- Email: [email protected]
Multi-temporal Multi-spectral and Radar Remote Sensing for Agricultural Monitoring in the Braila Plain
Available online at www.sciencedirect.com
ScienceDirect Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
“ST26733”, International Conference "Agriculture for Life, Life for Agriculture"
Multi-temporal multi-spectral and radar remote sensing for agricultural monitoring in the Braila Plain Violeta Poenaru1,2*, Alexandru Badea1, Sorin Mihai Cimpeanu2, Anisoara Irimescu3 1
2
Romanian Space Agency, 21-25 Mendeleev Str., 010362, District 1, Bucharest, Romania University of Agronomic Sciences and Veterinary Medicine of Bucharest, 59 Marasti Blvd, District 1, 011464, Bucharest, Romania 3 National Meteorological Administration, 97 Sos. Bucuresti-Ploiesti, 013686, Bucharest, Romania
Abstract The objective of the paper is to investigate the sensitivity of Landsat OLI and C-band radar signals to monitor an agricultural area affected by soil salinization and land degradation. The chosen test area - Braila Plain has the special particularities such as: dry climate, high annual average temperatures (9-110C), very dry and hot summers which cause a large potential evapotranspiration and conduct to a moisture deficit in soil, alkaline soils, winter winds with an average speed of 2.7 - 3.4 m/s. The soil type and climate conditions favor the culture of maize (50%), wheat and successive crops (16%), alpha-alpha (18%), sugar beet (6%), sunflower (7%), vegetables and other crops (3%). Taking into account the soil type, climate conditions and geomorphological characteristics of the studied area, the paper focuses on evaluation of Sentinel-1 sensor capabilities to monitor soil degradation and surface soil moisture. A multi-temporal series of Sentinel-1 data gathered from October 2014 until January 2015 is used. Crop growing stages are investigated with multi-temporal Landsat OLI and MODIS data. The normalized difference vegetation index (NDVI), specific leaf area index, land thermal index, soil moisture index and soil salinity information are retrieved from Landsat data. The potential evapotranspiration is computed from MODIS data to evaluate the effects of soil salinity on growing crops. The results confirm soil degradation and the synergy of using multi-spectral and radar data for crops monitoring.
© The Authors. Authors. Published Publishedby byElsevier ElsevierB.V. B.V. This is an open access article under the CC BY-NC-ND license © 2015 2015 The (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the University of Agronomic Sciences and Veterinary Medicine Bucharest. Peer-review under responsibility of the University of Agronomic Sciences and Veterinary Medicine Bucharest Keywords:multi-spectral data, SAR data, Braila Plain, soil degradation, vegetation parameters
* Corresponding author. +4-021-316-8722, Fax: +4-021-312-8804 E-mail address: [email protected]
2210-7843 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the University of Agronomic Sciences and Veterinary Medicine Bucharest doi:10.1016/j.aaspro.2015.08.134
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
1. Introduction The environmental factors that affect plant growing conditions with impact on the crop productivity. The description of degradation processes includes soil quality, geomorphologic conditions, terrain slope, landscape conditions, vegetation coverage, bio-diversity and applied irrigation solutions. The development of irrigation and drainage infrastructure and subsequently decreases of the irrigated areas by decommissioning altered the hydrologic regime of the soil and facilitated a secondary salinity. The first ameliorative measures were applied in Romania since 1977 when studies regarding the land reclamation and management practices for salt-affected soils have been conducted by National Research and Development Institute for Soil Science, Agricultural Chemistry and Environment (ICPA) (Dumitru et al. 2003). The solutions for controlling soil salinity and improving soil status and crop yield involve drainage, land levelling, leaching, fertilization and crop rotation. Intense drainage was determined significant yield enhancement in the experimental field LaculSarat during six years agricultural monitoring (Cotet, 2011). Saline soils located on low lands, in depressionary areas with low natural drainage, represents 4.2% of the arable land of Romania (about 614 000 ha).Remote sensing imagery has a significant potential for detecting and monitoring salt affected soils often in the advanced stage of salization processes (Wu et al., 2008; Allbed and Kumar., 2013; Shrestha and Farshad, 2008; Wu et al., 2012). In visual image interpretation, a poor crop stand on a salt affected soil shows a different surface reflectance than that of a healthy vegetation cover on a salt-free soil. In the microwave wavelengths, the signal is sensitive to water content and dielectric properties of the targets therefore the radar backscattering coefficients are modelled as function of imaginary part of dielectric constant of salt affected soil samples (Aly et al., 2007; Lasne et al., 2008; Lhissou et al., 2013). In this perspective, the aim of the paper is to investigate the sensitivity of Landsat OLI and Sentinel-1 C-band radar signals to monitor an agricultural area affected by soil salinity and land degradation (crusts), located in Braila Plain. The performance of several multispectral vegetation indices such as normalized difference vegetation index (NDVI), specific leaf area index (LAI), land thermal index, soil moisture index and soil salinity information are examined to assess salinity stress. Moreover, the potential evapotranspiration is computed from MODIS data to evaluate the effects of soil salinity on growing crops. Sensitivity of backscattering coefficients to soil moisture index and salinity index is inspected. 2. Materials and Methods Description of the test area and dataset The test area - Braila Plain, situated in south-eastern part of Romania, in the Romanian Plain, includes a part of Inferior Siret River meadow, a part of the Baraganului Plain, a small part of the Salcioara Plain and the Buzau Plain (Fig. 1a). The relief is generally homogeneous with rivers, tablelands and lake depressions. The chosen test area presents peculiar conditions: dry climate, high annual average temperatures (9-110C), very dry and hot summers which cause a significant potential evapotranspiration and conduct to a moisture deficit in soil, alkaline soils (Fig. 1b), winter winds with an average speed of 2.7 - 3.4 m/s. The soil type and climate conditions are favourable for the culture of maize (50%), wheat and successive crops (16%), alpha-alpha (18%), sugar beet (6%), sunflower (7%), vegetables and other crops (3%). The investigations have been focused on estimating of vegetation and soil land dynamics using time series Landsat TM data. 16 day Landsat NDVI composite with a 30m spatial resolution and 8-day evapotranspiration data with 1 km resolution 8-day MODIS global evapotranspiration (MOD.16) product were used. Also, a time series of IWS, GRD, dual polarized Sentinel 1 data acquired on ascending and descending nodes between August 2014 and January 2015 has been used to derive backscattering coefficients temporal profiles and to relate them on the soil moisture estimated from Landsat data and potential evapotranspiration estimated from MODIS data. The Landsat TM images gathered between 1986 and 2011 period were exploited to derived fifteen years of NDVI, LAI and SI (salinity index) temporal profiles. A Landsat OLI data collected on 25.08.2014 was used to validate Sentinel-1 results. In this study have been selected 3 salt-affected soil test areas, 4 agricultural test areas and 1 periurban salt-affected test area.
507
508
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Fig. 1. (a) Test area – Braila agricultural area; (b) Soil classification (left) and soil salinization classification (right)of the test area: Scale 1:200000. © National Research and Development Institute for Soil Science, Agricultural Chemistry and Environment (ICPA)
Remote sensing techniques and methodology Assessing soil salinity is time-demanding and laborious as the changes in salinity must be monitored over a long time. This is possible at a reasonable cost thanks to time series remotely sensed data.There are several techniques to map and monitor saline soils among which: supervised classification, spectral extraction and matching techniques (Howari, 2003), changes in surface characteristics expressed as vegetation indices and changes in radiometric thermal temperature (Masoud and Koike, 2006), SAR backscattering models to relate the dependence between soil moisture, soil salinity and the backscattering coefficient (Aly and Magagi, 2004; Dobson et al., 1985) and SAR polarimetry to examine the spatio-temporal dynamics of soil salinity (Barbouchi et al., 2014).
509
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Multispectral analysis The normalized difference vegetation index (NDVI) is sensitive to salinity because it normalizes green leaf scattering in the near infrared wavelength and chlorophyll absorption in the red wavelength (Tucker, 1979) (eq. 1).
NDVI =
ρ NIR − ρ RED ρ NIR + ρ RED
(1)
NDVI is an ambiguous indicator of the extent of salinity because it also relates to other crop variables such as: biomass, leaf area, plant water content, nitrogen and chlorophyll content. Moreover, this index is sensitive to the optical properties of soil and is suitable to estimate the vegetation cover fraction in the agricultural areas (Yang et al., 2011). The salinity index (SI), which combine blue and red band, is sensitive to the surface reflectance of the saltaffected land (eq. 2) (Douaoui et al., 2005).
SI = ρNIR * ρRED
(2)
Normalized difference soil index (NDSI), expressed as in eq. 3, leaf area index (LAI) (eq. 4, eq.5), brightness index (BI) (eq. 6), vegetation soil salinity index (VSSI) (eq.7), land surface temperature (eq. 8-9), normalized difference build-up area index (NDBI) (eq. 10) and potential evapotranspiration are computed in addition to NDVI to better understanding the relationship between soil and vegetation stress. ρ − ρ NIR NDSI = RED (3) ρ RED + ρ NIR
EVI = 2.5*
ρ NIR − ρRED ρ NIR −7.5 * ρBLUE +1
(4)
LAI = 3.618* EVI − 0.118
(5)
BI = ρ RED 2 + ρ NIR 2
(6)
VSSI = 2* ρGREEN − 5* ( ρ RED + ρ NIR )
(7)
L = Lmin + (( Lmax − Lmin ) / 255) * Q
(8)
T = k2 /(ln(k1 / L + 1))
(9)
Where: L - value of radiance in thermal infrared; T – ground temperature and Q – digital record. NDBI =
ρterm − ρ NIR ρterm + ρ NIR
(10)
510
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Evapotranspiration (ET) is a combined process of transpiration from surfaces of plant leaves and evaporation from the soil in which is growing. Plant transpiration is controlled by canopy conductance while soil evaporation reflects water balance of a crop, when consider water use efficiency. Potential evapotranspiration (PET) derived from MOD.16 ET/PET data uses Penman-Monteith (P-M) equation (Monteith, 1965) to calculate global remotely sensed ET (Mu et al., 2007; Mu et al., 2009) and integrates both P-M and Priestley-Taylor methods (Priestley and Taylor, 1972). Surface moisture is an important parameter used in modelling water plant coupled land surface processes and land-air climatic systems. A method of detecting soil moisture from Landsat TM data uses surface temperature and NDVI space to estimate spatial patterns of soil moisture (eq. 11) (Zeng and Xiang, 2004). SMI =
Ts max − Ts Ts max − Ts min
(11)
where: Tsmax, Tsmin are the maximum and minimum surface temperature for a given NDVI and Ts are the remotely sensed data-derived surface temperature for a given NDVI. Backscatter analysis In order to examine the backscatter behaviour of salt-affected soils, we extracted the backscattering coefficients of different type of soils from calibrated Sentinel-1 data. Images acquired on 10 August, 22 August, 14 November and 8 December 2014 has been analyzed as a one dataset. The images acquired on 9 August, 21 August, 8 October, 25 November 2014 and 12 January 2015 have been analyzed as a second dataset. As first step, we used 1x1 multilooked data to improve SAR image quality by reducing the SAR speckle. Secondly, we applied Range-Doppler terrain correction method to produce an orthorectified product in the WGS geographic coordinates. This method uses available orbit state information extracted from the metadata, the radar timing annotation, the slant to ground range conversion parameters together with the reference Digital Elevation Model (DEM) to derive the precise orbit geolocation. Thirdly, the radiometric calibration LUT to relate pixel value (digital number) with the radar backscatter of the scene (eq. 12) was applied.
value ( i ) =
DNi Ai2
2
where depending on the selected LUT: value (i) = ıi0; and Ai = ıi0
(12)
It is important to notice that a bilinear interpolation was used for the pixels that fall between points in the LUT. After calibration, the datasets were co-registered and a Gamma MAP multi-temporal filter with a 5x5 window was applied. 3. Results and Discussions Soil and vegetation indices analysis The images data considered in this study were acquired since July until end August when vegetation cover reflects more in near-infrared. Vegetation and soil indices were applied to all Landsat TM data for mapping and interpretation. Salinity maps are established based on a false color composite (Fig. 2a): vegetation appears in shades of green, urban infrastructures are white-cream while soils varying from grey, blue to white-blue. Inspection of Landsat TM images reveals a higher reflectance in all spectral bands for saline soils. Five soil salinity index classes were found in this study as follows: saline water and salty lake (0.04-0.08), slightly saline (0.08-0.10), moderately saline (0.10-0.15), strongly saline (0.15-0.20) and very strongly saline (> 0.20) (Fig. 2b). Two Landsat data acquired on July 2003 and July 2011 were used to detect and monitor soil salinity, evapotranspiration and NDVI in the Braila Plain. The majority of pixels in the 2003 image have salinity index values ranging between 0.033 and 0.66, whereas the values in the 2011 image ranged between 0.038 and 0.72. It is possible to identify a little increase of the soil salinity during 8 years spinning time (Table 1 and Table 2, Fig. 3).
511
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Fig. 2. (a) False color composite of Landsat data acquired; (b) Salinity index map derived from Landsat data acquired on 16.07.2011 (left) on 16.07.2011 (R: TM1; G: TM2, B: TM3): detection and 26.07.2003 (right)of salt affected soil represented with white–blue color. Table 1. Correlation salinity index for July 2003 Index
NDVI
SI
BI
NDSI
VSSI
Temp
LAI
NDVI SI BI NDSI VSSI Temp LAI
1 0.119841 -0.03889 0.466959 0.328887 0.407788 -0.47575
0.119841 1 0.937011 0.057419 -0.56979 -0.34761 0.024392
-0.03889 0.937011 1 -0.17945 -0.76883 -0.57208 0.2852
0.466959 0.057419 -0.17945 1 0.764838 0.868942 -0.98976
0.328887 -0.56979 -0.76883 0.764838 1 0.93227 -0.82895
0.407788 -0.34761 -0.57208 0.868942 0.93227 1 -0.9099
-0.47575 0.024392 0.2852 -0.98976 -0.82895 -0.9099 1
512
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516 Table 2. Correlation salinity index for July 2011 Index NDVI SI BI NDSI VSSI Temp LAI
NDVI 1 -0.25906 0.713327 -0.99725 -0.84414 0.101664 0.962644
SI -0.25906 1 0.315468 0.18918 -0.19631 0.292137 -0.07603
BI 0.713327 0.315468 1 -0.76041 -0.9759 -0.0694 0.874804
NDSI -0.99725 0.19291 -0.76041 1 0.880435 -0.10178 -0.9782
VSSI -0.84414 -0.19631 -0.9759 0.880435 1 -0.00618 -0.95451
Temp 0.101664 0.292137 -0.0694 -0.10178 -0.00618 1 0.010205
LAI 0.962644 -0.07603 0.874804 -0.9782 -0.95451 0.010205 1
Fig. 3. Dependence of salinity index on NDVI – linear regression
In 2003, the correlation coefficient is positively better for temperature and VSSI (93%) than SI and NDVI (12%). In 2011, the better correlation is found for NDSI and VSSI (88%) whereas SI and NDVI present lower correlation (25%). Similar results were obtained in the NDVI and SI analysis on agricultural field and salt-affected area (Fig. 4): negative values of NDVI correspond to the sowing with lower salinity index values (May) and positive values for salt-affected soil (winter wheat). This is possible due to rehabilitation measures and ameliorative works in Big Island of Braila and agricultural areas in the North Braila Terrace. a
b
Fig. 4. Multitemporal data: NDVI dependence on salinity index (a) agricultural area sample; (b) salt-effected soil.
Potential evapotranspiration had a linear dependence on the temperature acquired on Braila meteorological station: pick values of PET are followed by pick values of temperature (Fig. 5a). Vegetation cover type has also an important role in evaporation because different vegetation types have different rates of transpiration and produce different amounts of turbulence above the canopy (the greater the turbulence, the greater the evaporation) (Fig. 5b). Change detection statistics method based on class by class image difference was performed on NDVI data. As vegetation cover indicator, the NDVI based vegetation cover classification was produced by means of segmentation technique and compared with spatio-temporal changes during 1986-2011 periods. Segmentation of a time series attempts to divide it into homogeneous subsequence, such that each of these segments is different from each other (Im et al., 2008).
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
A high threshold was applied on classified data to separate real changes and seasonal or inter-annual variability. The result is shown in Fig. 6.
a)
b) Fig. 5. (a) Potential evapotranspiration derived from MOD.16 products (left); (b). Agreement between NDVI and PET: 60% (2003) and 61% (2011).
Fig. 6. Change detection, NDVI: areas unchanged appear in grey-dark tone, land changes are red, land cover changes are blue and seasonal changes are green
513
514
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Backscattering coefficients analysis For each test area, samples of 5 x 5 pixels are extracted from each SAR data to estimate the mean backscatter coefficient value. Mean, standard deviation, the minimum value, the maximum value and backscatter coefficient were inspected to ensure the sigma 0 value is representative (Fig. 7a). Also, the temporal behaviour of backscatter coefficients of a salt-affected soil were investigated: - 10 dB for bare soil that decreasing to -20 dB in moisture conditions (Fig. 7b). Multi-temporal SAR images have been classified by applying segmentation method. We used a maximum likelihood classifier on each of the homogeneous segments identified by the segmentation. These classifications highlight surface moisture content as is observed in Fig. 8a. Consequently, surface soil moisture and NDVI derived from Landsat OLI imagery acquired on 25 August were used to investigate the relationship between backscatter coefficient soil moisture and NDVI. The results shown in Fig. 8b depict a good agreement (about 33%).
a)
b) Fig. 7 (a). Multi-temporal filtering of Sentinel imagery: R: 14.11.2014; G: 10.08.2014; B: 22.08.2014. Temporal changes appear as blue-redgreen tone while unchanged areas are presented in grey tones. (b). Temporal profiles of the backscattered coefficients in a salt-affected soil.
a)
b)
Fig. 8. (a) SAR classification based on sigma nought values. Red: water bodies and surace soil moisture; azure: urban areas; blue: vegetated areas and yellow: bare soils. (Sentinel -1 image gathered on 22.08.2014). (b). Relationship between Sigma 0 acquired in GRD mode and soil moisture and LAI estimated from Landsat OLI data
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
515
4. Conclusions The study assesses the sensitivity of Landsat OLI and Sentinel-1 C-band radar signals to monitor an agricultural area from Braila Plain affected by soil salinity and land degradation. As the result of this analysis, multi-temporal images of Landsat (2003 and 2011) and Sentinel -1 (2014) proven to be very useful to identify salt affected soils. The results indicate that NDVI generates more accurate crop condition estimates science it eliminates influence of salt-affected soil and inter-annual variability in arable land utilization. This is possible due to rehabilitation measures and ameliorative works. The results confirm soil degradation and the synergy of using multi-spectral and radar data for crops monitoring. In the absence of in-situ measurements, the multi-temporal satellite data are helpful for continuous monitoring of the soil salinity dynamics in the region. Acknowledgements The paper was published under the frame of European Social Fund, Human Resources Development Operational Programme 2007-2013, project no POSDRU/159/1.5/S/132765. This research work was carried out with the support of Romanian Space Agency and also was financed from Project PN II Partnership No 171/2013. References Allbed A., Kumar L. 2013. Soil salinity mapping and monitoring in arid and semi-arid regions using remote sensing technology: A review. Advanced in Remote Sensing, 2, 373-385. Aly Z., Bonn F., Magagi R. 2004. Modelling the backscattering coefficient of salt-affected soils: Application to Wadi El-Natrun bottom. EARSeLeProc., vol. 3, no. 3, pp. 372–381. Aly Z.,Bonn F.J., Magagi R. 2007. Analysis of the backscattering coefficient of salt affected soils using modelling and Radarsat-1 SAR data. Geoscience and Remote Sensing, IEEE Transaction on, vol 45, no. 2, 332-341. Barbouchi M., Abdelfattah R., Chokmani K., Ben Aissa N., Lhissou R., El Harti A. 2014. Soil Salinity Characterization Using PolarimetricInSAR Coherence: Case Studies in Tunisia and Morocco. Selected Topics in Applied Earth Observations and Remote Sensing, IEEE Journal of,vol.PP, no.99,pp.1-10. Cotet V., 2011. Effect of ameliorative works in experimental field LacuSarat, Braila County. Scientific Papers, USAMV Bucharest, Series A, Vol. LIV. Dobson M. C., Ulaby F. T., Hallikainen M. T., El-Rayes M.A. 1985. Microwave dielectric behavior of wet soil—Part II: Dielectric mixing models. IEEE Trans. Geosci. Remote Sens., vol. GRS-23, no. 1, pp. 35– 46. Douaoui A.E.K., Nicolas H., Walter, C. 2006. Detecting salinity hazards within a semiarid context by means of combining soil and remotesensing data. Geoderma 134(1–2): 217–230. Dumitru M., Ciobanu C., Motelica M. D., 2003. Romania soil quality, in Rehabilitation and Management of polluted soils, Proceedings of an international workshop, Braila, Romania, p. 91-130. Howari, F.M. 2003. The use of remote sensing data to extract information from agricultural land with emphasis on soil salinity. Australian Journal of Soil Research 41(7): 1243–1253. Im J., Jensen J.R., Tullis J.A. 2008. ObjectǦbased change detection using correlation image analysis and image segmentation. International Journal of Remote Sensing, 29:2, 399-423. Lasne Y., Paillou P., Ruffie G., Serradilla C., Demontoux F., Freeman A., Farr T., Mcdonald K., Chapman B., Malezieux. 2008. Effect of salinity on the dielectric properties of geological materials: implication for soil moisture detection by means of remote sensing. IEEE Trans. On Geosc. And Rem. Sens., 46 (6), 1674-1688. Lhissou R., Chokmani K., El Harti A., Abdelfattah R., Barbouchi M. 2013. Soil salinity estimation using RADARAT 2 polarimetric data in arid and sub-arid regions: Marocco and Tunisia cases. Geophisical Research, vol 15, EGU. Masoud, A.A., Koike, K. 2006. Arid land salinization detected by remotely-sensed landcover changes: A case study in the Siwa region, NW Egypt. Journal of Arid Environments 66(1): 151–167. Monteith J. L. 1965. Evaporation and environment. The State and Movement of Water in Living Organisms. G. E. Fogg, Ed., Symposia of the Society for Experimental Biology, Vol. 19, Academic Press, 205–234 Mu Q., Heinsch F.A., Zhao, M., Running, S.W. 2007. Development of a global evapotranspiration algorithm based on MODIS and global meteorology data. Remote Sensing Environ., 111,519–536 Mu Q., Jones L.A., Kimball J.S., McDonald K.C., Running S.W. 2009. Satellite assessment of land surface evapotranspiration for the pan-Arctic domain. Water Resour. Res., 45, W09420. Priestley C.H.B., Taylor, R.J. 1972. On the assessment of surface heat flux and evaporation using large-scale parameters. Mon. Wea. Rev., 100, 81–92. Shrestha D. P., Farshad A. 2008. Mapping salinity hazard: An integrated application of Remote Sensing and Modelling-Based Techniques. Remote sensing of soil salinization: Impact on land management. CRC press, pp 257-270.
516
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516 Tucker C. J., 1979. Red and photographic infrared linear combinations for monitoring vegetation, Remote Sensing of the Environment, 8, 127150. Wu J., Vincent B., Yang J, Bouarfa S., Vidal A. 2008. Remote sensing monitoring of changes in soil salinity: a case study in Inner Mongolia, China. Sensor, 8, 7035 – 7049. Wu Y., Han X., Wang W. 2012.Discrimination soil salinization distribution in the middle region of Heihe River Basin using TM data. Geoscience and Remote Sensing Symposium, IEEE International, 4225-4228. Yang Y., Zhu J., Zhao C., Liu S., Tong S. 2011. The spatial continuity study of NDVI based on Kriging and BPNN algorithm. Mathematical and Computer Modelling. Vol 54, Issue 3-4, pp 1138 – 1144. Zeng Y., Feng Z., Xiang N. 2004. Assessment of soil moisture using Landsat ETM+ temperature/vegetation index in semiarid environment. Geoscience and Remote Sensing Symposium, IEEE International, vol.6, pp.4306 -4309.
ScienceDirect Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
“ST26733”, International Conference "Agriculture for Life, Life for Agriculture"
Multi-temporal multi-spectral and radar remote sensing for agricultural monitoring in the Braila Plain Violeta Poenaru1,2*, Alexandru Badea1, Sorin Mihai Cimpeanu2, Anisoara Irimescu3 1
2
Romanian Space Agency, 21-25 Mendeleev Str., 010362, District 1, Bucharest, Romania University of Agronomic Sciences and Veterinary Medicine of Bucharest, 59 Marasti Blvd, District 1, 011464, Bucharest, Romania 3 National Meteorological Administration, 97 Sos. Bucuresti-Ploiesti, 013686, Bucharest, Romania
Abstract The objective of the paper is to investigate the sensitivity of Landsat OLI and C-band radar signals to monitor an agricultural area affected by soil salinization and land degradation. The chosen test area - Braila Plain has the special particularities such as: dry climate, high annual average temperatures (9-110C), very dry and hot summers which cause a large potential evapotranspiration and conduct to a moisture deficit in soil, alkaline soils, winter winds with an average speed of 2.7 - 3.4 m/s. The soil type and climate conditions favor the culture of maize (50%), wheat and successive crops (16%), alpha-alpha (18%), sugar beet (6%), sunflower (7%), vegetables and other crops (3%). Taking into account the soil type, climate conditions and geomorphological characteristics of the studied area, the paper focuses on evaluation of Sentinel-1 sensor capabilities to monitor soil degradation and surface soil moisture. A multi-temporal series of Sentinel-1 data gathered from October 2014 until January 2015 is used. Crop growing stages are investigated with multi-temporal Landsat OLI and MODIS data. The normalized difference vegetation index (NDVI), specific leaf area index, land thermal index, soil moisture index and soil salinity information are retrieved from Landsat data. The potential evapotranspiration is computed from MODIS data to evaluate the effects of soil salinity on growing crops. The results confirm soil degradation and the synergy of using multi-spectral and radar data for crops monitoring.
© The Authors. Authors. Published Publishedby byElsevier ElsevierB.V. B.V. This is an open access article under the CC BY-NC-ND license © 2015 2015 The (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the University of Agronomic Sciences and Veterinary Medicine Bucharest. Peer-review under responsibility of the University of Agronomic Sciences and Veterinary Medicine Bucharest Keywords:multi-spectral data, SAR data, Braila Plain, soil degradation, vegetation parameters
* Corresponding author. +4-021-316-8722, Fax: +4-021-312-8804 E-mail address: [email protected]
2210-7843 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the University of Agronomic Sciences and Veterinary Medicine Bucharest doi:10.1016/j.aaspro.2015.08.134
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
1. Introduction The environmental factors that affect plant growing conditions with impact on the crop productivity. The description of degradation processes includes soil quality, geomorphologic conditions, terrain slope, landscape conditions, vegetation coverage, bio-diversity and applied irrigation solutions. The development of irrigation and drainage infrastructure and subsequently decreases of the irrigated areas by decommissioning altered the hydrologic regime of the soil and facilitated a secondary salinity. The first ameliorative measures were applied in Romania since 1977 when studies regarding the land reclamation and management practices for salt-affected soils have been conducted by National Research and Development Institute for Soil Science, Agricultural Chemistry and Environment (ICPA) (Dumitru et al. 2003). The solutions for controlling soil salinity and improving soil status and crop yield involve drainage, land levelling, leaching, fertilization and crop rotation. Intense drainage was determined significant yield enhancement in the experimental field LaculSarat during six years agricultural monitoring (Cotet, 2011). Saline soils located on low lands, in depressionary areas with low natural drainage, represents 4.2% of the arable land of Romania (about 614 000 ha).Remote sensing imagery has a significant potential for detecting and monitoring salt affected soils often in the advanced stage of salization processes (Wu et al., 2008; Allbed and Kumar., 2013; Shrestha and Farshad, 2008; Wu et al., 2012). In visual image interpretation, a poor crop stand on a salt affected soil shows a different surface reflectance than that of a healthy vegetation cover on a salt-free soil. In the microwave wavelengths, the signal is sensitive to water content and dielectric properties of the targets therefore the radar backscattering coefficients are modelled as function of imaginary part of dielectric constant of salt affected soil samples (Aly et al., 2007; Lasne et al., 2008; Lhissou et al., 2013). In this perspective, the aim of the paper is to investigate the sensitivity of Landsat OLI and Sentinel-1 C-band radar signals to monitor an agricultural area affected by soil salinity and land degradation (crusts), located in Braila Plain. The performance of several multispectral vegetation indices such as normalized difference vegetation index (NDVI), specific leaf area index (LAI), land thermal index, soil moisture index and soil salinity information are examined to assess salinity stress. Moreover, the potential evapotranspiration is computed from MODIS data to evaluate the effects of soil salinity on growing crops. Sensitivity of backscattering coefficients to soil moisture index and salinity index is inspected. 2. Materials and Methods Description of the test area and dataset The test area - Braila Plain, situated in south-eastern part of Romania, in the Romanian Plain, includes a part of Inferior Siret River meadow, a part of the Baraganului Plain, a small part of the Salcioara Plain and the Buzau Plain (Fig. 1a). The relief is generally homogeneous with rivers, tablelands and lake depressions. The chosen test area presents peculiar conditions: dry climate, high annual average temperatures (9-110C), very dry and hot summers which cause a significant potential evapotranspiration and conduct to a moisture deficit in soil, alkaline soils (Fig. 1b), winter winds with an average speed of 2.7 - 3.4 m/s. The soil type and climate conditions are favourable for the culture of maize (50%), wheat and successive crops (16%), alpha-alpha (18%), sugar beet (6%), sunflower (7%), vegetables and other crops (3%). The investigations have been focused on estimating of vegetation and soil land dynamics using time series Landsat TM data. 16 day Landsat NDVI composite with a 30m spatial resolution and 8-day evapotranspiration data with 1 km resolution 8-day MODIS global evapotranspiration (MOD.16) product were used. Also, a time series of IWS, GRD, dual polarized Sentinel 1 data acquired on ascending and descending nodes between August 2014 and January 2015 has been used to derive backscattering coefficients temporal profiles and to relate them on the soil moisture estimated from Landsat data and potential evapotranspiration estimated from MODIS data. The Landsat TM images gathered between 1986 and 2011 period were exploited to derived fifteen years of NDVI, LAI and SI (salinity index) temporal profiles. A Landsat OLI data collected on 25.08.2014 was used to validate Sentinel-1 results. In this study have been selected 3 salt-affected soil test areas, 4 agricultural test areas and 1 periurban salt-affected test area.
507
508
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Fig. 1. (a) Test area – Braila agricultural area; (b) Soil classification (left) and soil salinization classification (right)of the test area: Scale 1:200000. © National Research and Development Institute for Soil Science, Agricultural Chemistry and Environment (ICPA)
Remote sensing techniques and methodology Assessing soil salinity is time-demanding and laborious as the changes in salinity must be monitored over a long time. This is possible at a reasonable cost thanks to time series remotely sensed data.There are several techniques to map and monitor saline soils among which: supervised classification, spectral extraction and matching techniques (Howari, 2003), changes in surface characteristics expressed as vegetation indices and changes in radiometric thermal temperature (Masoud and Koike, 2006), SAR backscattering models to relate the dependence between soil moisture, soil salinity and the backscattering coefficient (Aly and Magagi, 2004; Dobson et al., 1985) and SAR polarimetry to examine the spatio-temporal dynamics of soil salinity (Barbouchi et al., 2014).
509
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Multispectral analysis The normalized difference vegetation index (NDVI) is sensitive to salinity because it normalizes green leaf scattering in the near infrared wavelength and chlorophyll absorption in the red wavelength (Tucker, 1979) (eq. 1).
NDVI =
ρ NIR − ρ RED ρ NIR + ρ RED
(1)
NDVI is an ambiguous indicator of the extent of salinity because it also relates to other crop variables such as: biomass, leaf area, plant water content, nitrogen and chlorophyll content. Moreover, this index is sensitive to the optical properties of soil and is suitable to estimate the vegetation cover fraction in the agricultural areas (Yang et al., 2011). The salinity index (SI), which combine blue and red band, is sensitive to the surface reflectance of the saltaffected land (eq. 2) (Douaoui et al., 2005).
SI = ρNIR * ρRED
(2)
Normalized difference soil index (NDSI), expressed as in eq. 3, leaf area index (LAI) (eq. 4, eq.5), brightness index (BI) (eq. 6), vegetation soil salinity index (VSSI) (eq.7), land surface temperature (eq. 8-9), normalized difference build-up area index (NDBI) (eq. 10) and potential evapotranspiration are computed in addition to NDVI to better understanding the relationship between soil and vegetation stress. ρ − ρ NIR NDSI = RED (3) ρ RED + ρ NIR
EVI = 2.5*
ρ NIR − ρRED ρ NIR −7.5 * ρBLUE +1
(4)
LAI = 3.618* EVI − 0.118
(5)
BI = ρ RED 2 + ρ NIR 2
(6)
VSSI = 2* ρGREEN − 5* ( ρ RED + ρ NIR )
(7)
L = Lmin + (( Lmax − Lmin ) / 255) * Q
(8)
T = k2 /(ln(k1 / L + 1))
(9)
Where: L - value of radiance in thermal infrared; T – ground temperature and Q – digital record. NDBI =
ρterm − ρ NIR ρterm + ρ NIR
(10)
510
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Evapotranspiration (ET) is a combined process of transpiration from surfaces of plant leaves and evaporation from the soil in which is growing. Plant transpiration is controlled by canopy conductance while soil evaporation reflects water balance of a crop, when consider water use efficiency. Potential evapotranspiration (PET) derived from MOD.16 ET/PET data uses Penman-Monteith (P-M) equation (Monteith, 1965) to calculate global remotely sensed ET (Mu et al., 2007; Mu et al., 2009) and integrates both P-M and Priestley-Taylor methods (Priestley and Taylor, 1972). Surface moisture is an important parameter used in modelling water plant coupled land surface processes and land-air climatic systems. A method of detecting soil moisture from Landsat TM data uses surface temperature and NDVI space to estimate spatial patterns of soil moisture (eq. 11) (Zeng and Xiang, 2004). SMI =
Ts max − Ts Ts max − Ts min
(11)
where: Tsmax, Tsmin are the maximum and minimum surface temperature for a given NDVI and Ts are the remotely sensed data-derived surface temperature for a given NDVI. Backscatter analysis In order to examine the backscatter behaviour of salt-affected soils, we extracted the backscattering coefficients of different type of soils from calibrated Sentinel-1 data. Images acquired on 10 August, 22 August, 14 November and 8 December 2014 has been analyzed as a one dataset. The images acquired on 9 August, 21 August, 8 October, 25 November 2014 and 12 January 2015 have been analyzed as a second dataset. As first step, we used 1x1 multilooked data to improve SAR image quality by reducing the SAR speckle. Secondly, we applied Range-Doppler terrain correction method to produce an orthorectified product in the WGS geographic coordinates. This method uses available orbit state information extracted from the metadata, the radar timing annotation, the slant to ground range conversion parameters together with the reference Digital Elevation Model (DEM) to derive the precise orbit geolocation. Thirdly, the radiometric calibration LUT to relate pixel value (digital number) with the radar backscatter of the scene (eq. 12) was applied.
value ( i ) =
DNi Ai2
2
where depending on the selected LUT: value (i) = ıi0; and Ai = ıi0
(12)
It is important to notice that a bilinear interpolation was used for the pixels that fall between points in the LUT. After calibration, the datasets were co-registered and a Gamma MAP multi-temporal filter with a 5x5 window was applied. 3. Results and Discussions Soil and vegetation indices analysis The images data considered in this study were acquired since July until end August when vegetation cover reflects more in near-infrared. Vegetation and soil indices were applied to all Landsat TM data for mapping and interpretation. Salinity maps are established based on a false color composite (Fig. 2a): vegetation appears in shades of green, urban infrastructures are white-cream while soils varying from grey, blue to white-blue. Inspection of Landsat TM images reveals a higher reflectance in all spectral bands for saline soils. Five soil salinity index classes were found in this study as follows: saline water and salty lake (0.04-0.08), slightly saline (0.08-0.10), moderately saline (0.10-0.15), strongly saline (0.15-0.20) and very strongly saline (> 0.20) (Fig. 2b). Two Landsat data acquired on July 2003 and July 2011 were used to detect and monitor soil salinity, evapotranspiration and NDVI in the Braila Plain. The majority of pixels in the 2003 image have salinity index values ranging between 0.033 and 0.66, whereas the values in the 2011 image ranged between 0.038 and 0.72. It is possible to identify a little increase of the soil salinity during 8 years spinning time (Table 1 and Table 2, Fig. 3).
511
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Fig. 2. (a) False color composite of Landsat data acquired; (b) Salinity index map derived from Landsat data acquired on 16.07.2011 (left) on 16.07.2011 (R: TM1; G: TM2, B: TM3): detection and 26.07.2003 (right)of salt affected soil represented with white–blue color. Table 1. Correlation salinity index for July 2003 Index
NDVI
SI
BI
NDSI
VSSI
Temp
LAI
NDVI SI BI NDSI VSSI Temp LAI
1 0.119841 -0.03889 0.466959 0.328887 0.407788 -0.47575
0.119841 1 0.937011 0.057419 -0.56979 -0.34761 0.024392
-0.03889 0.937011 1 -0.17945 -0.76883 -0.57208 0.2852
0.466959 0.057419 -0.17945 1 0.764838 0.868942 -0.98976
0.328887 -0.56979 -0.76883 0.764838 1 0.93227 -0.82895
0.407788 -0.34761 -0.57208 0.868942 0.93227 1 -0.9099
-0.47575 0.024392 0.2852 -0.98976 -0.82895 -0.9099 1
512
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516 Table 2. Correlation salinity index for July 2011 Index NDVI SI BI NDSI VSSI Temp LAI
NDVI 1 -0.25906 0.713327 -0.99725 -0.84414 0.101664 0.962644
SI -0.25906 1 0.315468 0.18918 -0.19631 0.292137 -0.07603
BI 0.713327 0.315468 1 -0.76041 -0.9759 -0.0694 0.874804
NDSI -0.99725 0.19291 -0.76041 1 0.880435 -0.10178 -0.9782
VSSI -0.84414 -0.19631 -0.9759 0.880435 1 -0.00618 -0.95451
Temp 0.101664 0.292137 -0.0694 -0.10178 -0.00618 1 0.010205
LAI 0.962644 -0.07603 0.874804 -0.9782 -0.95451 0.010205 1
Fig. 3. Dependence of salinity index on NDVI – linear regression
In 2003, the correlation coefficient is positively better for temperature and VSSI (93%) than SI and NDVI (12%). In 2011, the better correlation is found for NDSI and VSSI (88%) whereas SI and NDVI present lower correlation (25%). Similar results were obtained in the NDVI and SI analysis on agricultural field and salt-affected area (Fig. 4): negative values of NDVI correspond to the sowing with lower salinity index values (May) and positive values for salt-affected soil (winter wheat). This is possible due to rehabilitation measures and ameliorative works in Big Island of Braila and agricultural areas in the North Braila Terrace. a
b
Fig. 4. Multitemporal data: NDVI dependence on salinity index (a) agricultural area sample; (b) salt-effected soil.
Potential evapotranspiration had a linear dependence on the temperature acquired on Braila meteorological station: pick values of PET are followed by pick values of temperature (Fig. 5a). Vegetation cover type has also an important role in evaporation because different vegetation types have different rates of transpiration and produce different amounts of turbulence above the canopy (the greater the turbulence, the greater the evaporation) (Fig. 5b). Change detection statistics method based on class by class image difference was performed on NDVI data. As vegetation cover indicator, the NDVI based vegetation cover classification was produced by means of segmentation technique and compared with spatio-temporal changes during 1986-2011 periods. Segmentation of a time series attempts to divide it into homogeneous subsequence, such that each of these segments is different from each other (Im et al., 2008).
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
A high threshold was applied on classified data to separate real changes and seasonal or inter-annual variability. The result is shown in Fig. 6.
a)
b) Fig. 5. (a) Potential evapotranspiration derived from MOD.16 products (left); (b). Agreement between NDVI and PET: 60% (2003) and 61% (2011).
Fig. 6. Change detection, NDVI: areas unchanged appear in grey-dark tone, land changes are red, land cover changes are blue and seasonal changes are green
513
514
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
Backscattering coefficients analysis For each test area, samples of 5 x 5 pixels are extracted from each SAR data to estimate the mean backscatter coefficient value. Mean, standard deviation, the minimum value, the maximum value and backscatter coefficient were inspected to ensure the sigma 0 value is representative (Fig. 7a). Also, the temporal behaviour of backscatter coefficients of a salt-affected soil were investigated: - 10 dB for bare soil that decreasing to -20 dB in moisture conditions (Fig. 7b). Multi-temporal SAR images have been classified by applying segmentation method. We used a maximum likelihood classifier on each of the homogeneous segments identified by the segmentation. These classifications highlight surface moisture content as is observed in Fig. 8a. Consequently, surface soil moisture and NDVI derived from Landsat OLI imagery acquired on 25 August were used to investigate the relationship between backscatter coefficient soil moisture and NDVI. The results shown in Fig. 8b depict a good agreement (about 33%).
a)
b) Fig. 7 (a). Multi-temporal filtering of Sentinel imagery: R: 14.11.2014; G: 10.08.2014; B: 22.08.2014. Temporal changes appear as blue-redgreen tone while unchanged areas are presented in grey tones. (b). Temporal profiles of the backscattered coefficients in a salt-affected soil.
a)
b)
Fig. 8. (a) SAR classification based on sigma nought values. Red: water bodies and surace soil moisture; azure: urban areas; blue: vegetated areas and yellow: bare soils. (Sentinel -1 image gathered on 22.08.2014). (b). Relationship between Sigma 0 acquired in GRD mode and soil moisture and LAI estimated from Landsat OLI data
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516
515
4. Conclusions The study assesses the sensitivity of Landsat OLI and Sentinel-1 C-band radar signals to monitor an agricultural area from Braila Plain affected by soil salinity and land degradation. As the result of this analysis, multi-temporal images of Landsat (2003 and 2011) and Sentinel -1 (2014) proven to be very useful to identify salt affected soils. The results indicate that NDVI generates more accurate crop condition estimates science it eliminates influence of salt-affected soil and inter-annual variability in arable land utilization. This is possible due to rehabilitation measures and ameliorative works. The results confirm soil degradation and the synergy of using multi-spectral and radar data for crops monitoring. In the absence of in-situ measurements, the multi-temporal satellite data are helpful for continuous monitoring of the soil salinity dynamics in the region. Acknowledgements The paper was published under the frame of European Social Fund, Human Resources Development Operational Programme 2007-2013, project no POSDRU/159/1.5/S/132765. This research work was carried out with the support of Romanian Space Agency and also was financed from Project PN II Partnership No 171/2013. References Allbed A., Kumar L. 2013. Soil salinity mapping and monitoring in arid and semi-arid regions using remote sensing technology: A review. Advanced in Remote Sensing, 2, 373-385. Aly Z., Bonn F., Magagi R. 2004. Modelling the backscattering coefficient of salt-affected soils: Application to Wadi El-Natrun bottom. EARSeLeProc., vol. 3, no. 3, pp. 372–381. Aly Z.,Bonn F.J., Magagi R. 2007. Analysis of the backscattering coefficient of salt affected soils using modelling and Radarsat-1 SAR data. Geoscience and Remote Sensing, IEEE Transaction on, vol 45, no. 2, 332-341. Barbouchi M., Abdelfattah R., Chokmani K., Ben Aissa N., Lhissou R., El Harti A. 2014. Soil Salinity Characterization Using PolarimetricInSAR Coherence: Case Studies in Tunisia and Morocco. Selected Topics in Applied Earth Observations and Remote Sensing, IEEE Journal of,vol.PP, no.99,pp.1-10. Cotet V., 2011. Effect of ameliorative works in experimental field LacuSarat, Braila County. Scientific Papers, USAMV Bucharest, Series A, Vol. LIV. Dobson M. C., Ulaby F. T., Hallikainen M. T., El-Rayes M.A. 1985. Microwave dielectric behavior of wet soil—Part II: Dielectric mixing models. IEEE Trans. Geosci. Remote Sens., vol. GRS-23, no. 1, pp. 35– 46. Douaoui A.E.K., Nicolas H., Walter, C. 2006. Detecting salinity hazards within a semiarid context by means of combining soil and remotesensing data. Geoderma 134(1–2): 217–230. Dumitru M., Ciobanu C., Motelica M. D., 2003. Romania soil quality, in Rehabilitation and Management of polluted soils, Proceedings of an international workshop, Braila, Romania, p. 91-130. Howari, F.M. 2003. The use of remote sensing data to extract information from agricultural land with emphasis on soil salinity. Australian Journal of Soil Research 41(7): 1243–1253. Im J., Jensen J.R., Tullis J.A. 2008. ObjectǦbased change detection using correlation image analysis and image segmentation. International Journal of Remote Sensing, 29:2, 399-423. Lasne Y., Paillou P., Ruffie G., Serradilla C., Demontoux F., Freeman A., Farr T., Mcdonald K., Chapman B., Malezieux. 2008. Effect of salinity on the dielectric properties of geological materials: implication for soil moisture detection by means of remote sensing. IEEE Trans. On Geosc. And Rem. Sens., 46 (6), 1674-1688. Lhissou R., Chokmani K., El Harti A., Abdelfattah R., Barbouchi M. 2013. Soil salinity estimation using RADARAT 2 polarimetric data in arid and sub-arid regions: Marocco and Tunisia cases. Geophisical Research, vol 15, EGU. Masoud, A.A., Koike, K. 2006. Arid land salinization detected by remotely-sensed landcover changes: A case study in the Siwa region, NW Egypt. Journal of Arid Environments 66(1): 151–167. Monteith J. L. 1965. Evaporation and environment. The State and Movement of Water in Living Organisms. G. E. Fogg, Ed., Symposia of the Society for Experimental Biology, Vol. 19, Academic Press, 205–234 Mu Q., Heinsch F.A., Zhao, M., Running, S.W. 2007. Development of a global evapotranspiration algorithm based on MODIS and global meteorology data. Remote Sensing Environ., 111,519–536 Mu Q., Jones L.A., Kimball J.S., McDonald K.C., Running S.W. 2009. Satellite assessment of land surface evapotranspiration for the pan-Arctic domain. Water Resour. Res., 45, W09420. Priestley C.H.B., Taylor, R.J. 1972. On the assessment of surface heat flux and evaporation using large-scale parameters. Mon. Wea. Rev., 100, 81–92. Shrestha D. P., Farshad A. 2008. Mapping salinity hazard: An integrated application of Remote Sensing and Modelling-Based Techniques. Remote sensing of soil salinization: Impact on land management. CRC press, pp 257-270.
516
Violeta Poenaru et al. / Agriculture and Agricultural Science Procedia 6 (2015) 506 – 516 Tucker C. J., 1979. Red and photographic infrared linear combinations for monitoring vegetation, Remote Sensing of the Environment, 8, 127150. Wu J., Vincent B., Yang J, Bouarfa S., Vidal A. 2008. Remote sensing monitoring of changes in soil salinity: a case study in Inner Mongolia, China. Sensor, 8, 7035 – 7049. Wu Y., Han X., Wang W. 2012.Discrimination soil salinization distribution in the middle region of Heihe River Basin using TM data. Geoscience and Remote Sensing Symposium, IEEE International, 4225-4228. Yang Y., Zhu J., Zhao C., Liu S., Tong S. 2011. The spatial continuity study of NDVI based on Kriging and BPNN algorithm. Mathematical and Computer Modelling. Vol 54, Issue 3-4, pp 1138 – 1144. Zeng Y., Feng Z., Xiang N. 2004. Assessment of soil moisture using Landsat ETM+ temperature/vegetation index in semiarid environment. Geoscience and Remote Sensing Symposium, IEEE International, vol.6, pp.4306 -4309.