Home Articles Studying saline soils through satellite images

Studying saline soils through satellite images

Salt-affected soils occur in every continent and under almost all climatic conditions. The presence of excess salts on the surface of the soil and in the root zone characterises all saline soils. The main source of salts in soil is the primary minerals in the exposed layer of the earth’s crust. Salinity occurs through natural or human-induced processes that result in the accumulation of dissolved salts in the soil water to an extent that inhibits growth of plants. Soil salinity and waterlogging are two main constraints present in irrigated agricultural lands. In India, the problem of salinity and alkalinity increases every year as a result of secondary salinisation. In India, about 8.6 mha (Pathak 2000) of land area is affected by soil salinity. Salt affected soils occur in states like Uttar Pradesh, Gujarat, West Bengal, Rajasthan, Punjab, Maharashtra, Haryana, Orissa, Delhi, Kerala and Tamil Nadu. Almost 2.8 million hectares of salt-affected soil are present within the Indo-Gangetic alluvial plain occupying parts of Punjab, Haryana, Uttar Pardesh, Delhi, Bihar and Rajasthan (Abrol et al. 1971). In south-west Punjab which includes districts of Ferozpur, Faridkot, Muktsar and Bhatinda, groundwater is of poor quality and cannot be used for irrigation.

Waterlogging and salinity problems are temporal as well as spatial in nature. They serve as a major soil degradation problem in Punjab. Waterlogging in the low lying areas is created due to seepage from irrigated uplands and seepage from canal water with inadequate water management practice (Goyal, et al. 2005). Sethi (1993) used IRS 1A, LISS II images for identification of salt-affected soils in Etah district. Apart from inventorying salt – affected soils, studies were also carried out to monitor their temporal behaviour using concurrent and historical satellite data/ aerial photographs (Singh, 1994). The approach to the problem of delineating saline soils using remote sensing data and GIS techniques has proved very efficient as per several recent studies (Dwivedi et al., 1998).

According to Jaishanker et al. (2005), vegetation indices are widely used to quantitatively assess the biophysical characteristics of vegetation from remote sensing measurements. Different indices have their own advantages in retrieving vegetation information. The study examines the correlations among different vegetation indices derived from a set of mustard, gram and wheat fields at three different phenological growth stages. Abdullah et al. (1978) investigated the effect of salinity on reproductive physiology of wheat. Salinity and sodicity stress bring about a reduction in plant growth and crop yield. Under both these conditions, plants exhibit stunted growth, poor tillering and branching and their flowering and maturity cycle gets delayed. Reduction in seed number and size also occurs in some plants.

Remote sensing, GIS and GPS have emerged as powerful survey tools in the natural resource inventory and data handling. Application of remote sensing technology in mapping and monitoring of degraded lands, especially salt-affected soils, has shown great promise of enhanced speed, accuracy and cost effectiveness (Dwivedi, 1996). A wide variety of satellite remote sensing data from Landsat – TM, SPOT, IRS 1C & 1D are now available to earth resource scientists for generating information on natural resources. Districts of Bhatinda and Muktsar were identified as intensive area study sites for this work.

Location and extent
The study area lies in Eco region 2 (M9E1), between geo-coordinates 30? 00` to 30? 15` N and 76? 30` to 76? 45` E. It is located in south-western part of Punjab (Fig.1).

Materials and methodology
For the purpose, multi-spectral and multi-date IRS 1D, LISS III, March 2000 images of the study area were procured from the National Data Center (NDC), Balanagar, National Remote Sensing Agency (NRSA), Hyderabad.

(i) Ancillary data

  1. Survey of India (SOI) Toposheet Maps (44-J/12). Scale –1:50,000
  2. District maps of Bhatinda and Muktsar: their Village maps.
  3. Cadastral map of study area.

The published soil survey reports, soil maps, atlas of Punjab, census report, water quality reports for the study area were collected and utilised during interpretation and field work.

(ii) Preparation of base map
The base maps indicating permanent cultural features such as major canals, highways, roads, railways, villages and sand dunes were prepared in Arc View 3.1 and ARC GIS 8.0 software. For this, Survey of India toposheet (44J/12) was used as reference. The Survey of India toposheet was georeferenced using the geographic latitude/ longitude WGS 84 coordinate system. The coordinate of the toposheet was converted into the geographic latitude/ longitude values in meters using ERDAS IMAGINE–8.1software. Finally one toposheet on 1:50,000 scale covering the study area was digitised to prepare the base map. A part of this irrigation network was digitised and added to the base map from satellite data to facilitate mapping of salt affected soils and secondary salinity.

(iii) Generation of False Colour Composite (FCC)
IRS 1D LISS III data has four bands: blue (0.45- 0.52?m), green (0.52-0.59?m), red (0.62- 0.68?m) and near infrared (0.77-0.86?m). The FCC (Fig.2) was generated by combination of three bands: infrared, red and green bands projecting as red, green and blue image planes. The standard false colour composite was used. The vegetation was represented by red colour instead of green colour in the false colour composite.

Methodologies (Fig.3) comprises Pre-field interpretation, Field work (Ground truth) and Post-field work (Analysis).

Fig.3 Overview of Methodology of digital data analysis for mapping Land use/land cover map
The enhanced false colour composite image of the study area (1:50,000 scale) was displayed on monitor. Standard FCC was visually interpreted for salt affected soils and waterlogged areas with the help of image elements like tone, texture, shape, size, pattern and association, etc. The salt-affected soils were depicted in tones of bright white to dull white with medium to coarse texture on Standard FCC as per the presence of salts on soil surface. The landforms associated with the occurrence of salt-affected soils were also considered during interpretation. The obstructions to natural drainage like roads, railway lines, distributaries, etc. could easily be identified on FCC images. The waterlogged areas appeared on the FCC image in tones of dark blue to black with a smooth texture. Additional pre- and post- monsoon images were used to permanently identify waterlogged area.

Initially rapid traverse of the study area was made to identify the sampling points in the area. Detailed field investigations were carried out in various physiographic units to observe the broad physiographic-soil relationship. The study area was surveyed three times in a year – December-January, February-March and April-May. The villages surveyed were Giddarbaha, Thiri, Malaut and Lambi. Soil and plant samples were collected with their geographic location using GPS. A reconnaissance survey of the study area was done using satellite images (FCC).

Salt affected lands were identified on the ground and ascertained on the satellite image by characterising their images (Fig.4a and 4b). Satellite image of IRS 1D LISS III of March 2000 was used for the purpose.

Fig.4 (a) Crop affected by salinity and Waterlogging

Fig.4 (b) Salt scalds on the soil surface
Modification of the post-field work primarily involved interpreting the mapping units on the FCC of the satellite data in the light of field information and soil physico-chemical data. On the basis of this information, a final map showing salt-affected soils was prepared.

Image classification, that is, categorisation of pixels based on their spectral characteristics is one of the fundamental analysis techniques for remotely sensed data. The two common methods of digital classification, that is, unsupervised and supervised methods were utilised to prepare the land use/ land cover from IRS 1 D LISS III satellite data of March 200).

During March, in the south-west Punjab, wheat and other vegetation cover are at an optimum growth condition. Therefore, various land cover types can be distinctly separated during classification. The application of unsupervised classification has had a significant advantage for the area under study because similar results were obtained for the same data set. The unsupervised classification was carried out for understanding the pixel separability of classes based on the clustering of DN values and respective cover type on the ground.

Supervised classification procedures are the most important analytical tools used for the extraction of more information from remotely sensed digital image data. The classification was carried out using the maximum likelihood classifier. Before starting this procedure, the possibility of separation of each class was determined from various enhancement techniques and the unsupervised classification techniques. The supervised classification using maximum likelihood algorithm quantitatively evaluated both the variance and co-variance categories of spectral responses. The training samples were assigned based on the ground truth information collected from each of the variable crop cover types in order to ensure the planimetric accuracy of the samples. The GPS points indicating geographical coordinates of each cover type were also considered.

IRS –1D Image data have been used to detect and assess the salt-affected land and waterlogged land and crop through a combination of visual and digital techniques. Methods commonly applied for the classification of digital remote sensing data are based on the radiometric information contained in the image bands. Ideally, pixels are expected to be, to a degree, more or less grouped in the multispectral space in clusters corresponding to different land cover types (Price 1994). The maximum likelihood classification algorithm was applied in this study. Both the unsupervised classification procedure and the supervised classification processes are discussed below.

Unsupervised classification of digital satellite data
The unsupervised classification is computer automated. The computer algorithm (unsupervised) identifies statistical part in the data without using any ground truth data. Therefore, the unsupervised ISODATA algorithm was used to perform this classification. Initially the land use/ land cover was classified with unsupervised classification into 16 clusters (spectral classes). The maximum number of iterations was kept at six. 16 clusters (classes) were enlarged after the iteration. These clusters were specified to classify the spectral variability representing the land use/ land cover in the study area. These classes represented wheat crop conditions influenced by salinity and normal conditions and were identified based on the field observations for the major land use /land cover classes. Subsequently, these clusters were studied to identify the land use/ land cover classes based on field data collected with GPS. The spectral groupings were assigned to land use/ land cover classes. Eventually seven land use/ land cover classes (clusters) were classified in the area. Spectral clusters of crops affected by moderate salinity were not distinguished by the unsupervised classification and merged with the normal crop. The crop affected by severe salinity could be separated into a cluster. Areas under fallow (sand dunes) were also distinguished clearly as a separate cluster.

The unsupervised classification (mainly wheat) affected by severe salinity, moderate salinity and normal crop were 9.65 per cent, 8.12 per cent and 53.44 per cent areas, respectively. 6.43 per cent area was classified as waterlogged. Plantation/ orchards, settlement and fallow land (sand dunes) classes were 3.26 per cent, 11.14 per cent and 7.96 per cent of total area, respectively (Table 1 and Fig.5).

Table1. Area statistics of various land use/ land cover based on unsupervised classification (IRS 1D, LISS III, 1st March 2000)
Supervised classification of digital satellite data
There were three basic stages involved in the supervised classification method: training stage, classification stage and accuracy assessment stage. In the training stage, pixels in the image that represented the typical spectral information about severe and moderate salinity were examined for studying the classification.

The maximum likelihood decision rule is based on a normalized (Gaussian) estimate of the probability density function of each class. The maximum likelihood classifier quantitatively evaluates both the variance and covariance of the category spectral response patterns when classifying an unknown pixel. These classes appeared distinctly on the standard FCC and their GPS locations were recorded to generate training sets. Crops affected by salinity were divided in two classes – crops affected by severe salinity and crops affected by moderate salinity. After generation of training sets, supervised classification was performed using the maximum likelihood classifier algorithm by ERDAS imagine software. Training sets were collected from the field survey by recording the geographic coordinates of homogeneous land-cover area with GPS. In the supervised classification, classes were identified and training sets were generated. These training sets were generated with the help of ground truth data.

The land use land cover classes classified (Table 2 & Fig.6) in the area are discussed below:

  1. Crops affected by severe salinity – The area under crops affected by severe salinity was identified to be about 11.14 per cent (64.37 sq km) of the total area with very low yield. The crops of this category have a distinct spectral signature and were easily marked in the training set generation.
  2. Crops affected by moderate salinity – The area under this category was mapped as 8.18 per cent (47.27 sq. km.) of the total area. The spectral signature was not very distinct and training sets were provided with GPS locations. The resulting spectra and locations were accurate and crops affected by moderate salinity were classified.
  3. Normal crops – Normal crops were depicted on the image (Std. FCC) with a red and bright red tonal variation. Training sets of this class were very well identified on the image. The crops with normal spectra were also growing on normal soils. The crop growth was observed to be in very good condition and a high crop yield was predicted. This class constituted about 54.21 per cent (313.26 sq. km) of the total area.
  4. Waterlogged/ canal area – In this category, areas where water accumulation was found on the surface was classified. The waterlogged area along with the canal area occupied nearly 36.0 sq. km. (6.23 per cent) of the total area.
  5. Plantation/ orchards – This land cover appeared in brownish tone and smooth texture. The training set was easily generated for this class. The area classified under this category was 24.04 sq km (4.16 per cent) in the classified image.
  6. Settlement – The area classified in this class covered an area of 59.75 sq. km (10.34 per cent) of the total area.
  7. Sand dunes – The sand dunes appeared as a bright cream white tone on the imagery. These were fallow or barren. They covered an area of 33.17 sq km (5.74 per cent) of total area.

Table 2. Area of various Land use / and land cover based on supervised classification (IRS 1D, LISS III, 1st March 2000)
(c) Accuracy assessment
An accuracy assessment determines the quality of information derived from remotely sensed data. Seven land cover classes were classified in the area. The accuracy of each class was assessed with reference pixels. Reference pixels were identified based on the ground truth survey.

The accuracy of each salt affected land use/ land cover class is provided in Table 3. The crops affected by severe and moderate salinity were classified with an accuracy of 81.65 per cent and 82.41 per cent, respectively. The normal crop class was classified with maximum accuracy of 91.71 per cent. Waterlogged area, plantation/ orchards, settlement and sand dunes were classified with an accuracy of 88.69 per cent, 82.05 per cent, 86.06 per cent and 84.81 per cent, respectively. The overall accuracy for the salt affected/ waterlogging land use/ land cover map was found to be 85.47 per cent. The accuracy assessments of the supervised classification of March data indicate the high level of accuracy that was achieved by correctly providing data and ground information. The supervised classification (maximum likelihood algorithm) provided vastly improved results over the unsupervised classification. However, the unsupervised classification was found to be an important tool in the initial analysis.

Table 3. Accuracy assessment of land use/land cover map (Producer and User accuracy percentage)
5.0 Conclusion
Using IRS 1D images of March 2000, studies were conducted to assess the effects of secondary salinisation on cereal crops in 577.86 sq km area of south-west Punjab (Bhatinda and Muktsar districts), a part of the vast Trans-Gangetic Plain region of the Indo-Gangetic alluvial plain. Recent advances in remote sensing technology have opened new vistas in inventory, characterisation and monitoring of degraded lands. Remote sensing and GIS have been effectively used to study the dynamic behaviour of salinity and waterlogging in this study. In south-west Punjab, areas affected by waterlogging due to seepage of water from canal and salinity due to salts on the surface appeared as a white salt encrustation. In districts of Bhatinda and Muktsar, salinity and waterlogging affected the low lying villages of Giddarbaha, Thiri, Fakarsar, Jandwala Chakatarsingh wala, Shekhu, Malaut, Danewala, Rathayan, Abulkharana, Mahuana, Tappakhera, Dewankhera, Adhnian, Sahnakhera, Kheowali, and Pajawa. Due to irregular irrigation, seepage and high water requirement of crops, the arid environment has turned into salt affected/ secondary salinised and waterlogged area. In the salt-affected lands, crops appear to wither away and there is a heavy loss of yield. Salt tolerant grasses and weeds cover the waterlogged areas. The digital unsupervised classification took into account all three bands – green, red and near infrared. Crops affected by moderate salinity were distinguished by unsupervised classification. The crops affected by severe salinity were clearly separated into a cluster. Area under fallow (sand dunes) was distinguished clearly as separate cluster. The unsupervised classification showed crops (mainly wheat) affected by severe salinity, moderate salinity and normal crop covered 9.65 per cent, 8.12 per cent and 53.44 per cent area, respectively. The waterlogged areas constituted 6.43 per cent. Plantation/ orchards, settlement and Fallow land (sand dunes) classes covered 3.26 per cent, 11.14 per cent and 7.96 per cent of total area, respectively. In the supervised classification, seven land use/ land covers classes were identified and training sets were generated. The classify land cover classes appear distinctly on the standard FCC and their GPS locations were recorded to generate training sets. Crops affected by salinity were classified in two classes – crop affected by severe salinity and crop affected by moderate salinity. The area under crops affected by severe salinity was classified of 11.14 percent (64.37 sq. km.). Crops affected by moderate salinity were mapped at 8.18 per cent (47.27 Sq. km.) of total area. The land cover class was classified with accuracy of 85.63 per cent. It is important to take into consideration that remote sensing and GIS are an efficient and accurate source of information especially in the study of salt affected and waterlogged areas. The present study has utilised satellite remote sensing data and GIS in characterisation and mapping of salt affected and waterlogged area, the GIS for generating digital database (spatial and non- spatial) and in identifying and mapping the crop area affected by salinity and waterlogging. The image data and the GPS enabled accurate collection of soil and plant samples for the study of influences of salinity on plant parameters (physiological and chemical). It is emphasised that waterlogging, salinity and secondary salinity in the villages of Muktsar and Bhatinda have reached a critical situation. In the upcoming days, the environmental crisis will escalate further.


  • Abdullah Z., Ahmed, R. and Ahmed, J. (1978). Salinity induced changes in the reproductive physiology of wheat plants. Plant and Cell Physiol., 19 (1): 99-106.
  • Abrol, A.P. and Bhumbla, D. R. (1971). Saline and alkali soils in India- their occurrence and management. World Soil Resources Report. 41-42.
  • Dwivedi, R. S. (1996). Monitoring of salt – affected soils of the Indo-Gangetic alluvial plains using principal component analysis. Int. J. Remote Sensing, 17(10): 1907-1914.
  • Dwivedi, R. S., and Sreenivas, K. (1998). Delineation of salt-affected soils and waterlogged areas in the Indo-Gangetic plains using IRS-1C LISS-III data. Int. J.
  • Goyal, V.C., Jain, S.K., Pareek, N. (2005). Water logging an drainage assessment in Ravi- Tawi irrigation command (J&K) using remote sensing approach. J. Indian Soc. Remote Sensing, 33 (1):17-23.
  • Jaishanker, R., Senthivel, T. and Sridhar, V.N. (2005). Comparison of vegetation indices for practicable homology. J. Indian Soc. Remote Sensing, 33(3): 395-404.
  • Pathak P. S., 2000. Agro forestry: A tool for arresting land degradation. Indian Farming, 49(11): 15-19.
  • Price, J. C. (1994). How unique are spectral signatures. Remote Sensing of Environ. 49: 181-186.
  • Sethi, M. (1993). Salt affected soils of Etah district extent, characteristics and agricultural evaluation. Ph.D. Thesis, J.N.U., New Delhi, India.
  • Singh, A. N. (1994). Monitoring change in the extent of salt affected soils in northern India. Int. J. Remote Sensing, 19(14): 2739-2751.