Home Articles Integrated Remote Sensing and factor analytic GIS model for evaluating groundwater pollution...

Integrated Remote Sensing and factor analytic GIS model for evaluating groundwater pollution potential

Integrated Remote Sensing and factor analytic GIS model for evaluating groundwater pollution potential

O. P. Dubey, D. C. Sharma
Irrigation Research Institute, Roorkee,
Roorkee 247667, Uttaranchal

Background

In order to fulfill growing needs, pollutants are being increasingly added to the groundwater system through various human activities and natural processes. Applications of fertilizers and pesticides to enhance crop production have become a common practice. In case fertilizer application exceeds the plant uptake, the residual joins the water table. This increases nitrate concentration in groundwater. Similarly, excess applications of pesticides that are complex organic chemicals may have adverse health effects. Long-term use of saline irrigation water combined with poor management and adverse climatic conditions for example, low rainfall and high evaporation, leads to accumulation of salts in the root zone. Poor agricultural practice results in a loss of crop yield and deterioration of soil structure. Poorly designed and improperly managed waste disposal sites contribute significant amount of leachate. This leachate may affect the water quality. Improperly designed and maintained septic tank becomes a threat to groundwater quality. Disposal of waste through wells adds pollutant to the groundwater and also accelerate their movement towards a production well. In and around major pollutant is industrial waste that includes heavy metals, toxic compounds and radioactive material. Another significant source of metal contaminants are tailing produced at mining sites. An indiscriminate groundwater development in coastal aquifers leads to an excessive saltwater intrusion. The impact of the intrusion is further aggravated by an excessive upcoming of the freshwater-saltwater interface below a pumping well, caused by faulty well design or operation (too long a pumping spell and/or too little a rest period between the two successive pumping spells). In India, even during ancient time, well-defined legislation was in existent to control the water pollution. The Athervaveda provides panel action for polluting water.

“}k u vkiLrUos {jUrq ;ks u% nqjfiz;s ra fun~/ke: A ifo=s.k I`fFkfo eksr iqukfe AA ” AvFkZosn dk.M 12 & ‘yksd 30 A

Which means, let us have the life-sustaining water through good conduct and proper means (adequate resources), ” O goddess Earth give punishment to those who are responsible for polluting water”.

In recent years, land and water sectors were put to stresses in order to meet the demand of the growing population. Groundwater is more dependable source of water as compared to surface water (Viverkar 1999). Gradually, quality of water in addition to quantity is gaining importance in the selection of suitable sites for the groundwater development. Groundwater Pollution Potential (GWPP) of any given geographical location depends upon a wide range of above surface, surface and subsurface environmental parameters Brown 1972, Jackson1986). Evaluation of GPP involves decision-making keeping in view multiple interwoven criteria (ESCAP 1996). Generally, data required for GWPP is either not available or not sufficient. Collection, storage, and processing of required data is difficult, costly, and time consuming. As a result GWPP studies are generally based on questionable data. Remotely sensed data has proven capability in providing many above surface, surface and sub surface characteristics of a land unit. Synergistic use of remote sensing and ancillary data can be made for the development of the database required for GWPP (Daniel.et.al.1994, ESCAP 1996) GIS can be used to store, process and retrieve the developed database. GPP. Groundwater and its pollution system is a complex system (Biawas 1971, Hamil et. al. 1996). In this study an attempt has been made to represent the GWP system by Factor Analytic Model (FAM).


Integrated Remote Sensing and factor analytic GIS model for evaluating groundwater pollution potential

Study area & data used

Present study has been carried in a part (Fig. 1) of Hardwar district of Uttaranchal State, India. Synergistic use of, (1) SOI topographic sheet at 1: 250,000 scale, (2) Landsat TM FCC at 1: 50,000 scale for April, June, August and November months of 1988 and 1998 along with field survey have been made to asses the GPP. The area under investigation Ratmau watershed, covering about 500 km Sq. is bounded between latitudes 290 55′ N to 300 10′ N and longitude 770 55′ to 780 05′ E. The area is drained by Ratmau stream system. In general area is slopping towards south. Summer season begins from the end of April continues up to middle of June. The temperature variation during summer months is between 130 C to 450C. During winter it is (-) 20 C to 340 C. Hills are generally covered with Sal, Shisham, Khair and Bambo. Ground elevation in the area varies between 810 m to 280 m above mean sea level. Ground slope varies between 25m/Km to 0.3m/Km.. The present study aims to study the response of Ratmau watershed by formulating rainfall-runoff relationship. Synergistic use of remote sensing techniques and conventional techniques been utilized to extract the required data. The soil of the study area is generally coarse grained and aquifers are water table type. The area receives about 1000 mm annual rainfall; vegetation density is highly variable in spatial and temporal domain. It varies from 0 % to 95 %. Ground elevation varies from 215 m to 400 m; land slope varies from 0.6 % to 6 %, depth of groundwater table varies from less than 1 m to more than 20 m. The database for the study was generated through the integrated used of remote sensing and conventional techniques. Capabilities of GIS have been used to analyze and process the database and evaluation of GPP. Various themes were integrated using the concepts of linear mixing of influential parameters.

The Factor Analytic Model (FAM)

The FAM involves decomposing (Satty 1988, Mandoza 1997) the complex groundwater pollution system in to a number of simpler components forming a cascade. At each cascade level decision have been taken in simpler manner. The decision process moves from one cascade to another to arrive at the final decision. FAM has been used to develop a decision support system for weighting a particular land characteristic keeping in view its GPP. The FAM finally ranks a land unit in to a predefined GPP Class, based on its attributes. FAM was developed and calibrated using historical database consisting of 200 observed records. The FAM is mathematically sound but pair wise comparison is highly subjective. In order to get optimal results with minimum subjectivity the FAM was further modified to accommodate multi criteria. The decision cascade process starts from the lowest cascade level and progressively moves upwards until final decision is made. At each level pair wise comparisons have been made between factors at that level. These comparisons lead to priority vectors that are propagated up the cascade to arrive at a final priority vector. Decision cascading has been carried out in the following steps (Table 1).

  1. The decision making process has been decomposed in to a set of cascades. At the top level is the goal of the analysis. The elements of the lower level include the attribute such as objectives perhaps even more redefined attributes follows at the next lower level – until the last level.
  2. In the second phase, pair wise comparisons of the attributes or elements at a particular cascade level relative to their contribution or significance to the elements of the next higher cascade level is made. This phase constitutes much of the evaluation (qualitative) or assessment (quantitative) of the decision making process. Specifically the input matrix of pair wise comparisons express the relative of influence of an element over the others.
  3. In the third phase, the pair wise input matrix is decomposed spectrally. Spectral decomposition provides an estimate of the relative influence weight (RIW) of the elements at a particular cascade.
  4. Groundwater Pollution Potential (GPP) of a land unit was determined
    by linear mixing the above surface, surface, and subsurface parameters
    influencing the GPP. Linear mixing modeling is a branch of statistical
    science (Wang 1990, Maselli et.al. 1996, Bryant 1996, Kant.and Badrinath 1998).
    It is a method of analyzing a set of observations (obtained from a given sample)
    from their inter correlation to determine whether the variations can
    be accounted adequately by a number of basic categories smaller than
    that which the investigation was started (Fruchter, 1967). Let us
    consider a multivariate system consisting of ‘p’ responses described
    by the observable random variables X1, X2, X3.. Xp.
    The observable random vectors have mean x and co variance ‘S’. The
    Linear Mixing Model (LMM) postulates that ‘X’ is linearly dependent
    upon few unobservable variables F1, F2, ….. Fm called
    linear composite (LC) and additional source of variations e1,
    e2, ….. ep called specific factors. Hence LMM
    in matrix form may be written as :

    X – x =L F + e

    Where, (X-x) is a vector having p elements containing deviations of
    observed variable X and its mean value x, L is matrix of LC loading
    having prows and m column, C is vector of Composite having m rows and
    e is error vector having p elements. From the above equation it is
    evident that “p” deviations (X1 – x1) …. (xp – xp )
    are expressed in terms of (p + m) random variables, c1, c2, …. cm, e1 …. ep.
    With so many unobservable quantities a direct solution of LMM from the
    observations on x1, x2 …. xp is
    difficult. However, with the help of following assumptions about the
    random vectors c’ and e’, the model reduces to simple and easy form.
    These assumptions are (1) Original variables are linearly related. (2)
    Common composite ‘c’ and unique factors ‘e’ have mean zero and standard
    deviation unity. (3) Common factor ‘c’ and unique factor ‘e’ is independent.
    The LMM proceeds by imposing conditions that allow one to uniquely estimate
    the loading and the specific variance matrix. The loading matrix is
    then rotated, where the rotation is determined by some, ‘ease of interpretation’,
    method. Once the loading and the specific variance matrix are obtained composites
    are identified and estimated values for the composites themselves
    (called composites scores) are frequently constructed. (Johnsons, and Wichern, 1988).
    In LMM method composites are determined so as to account for maximum variance
    of all the observed variables. The residual terms (i.e. specific factors e1) are assumed to
    be small in this method. (Joreskog, Klovan and Reyment ; 1976).

    Min [{(X-x) – LF}T { (X-x)-LF}]-1

    Subject to Sfi = 1, fi > 0

    Developed FAM was calibrated using historical database consisting of 200 memory records. The database was subjected to Factor Analysis. Weights or membership function each cascade level for different land characteristics has been developed. The importance of a particular land characteristic was decided on the basis of a linguistic measure of importance. A comparison was made between various themes, land elements on a common scale and a confusion matrix representing their relative importance was developed. The confusion matrix was decomposed spectrally in to components. The first component accounts about 90 % variation in the data. The vector corresponding to this component represents the weights to different land characteristic that were considered influencing the decision. For data mixing RIW at different cascade level is shown in the Table 1. These weights were used to map the GPP of the study area. GWPP map is shown in Fig. 2

Integrated Remote Sensing and factor analytic GIS model for evaluating groundwater pollution potential

Conclusions
For over all developments of a region reliable estimate of groundwater quality and quantity is of paramount importance. Generally sufficient data required for groundwater pollution potential mapping are not available for Indian watersheds. Satellite data can be analyzed to generate database required for GWPP studies. Generated database can be put to FAM for extracting the most influential composite and subsequently the variable loading. Using the proposed FAM the study area was classified in to different classes in terms of their potential to pollute the groundwater. The model efficiency was tested by carrying out field surveys and found to above 80 percent. The model can be used for evaluating the GPP in any area after calibration. The added advantage of the proposed approach is that it compresses the data up to 70% that helps in efficient analysis and prediction.

Table.1 Variable Loading (RIW)

Goal Level
Level I Level II Level III
Feature RIW Feature RIW Feature RIW
GWPP Surface (0.65) Land use (0.44) Agriculture
Water Body
Barren Land
Thin forest
Thick
forest
Settlement
0.40
0.25
0.18
0.10
0.05
0.02
Land Slope (0.26) Low Slope
Mild Slope
Milder Slope
0.72
0.21
0.07
Distance from Paleo Channel (0.18) Less than 50m
More than 50m
0.90
0.10
Distance from Flood Plain 0.04 Up to 50m
More than 50m
0.90
0.10
Soil 0.06 Sand
Sandy
loamLoamy
sandClay
0.56
0.27
0.13
0.04
Distance from Urban areas 0.02 Less than 0.5 km
0.5km – 1.0m
More than 1.0 km
0.65
0.28
0.07
Sub Surface 0.24 Aquifer Media (0.5) Sand and Boulder
Sand Boulder and Clay
Sand and Clay
0.63
0.28
0.09
Permeability in Vertical Direction (0.5) High
Low
0.90
0.04
Ground Water (0.11) Groundwater Depth (0.60) < 5m
5-15m
> 15m
0.73
0.19
0.08
Rainfall Recharge (0.32) High
Medium
Low
0.65
0.28
0.07
Water Quality (0.08) SAR Value Low
SAR Value High
0.75
0.25