the filter in the first scene (March 25, 2012). The reason is data acquisition before or at the very beginning of spring work. This is evidenced, for example, by the Summary Reports of Regional Departments of Brno and Tábor of the State Phytosanitary Administration on the occurrence of harmful organisms and disturbances in 2012, which talk about many dislocations during March and April due to drought and frozen ground in March.

The image taken on April 27, 2012 already captures more than a quarter of the analyzed area. The RapidEye scenes, taken after the harvest, capture the largest area of soil selected by the filter, and the analyzed scenes represent the best option for mapping bare land. Almost 30 % of the BS obtained from the post‑harvest scene of Sentinel 2 then confirms the suitability of September terms. Most BS is capable of capturing RapidEye (average 26.3 % within one year), Sentinel 2 has slightly worse results (22.5 %), probably due to cloudiness and lower spatial resolution.

(6)

86.7 % for the 19.9 2015 slide). The binding strength between the input data (evaluated by correlation coefficients) was affected by the scanning time or the amount of identified arable land in the scene. This was true for both RapidEye and Sentinel 2 data.

The following tables 5 – 8 show the percentage of variables in each scenario.

NDVI 50

In the case of more than 50 % of the BS representation in the block (Tab. V), Sentinel 2 identified a larger range of variable soil (VS) (17422 ha) than RapidEye (14631 ha). This may be due to a higher resolution of RapidEye. The cause may also be the different crop composition in a given year or drought in the area.For both data sources, autumn imaging dates were richer in information on VS occurrence. According to the selected key, the largest surface of VS captured the second autumn term of Sentinel 2. Data RapidEye identifies the largest area of VS in the second spring time (27. 4. 2012 – 6492 ha). Despite the almost double area of arable land captured by RapidEye (156918.1 ha BS), the captured VS area is lower than that of Sentinel 2 (84.821 ha BS).

PCA 50

When determining the variability with a model with more than 50 % BS based on PCA (Tab. VI), the VS area in the RapidEye (16680.1 ha) and the Sentinel 2 (19101 ha) increased compared to the NDVI evaluation. The increase was mainly reflected in the moderate category for the spring

dates of both data sources. On the other hand, the BS area is lower than in the case of the NDVI‑only variant, in the second autumn term in RapidEye, even in the autumn term of Sentinel 2. In this data source, the SWIR band increases in the category high and Very high in both scenes.

NDVI 95

A model with a 95 % representation of arable land in a block, based on NDVI (Tab. VII), identified in the RapidEye scenes a larger BS area (5749.3 ha) compared to Sentinel 2 (4362.5 ha). Autumn terms are again richer for both data sources. Most BS is in moderate category. In the case of RapidEye, the BS area has slightly increased in area categories up to 20 ha and up to 30 ha. On the other hand, in the largest area category, both BS and low variability were low. In the case of Sentinel 2, the BS area increases in area categories up to 20 ha and up to 30 ha in autumn images only.

PCA 95

The last model, based on a 95 % representation of bare soil in a block (Tab. VIII), evaluates the smallest area of the listed variants as variable. Compared to the version based only on NDVI for RapidEye (5749.3 ha VS) and Sentinel 2 (4362.5 ha VS), it indicates a variable area of 4327.64 ha for RapidEye and an area of 3597.5 ha of BS as variable for Sentinel 2. In scenes made by the RapidEye, the occurrence of variable soil is random. In the case of Sentinel 2, the variable area increases only in the spring image on all Moderate areas.

III: Classification clue

Land AREA (ha)

CV

Low Moderate High Very High

< 20.0 CV < 15 % CV 15 %–35 % CV 35 %–75 % > 75 %

20.1–30.0 CV < 15 % CV 15 %–35 % CV 35 %–75 % > 75 %

> 30.0 CV <1 5 % CV 15 %–35 % CV 35 %–75 % > 75 %

IV: Overview of selected bare soil

Satellite RapidEye – 2012 Sentinel 2 – 2015/16

25. 3. 2012 27. 4. 2012 9. 9. 2012 11. 9. 2012 27. 3. 2016 19. 9. 2015

Analyzed area (ha) 153454.3 165103.6 152455.5 128833.0 258869.1

ha % ha % ha % ha % ha % ha %

50 % of bare soil in field 23035.3 15 42825.6 25.9 51985.2 34.1 39002.4 30.3 40293 16.5 44528 28.6

95 % of bare soil in field 6805 4.4 13257 9.1 20738.4 13.5 15741.5 9.8 10514 5.32 11012 7.1

Area covered by clouds

(7)
(8)

V:

E

valua

tion o

f soil v

ari

abili

ty on mo

re then 50

% bar

e soil in b

lo

ck based up

on ND

VI la

yer

NDVI 50 Satellite

Rapid Eye 2012 Sentinel 2 2015/16 Date

25. 3. 2012

27. 4. 2012

9. 9. 2012

11. 9. 2012

27. 3. 2016

19. 9. 2015

area (ha) 23035.3 42825.6 52054.8 39002.4 40293 44528 Categories L M H VH L M H VH L M H VH L M H VH L M H VH L M H VH < 20.0 ha 22.38 0.11 0.00 0.00 15.56 2.33 0.05 0.00 19.12 1.51 0.07 0.00 16.18 0.69 0.00 0.00 19.41 2.06 0.25 0.00 17.06 4.29 0.32 0.00 20.1–30.0 ha 19.49 0.00 0.00 0.00 13.91 2.52 0.00 0.00 16.67 1.60 0.14 0.00 15.25 0.65 0.00 0.00 16.40 2.65 0.00 0.00 15.71 3.58 0.23 0.00 > 30.0 ha 58.02 0.00 0.00 0.00 55.37 10.26 0.00 0.00 52.05 8.84 0.00 0.00 63.99 3.24 0.00 0.00 47.14 10.55 1.54 0.00 43.53 14.66 0.62 0.00 For eac h scene , the v ari abili

ty is e

xpr

essed as a

p

er

cent

ag

e) ( L

‑L ow , M ‑ M o der at e, H – Hig h, VH – V ery Hig h. VI: E valua tion o

f soil v

ari

abili

ty on mo

re then 50

% bar

e soil in b

lo

ck based on firs

t c

omp

onent o

f PCA anal

yz e PCA50 Satellite Rapid Eye 2012 Sentinel 2 2015/16 Date

25. 3. 2012

27. 4. 2012

9. 9. 2012

11. 9. 2012

27. 3. 2016

19. 9. 2015

area (ha) 23035.3 42825.6 52054.8 39002.4 40293 44528 Categories L M H VH L M H VH L M H VH L M H VH L M H VH L M H VH <20. 0 21.94 0.55 0.00 0.00 16.03 1.98 0.03 0.00 19.44 1.23 0.03 0.00 14.90 1.98 0.00 0.00 17.14 3.29 1.25 0.03 19.50 1.87 0.24 0.06 20.1–30.0 19.00 0.49 0.00 0.00 14.46 2.09 0.00 0.00 17.17 1.24 0.00 0.00 14.32 1.57 0.00 0.00 14.43 3.43 1.07 0.12 17.52 1.62 0.38 0.00 >30.0 55.90 2.13 0.00 0.00 52.65 12.76 0.00 0.00 54.03 6.86 0.00 0.00 60.89 6.34 0.00 0.00 38.14 15.55 5.09 0.45 47.49 10.17 1.06 0.09 For eac h scene , the v ari abili

ty is e

xpr

essed as a

p

er

cent

ag

e) ( L

(9)

VII:

E

valua

tion o

f soil v

ari

abili

ty on mo

re then 95

% bar

e soil in b

lo

ck based on ND

VI la

yer

NDVI 50 Satellite

Rapid Eye 2012 Sentinel 2 2015/16 Date

25. 3. 2012

27. 4. 2012

9. 9. 2012

11. 9. 2012

27. 3. 2016

19. 9. 2015

area 6805 14032 20738.4 15741.5 10513.8 11011.6 Categories L M H VH L M H VH L M H VH L M H VH L M H VH L M H VH <20.0 ha 24.23 0.00 0.00 0.00 19.27 3.42 0.14 0.00 20.80 2.21 0.08 0.00 14.15 0.68 0.00 0.00 24.76 1.09 0.17 0.00 20.98 5.40 0.75 0.00 20.1–30.0 22.50 0.00 0.00 0.00 16.29 3.42 0.00 0.00 16.53 2.00 0.13 0.00 14.08 0.84 0.00 0.00 18.62 1.90 0.00 0.00 18.33 4.99 0.43 0.00 >30.0 ha 53.26 0.00 0.00 0.00 47.70 9.76 0.00 0.00 51.16 7.08 0.00 0.00 65.33 4.93 0.00 0.00 45.40 7.62 0.44 0.00 31.78 17.34 0.00 0.00 For eac h scene , the v ari abili

ty is e

xpr

essed as a

p

er

cent

ag

e) ( L

‑L ow , M ‑ M o der at e, H – Hig h, VH – V ery Hig h. VIII: E valua tion o

f soil v

ari

abili

ty on mo

re then 95

% bar

e soil in b

lo

ck based up

on firs

t c

omp

onent o

f PCA anal

yz

e

PCA 50 Satellite

Rapid Eye 2012 Sentinel 2 2015/16 Date

25. 3. 2012

27. 4. 2012

9. 9. 2012

11. 9. 2012

27. 3. 2016

19. 9. 2015

area (ha) 6805 14032 20738.4 15741.5 10513.8 11011.6 Categories L M H VH L M H VH L M H VH L M H VH L M H VH L M H VH <20. 0 24.23 0.00 0.00 0.00 20.47 2.24 0.00 0.00 22.06 1.04 0.00 0.00 13.32 1.50 0.00 0.00 22.71 2.97 0.34 0.00 24.68 1.71 0.60 0.13 20.1–30.0 22.20 0.31 0.00 0.00 17.23 2.25 0.00 0.00 17.66 1.01 0.00 0.00 13.72 1.19 0.00 0.00 17.00 3.26 0.27 0.00 20.30 3.01 0.44 0.00 >30.0 53.26 0.00 0.00 0.00 50.88 6.94 0.00 0.00 53.61 4.64 0.00 0.00 64.58 5.68 0.00 0.00 41.74 11.72 0.00 0.00 40.06 9.07 0.00 0.00 For eac h scene , the v ari abili

ty is e

xpr

essed as a

(10)

CONCLUSION

Large NDVI variability with a maximum value of 0.2 for BS was found for ground measurements. This value is in line with some literary sources, but others exceed.

More bare soil was able to identify RapidEye with better spatial resolution. The reasons for the differences in the captured variability between the two data sources may also be the different crop composition in the given years or the occurrence of drought in the given locality, which was manifested mainly in 2012 (Malý 2012). Higher amounts of variable soils occur in autographs taken in the autumn, both for RapidEye and Sentinel 2. In versions of models that work with more than 50 % of the bare soil cover in a block, it means more variable land RapidEye. With more than 95 % of the bare soil representation in the block, it identifies more variable land of Sentinel 2.

Using PCA enhances variability and increases the area of variable plots. This change does not appear in all categories equally and is independent. More is shown in the PCA 50 version. This version of the model also makes it more apparent to use the SWIR band for Sentinel 2, which highlights the variability mainly in the high and very high land categories. In the PCA 95 version, highlighting is only visible in the spring image. The use of the first component was demonstrated, for example, by Mzuku et al. (2015) when they found that the first major component of the green, red and NIR bands had significant statistical relationships with organic carbon and sand, dust particles and some clay fractions. The individual spectral bands explained a significant part of the variability of soil moisture, its color, dry soil color, organic carbon, sand, dust and clay. (Chen et al., 2004) recalls the PCA limitation where the soil characteristics estimation by the PCA method is valid only for the analyzed area and can not be applied to other areas.

From the models used, the best results of the PCA 50 version, which has a large number of BSs from which it identifies more variable soils, achieves the best results. It therefore appears to be a more appropriate tool for identifying variable areas for locally targeted management.

This method of indirectly identifying soil variability can assist in the application of fertilizer or soil treatment in the area of site‑specific management.

Acknowledgements

This paper was supported by the project IGA FFWT MENDELU No. 59/2013, entitled “Evaluation of soil variability of the selected area with remote sensing data “ Data from the system iLPIS, in the form of spatial and descriptive representation of blocks of arable land as SHP format, was provided by the Ministry of Agriculture.

REFERENCES

ADELABU, S. et al. 2013. Exploiting machine learning algorithms for tree species classification in a semiarid woodland using RapidEye image. Journal of Applied Remote Sensing, 7(1): 73480.

BARET, F., JACQUEMOUD, S. and HANOCQ, J. F. 1993. The soil line concept in remote sensing. Remote

Sensing Reviews, 7(1): 65–82.

BERÁNEK, K. and KLEMENT, V. 2007. Variabilita agrochemických vlastností zemědělské půdy dle výsledků agrochemického zkoušení zemědělských půd v období 1999‑2004. Bulettin:Sekce ústřední kontroly, 4(1): 4–7. BERNSTEIN, L. S. 2012. Quick atmospheric correction code: algorithm description and recent upgrades.

Optical Engineering, 51(11): 111719.

BRAIDEK, J., PENNOCK, D. and YATES, T. 2007. Soil Sampling Designs. In: CARTER, M. R. and GREGORICH, E. G. (Eds.). Soil Sampling and Methods of Analysis. 2nd Edition. CRC Press, pp. 25–38.

BRÜSER, K. et al. 2014. Discrimination and characterization of management systems in semi‑arid rangelands of South Africa using RapidEye time series. International Journal of Remote Sensing, 35(5): 1653–1673.

CASTALDI, F. et al. 2016. Evaluation of the potential of the current and forthcoming multispectral and hyperspectral imagers to estimate soil texture and organic carbon. Remote Sensing of Environment, 179: 54–65. CHABRILLAT, S. et al. 2002. Use of hyperspectral images in the identification and mapping of expansive clay

soils and the role of spatial resolution. Remote Sensing of Environment, 82(2–3): 431–445.

CHEN, F. et al. 2004. Field Scale Mapping of Surface Soil Clay Concentration. Precision Agriculture, 5(1): 7–26. GOSWAMI, S., GAMON, J., VARGAS, S. and TWEEDIE, C. 2015. Relationships of NDVI, Biomass, and Leaf

Area Index (LAI) for six key plant species in Barrow, Alaska. PeerJ PrePrints, 3: e913v1.

JENSEN, J. R., 2014. Remote sensing of the environment: an earth resource perspective. 2nd Edition. Edinburgh: Pearson

Higher Education.

JONES, J. R. et al. 2015. Influence of soil, crop residue, and sensor orientations on NDVI readings. Precision

Agriculture, 16(6): 690‑704.

(11)

KUMAR, P. et al. 2016. Estimation of accumulated soil organic carbon stock in tropical forest using geospatial strategy. The Egyptian Journal of Remote Sensing and Space Science, 19(1): 109–123.

LI, S. and CHEN, X., 2014. A new bare‑soil index for rapid mapping developing areas using LANDSAT 8 data. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., 40(4): 139–144.

LIU, S. et al. 2015. Prediction of soil organic matter variability associated with different land use types in mountainous landscape in southwestern Yunnan province, China. CATENA, 133: 137–144.

MALÝ, A. 2012. Hydrological impacts of drought 2012 in southern Moravia. Brno: ČHMI. [Online]. Available at: http://www.tpuvz.cz/STUDIE/15.pdf [Accessed:2017, September 15].

METTERNICHT, G. and ZINCK, J. 2003. Remote sensing of soil salinity: potentials and constraints. Remote

Sensing of Environment, 85(1): 1–20.

MONTANDON, L. and SMALL, E. 2008. The impact of soil reflectance on the quantification of the green vegetation fraction from NDVI. Remote Sensing of Environment, 112(4): 1835–1845.

MZUKU, M., KHOSLA, R. and REICH, R. 2015. Bare Soil Reflectance to Characterize Variability in Soil Properties. Communications in Soil Science and Plant Analysis, 46(13): 1668–1676.

OUERGHEMMI, W. et al. 2016. Semi‑blind source separation for the estimation of the clay content over semi‑vegetated areas using VNIR/SWIR hyperspectral airborne data. Remote Sensing of Environment, 181: 251–263.

QI, J. et al. 1994. A modified soil adjusted vegetation index. Remote Sensing of Environment, 48(2): 119–126. QI, J., KERR, Y. and CHEHBOUNI, A., 1994. External factor consideration in vegetation index development.

In: Proc. of Physical Measurements and Signatures in Remote Sensing. ISPRS, pp. 723–730.

ROUSE, J. W. et al. 1974. Monitoring the vernal advancement and retrogradation (Greenwave effect) of natural vegetation. NASA/GSFCT Type III final report.

STEVENS, A. et al. 2010. Measuring soil organic carbon in croplands at regional scale using airborne imaging spectroscopy. Geoderma, 158(1–2): 32–45.

SUMFLETH, K. and DUTTMANN, R., 2008. Prediction of soil property distribution in paddy soil landscapes using terrain data and satellite information as indicators. Ecological Indicators, 8(5): 485–501.

TAPSALL, B., MILENOV, P. and TADEMIR, K., 2010. Analysis of Rapideye Imagery for Annual Landcover. Mapping as an Aid to European Union (EU) Common Agricultural Policy. In: WAGNER W. and SZÉKELY, B. (Eds.). ISPRS TC VII Symposium – 100 Years ISPRS. Vienna, Austria, July 5–7, 2010. IAPRS, Vol. XXXVIII, Part 7B

UWE, M.‑W. et al. 2013. Sentinel‑2 Level 2a Prototype Processor : Architecture , Algorithms and First Results.

In: ESA Living Planet Symposium 2013. Edinburgh, UK.

VIVONI, E. R. et al. 2008. Comparison of ground‑based and remotely‑sensed surface soil moisture estimates over complex terrain during SMEX04. Remote Sensing of Environment, 112(2): 314–325.

WOLLENHAUPT, N. C., MULLA, D. J. and CRAWFORD, C. A. G. 1997. Soil Sampling and Interpolation Techniques for Mapping Spatial Variability of Soil Properties. In: The State of Site-Specific Management for

Agriculture. Madison, WI: American Society of Agronomy, Crop Science Society of America, Soil Science

Society of America, 19–53.

References

Related documents

Another study that included patients with non-valvular persistent AF, but no other cardiovascular disease, reported that a low vitamin D level was associated with AF [16]..

The objective of this study was to assess the impact of implemented periodic safety update report (PSUR) system in our hospital via PSUR function assessment questionnaire (PFAQ)

Refinery margins, caught as they are be- tween the ever rising crude acquisition costs and the depressed ex-refinery prices, have thus tended to shrink and, on average, have

Химийн уриншийн талбайд гербицидын төрөл, цацах хугацааг зөв сонгосон нөхцөлд хог ургамлыг ихээхэн бууруулж тэр тусмаа олон наст хог ургамалтай тэмцэх нь илүү үр

3: The effect of PTU-exposure (1.5, 3, and 6 ppm) and levothyroxine therapy (Hypo 6 ppm + Levo) on the distance moved (A) and the escape latency (B) of the male

Мөн БЗДүүргийн нохойн уушгины жижиг гуурсанцрын хучуур эсийн болон гөлгөр булчингийн ширхгийн гиперплази (4-р зураг), Чингэлтэй дүүргийн нохойн уушгинд том

The high level of coastal litter in the study region adds to the findings of the Report on the State of the Mediterranean Marine and Coastal Environment, which

Sponge (Porifera) from the Mediterranean Coast of Turkey (Levantine Sea, eastern Mediterranean), with a Checklist of Sponges from the Coasts of Turkey..