Coastal landuse and land cover change and transformations of Kanyakumari coast, India using remote sensing and GIS

Coastal landuse and land cover change and transformations of Kanyakumari coast, India using remote sensing and GIS

S. Kaliraj

^a,⇑

, N. Chandrasekar

^b

, K.K. Ramachandran

^a

, Y. Srinivas

^b

, S. Saravanan

^b

aCentral Geomatics Laboratory (CGL), ESSO – National Centre for Earth Science Studies, Thiruvananthapuram 695011, India

bCentre for GeoTechnology, Manonmaniam Sundaranar University, Tirunelveli 627012, Tamil Nadu, India

a r t i c l e i n f o

Article history:

Received 6 January 2016 Revised 1 March 2017 Accepted 13 April 2017 Available online 28 April 2017

Keywords:

Landuse and land cover Change detection MLC algorithm Remote sensing and GIS

South West coast of Kanyakumari, South India

a b s t r a c t

The coastal landuse and land cover features in the South West coast of Kanyakumari are dynamically reg- ulated due to marine and terrestrial processes and often controlling by natural and anthropogenic activ- ities. The primary objective of this study is to estimate the decadal changes and their transformations of landuse and land cover (LULC) features under Level II category of USGS-LULC Classification System using Landsat ETM+ and TM images using Maximum Likelihood Classifier (MLC) algorithm for the period 2000–

2011. The classified LULC features are categorized as beachface land cover, cultivable lands, plantation and shrub vegetation, fallow land, barren land, settlements and built-ups, water bodies, and mining area, etc. The geo-database is prepared for LULC feature class with an attributes of name, location, area and spatial distribution, etc. It shows the larger area in beachface land cover (sandy beaches, foredunes, uplands, Teri dunes (laterite) and associated nearshore landforms), plantations, cultivable lands, fallows, and barren lands are converted into built-ups and it increases more than twice in the period of 10 years.

Using GIS techniques, the analysis of change detection matrix reveals that the total area of 45.90 km²in different LULC features periodically shifted or transformed from one state to another one or more states, i.e. the beachface land cover area of 1.24 km²is encroached for built-ups and 0.63 km²for placer mining during the decade. Meanwhile, the area of 0.21 km²in this cover is transformed into wetlands and salt- water bodies. During the past decade, the expansion of area in the built-ups and settlements are directly proportional to the growth of population, which produces severe threat to the coastal resources. Accuracy assessment of classified images shows the overall accuracy is estimated as 81.16% and 77.52% and overall Kappa coeffient statistical values of 0.83 and 0.76 for the year 2000 and 2011 respectively. Ground truth verification of the extracted LULC features performed using 120 samples (10 samples per class) reveals that the accuracy of classified features is 89%. This indicates the acceptable accuracy of the classified LULC features for landuse and land cover change studies. The geodatabase of LULC features is used as pri- mary source for sustainable land resource management in the coastal region.

Ó2017 National Authority for Remote Sensing and Space Sciences. Production and hosting 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/).

1. Introduction

In recent decades, the South-West coast of Kanyakumari district experience remarkable changes in landuse and land cover features due to impacts of marine and terrestrial factors and human activ- ities (Chandrasekar et al., 2013). The changes in landuse and land cover due to various physical, climatic and socio-economic factors

are directly influencing the socio-economic status of local people along the coastal area in space and time. Rapid growth of popula- tion, urbanization and tourism developmental activities are alter- ing the existing state of landuse and land cover features (Clark, 1982; Jaiswal et al., 1999; Chilar, 2000; Yuan et al., 2005; Joshi et al., 2011; Rawat et al., 2013). The landuse changes without con- sidering environmental sustainability is increasing the demand for land resources like agriculture, minerals, soil and water (Gibson and Power, 2000; Lu et al., 2004; Zoran, 2006; Santhiya et al., 2010). Nowadays, the rate of demand is exceeding than the rate of earth can provide sustainably causes more serious long-term sustainability issues along the coastal regions in worldwide (Xiuwan, 2002). Global assessment of percentage of land cover

http://dx.doi.org/10.1016/j.ejrs.2017.04.003

1110-9823/Ó2017 National Authority for Remote Sensing and Space Sciences. Production and hosting 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 National Authority for Remote Sensing and Space Sciences.

⇑ Corresponding author.

E-mail addresses:s.kaliraj@ncess.gov.in(S. Kaliraj),profncsekar@gmail.com(N.

Chandrasekar),raman.kk@ncess.gov.in(K.K. Ramachandran),drysv@yahoo.co.in(Y.

Srinivas),geosaravanan2000@yahoo.co.in(S. Saravanan).

Contents lists available atScienceDirect

The Egyptian Journal of Remote Sensing and Space Sciences

j o u r n a l h o m e p a g e : w w w . s c i e n c e d i r e c t . c o m

affected by human action vary from 20% to 100%, whereas, the humans appropriate 20% to 40% of the earth’s potential net pri- mary biological production (Richards, 1990; Nemani and Running, 1995; Brown et al., 1999; Small and Nicholls, 2003).

Nearly 24% of the earth surface area is likely experienced decline in ecosystem function and productivity during last two decades (Meyer and Turner, 1994; Kalensky et al., 2003), while the 43% of the earth surface area is experienced human-induced degradation (Daily, 1995).

UNPD (2007)has stated that the human population has doubled in the coastal regions during past 40 years, and to increase by the same amount again in the next 40 years. Rapid population growth leads quickened the pace of land transformation on the natural lan- duse and land cover features sue to expansion of settlements and infrastructure (Mujabar and Chandrasekar, 2012). In recent dec- ades, the changes in coastal landuse and land cover due to human activities have wrought in the earth’s life support system causes major issues worried by many people (Luong, 1993; Jaiswal et al., 1999; Misra et al., 2013). Such negative changes are deter- mine, in full or part of the vulnerability of places and people to cli- matic, economic or socio-political perturbations of the regions (Nicholls and Small, 2002; Yagoub and Kolan, 2006; Kaliraj et al., 2014). The landuse and land cover in the coastal areas are extre- mely occurring substantial transformation due to waves, wind, tide, saltwater, saltation, sea level rise, storm surge, cyclones and human interference activities (Weismiller and Momin, 1977;

Meyer and Turner, 1994; Nemani and Running, 1995; Muttitanon and Tripathi, 2005; Li et al., 2010; Mani Murali and Dinesh Kumar, 2015). The pace of population along the Indian coastal tracts is increasing from 11.42% in 1991 to 14.20% in 2014. It is directly producing negative impacts on land resources and causes rapid land covers transformation into built-ups, settlements and other recreational uses (Coppin et al., 2004; Bhatta et al., 2010).

Therefore, the Indian Government has launched various landuse and land cover change assessment projects throughout the nation in order to monitor and management the natural and environmen- tal resources (ICMAM, 2002). Moreover, the same phenomena is undergoing in the South West coast of Kanyakumari district of Tamil Nadu, where the coastal landforms are severely altering by the impact of landuse and land cover changes on the coastal and marine processes (Chandrasekar et al., 2000; Choudhary et al., 2013). Over the last decade, the landuse and land cover changes in the coastal area are adapted beachface landforms, sand dunes, agriculture and vegetative cover due to development of settle- ments and other infrastructures. The natural factors like waves, storm surge, cyclone, erosion and subsidence, etc., are gradually altering the landuse and land covers along the coastal regions (Nayak, 2002; Short and Trembanis, 2004; Zhang, 2011; Kaliraj and Chandrasekar, 2012; Mujabar and Chandrasekar, 2013).

Furthermore, the vast area of the coastal plantations and woody vegetations are damaged during the year 2004 due to the effect of Tsunami occurred on 26th December 2004 (Chandrasekar, 2005;

Hentry et al., 2010). Increasing of population and climatic variabil- ity produces pressure on the landuse and land cover (LULC) causing the greatest environmental impact on vegetative cover, shoreline change, landform degradation, loss of biodiversity, seawater intru- sion and groundwater pollution, deterioration of soils and air along the coastal regions (Chandrasekar et al., 2001; Chauhan and Nayak, 2005; INCOIS, 2009; Mahapatra et al., 2013; Kaliraj et al., 2014).

Analysis of LULC change is necessary for both scientific and policy actions on coastal vulnerability assessment (Brown et al., 1999;

Benoit and Lambin, 2000; Ayad, 2004; Baby, 2015). The LULC change analysis provides a primary information for sustainable coastal zone management, therefore, it is very essential for the obtaining the reliable information in order to support proper

coastal resource management planning in national or regional scale under immense population pressure (Weismiller and Momin, 1977; Wu et al., 2002; Zoran, 2006; Zhang and Zhu, 2011; Butt et al., 2015a). The integrated remote sensing and GIS techniques provide a successful platform for mapping the LULC features as per user requirement (Jaiswal et al., 1999;

Chandrasekar et al., 2000; Yagoub and Kolan, 2006; Kawakubo et al., 2011; Misra et al., 2013; Rawat and Kumar, 2015). In the con- ventional methods, the mapping of LULC features is performed using available records, maps and field survey. It is often time- consuming process, exhaustive and expensive and the output maps soon become outdated with the passage of time particularly in the rapidly changing environment (Anderson et al., 1976; Wickware and Howarth, 1981; Singh, 1989; Nemani and Running, 1995;

Nayak, 2002; Wang et al., 2004, 2008; Rawat et al., 2013). Satellite image provides a synoptic coverage of the earth surface in spatial and temporal scale help us to understand how the changes hap- pened in various parts of the environment including coastal waters (Misra et al., 2013). Integrated GIS and remote sensing technique deals with the spatio-temporal information of LULC features and well recognized for decision making in the scientific realm (Rawat et al., 2013; Misra and Balaji, 2015). Kaliraj and Chandrasekar (2012)have stated that the GIS and remote sensing combines the multiple spatial datasets such as maps, aerial pho- tographs, and satellite images for preparing the quantitative, qual- itative and descriptive geo-databases for periodical changes of the LULC features. Many researchers have stated that the remote sens- ing techniques for LULC change detection analysis in local, regional and global scales (Wickware and Howarth, 1981; Avery and Berlin, 1992; Jaiswal et al., 1999; Chandrasekar et al., 2000; Alam et al., 2002; Jayappa et al., 2006; Santhiya et al., 2010; Mujabar and Chandrasekar, 2012). Landsat images namely MSS, TM and ETM+

are used worldwide for LULC change detection analysis (Baby, 2015). The advantage of Landsat images with adequate spectral properties provide better information on LULC changes compared to point data collected by on-site instruments during in-situ survey (USGS, 2004; Muttitanon and Tripathi, 2005; Kawakubo et al., 2011). Many researchers have stated the use of Landsat TM and ETM+ images LULC change analysis (Toll, 1985; Vogelmann et al., 1998; Dwivedi et al., 2005; Akbari et al., 2006; Dewidar and Frihy, 2010; Hereher, 2011).

Recently developed algorithms and techniques and software provide relatively accurate information on change detection and transformations of LULC features in the area of interest (Richards and Jia, 2006; Amin and Fazal, 2012; Mohammady et al., 2015).

Algorithm based image classification techniques like SVM and ANN and ABC provides obtained satisfactory result while using more samples, but misclassified or overlapped the pixels while using minimum training samples at 4th order polynomial function (Jayanth et al., 2015).Weller et al. (2006)have pointed out the ANN algorithm mainly suitable for dinoflagellate cyst images compiled with self-organizing map clustering algorithm. However, the MLC classifier is used per-pixel signature files of training samples col- lected from the real ground and used as knowledge engine for clus- tering the maximum likelihood digital numbers (DN) of pixels in the image (Afify 2011; El-Asmar et al., 2013; Iqbal and Khan, 2014; Butt et al., 2015b). Using Maximum Likelihood Classifier (MLC) algorithm in supervised classification technique, the Landsat image provides relatively accurate result in change detection assessment (Anderson et al., 1976; Jensen, 1996; Di Gregorio and Jansen, 2000; Gibson and Power, 2000; Foody, 2002; Coppin et al., 2004; Lu et al., 2004; Joshi et al., 2011). The present study delivers the basic pre-requisite information for understanding the trends in landuse and land cover changes and transformation in the coastal area.

2. Study area

In the present study, the LULC features changes and transforma- tions in the south west coast of Kanyakumari district, Tamil Nadu, India is performed using remote sensing and GIS technology. The geographical coordinates of the study area extend from 77° 9⁰ 32.71⁰⁰ E to 77°35⁰18.67⁰⁰ E longitude and 8°4⁰37.01⁰⁰N to 8°16⁰ 44.83⁰⁰ N latitude covering the total area of 292 km² with the length of 58 km along the shoreline (Fig. 1). The topographical structure of this area is gently sloped southward with the altitude range of 1.5–138 m due to presence of the Western Ghats in the northern side (Kaliraj et al., 2013). The coastal area consists of well-developed drainage systems including rivers, streams and water bodies, in which, the major rivers namely Thamirabarani, Valliyar and Pazhayar are flowing southerly direction from the Western Ghats and forming various coastal and alluvial landforms at the mouth point (CGWB, 2008). Geologically, the coastal setting is composed of crystalline rocks of Late Quaternary deposits of clay sand and sandy materials covered by thick laterite soil dotted with a few rocky outcrops. Teri sand dunes (reddish brown) are located along the coastal stretch from Kovalam to Manakudi; the thickness increases from 1.5 m in the coastal headlands to a maximum of 7.0 m in the interior parts of the area. Rocky shoreline extends along the sea face of coastal uplands and the patches of alluvium intervening crystalline outcrops of charnockite, garnet and gar- net–biotite gneiss are noticed in the river course (Perumal et al., 2010). The geomorphic features are strongly influencing the spatial distribution of LULC feature along the coastal area. The dune vege-

tative cover and coconut plantations are mainly distributed along the nearshore areas and coastal plains, whereas, the cultivable lands are predominantly noticed in the alluvial plains and flood plains in the river course. In the study area, the nearshore features such as sandy beaches, dune complexes, vegetations, wetlands, and shallow marshes are frequently regulated by physical, climatic and socioeconomic factors (Hentry et al., 2010). The study area is pre- vailing the sub-tropical climatic conditions with the annual aver- age rainfall ranging from 826 mm to 1456 mm and the temperatures ranging from 23.78°C to 33.95°C. Demographically, the study area covers both urban and rural settlements with dense population of 17.2% of the total population of the district and the increasing of population causes rapid changes in LULC features of the coastal region.

3. Materials and method

3.1. Landuse and land cover feature extraction

The present study is carried out for assessment of the decadal changes and transformations in LULC features in the South West coast of Kanyakumari district using remote sensing and GIS tech- nology. The Landsat ETM+ images (30 m) covering the study area are obtained from the Global Land Cover Facility (GLCF) (http://

glcfapp.glcf.umd.edu:8080/esdi/) and Earth Explorer (http://

earthexplorer.usgs.gov/) online portal for the year 2000 and 2011. Fig. 2shows the functional flow of methodology used for LULC change analysis. The study area covers a single scene of Land-

Fig. 1.Geographical location of the study area.

sat image with path 143 and row 054 under the WRS (world refer- ence system) of reference datum. Landsat ETM+ image with the scene id – LE71430542000101SGS00 acquired on April 10, 2000

and the Landsat TM with scene id – LT51430542011043BKT00 acquired on February 02, 2011 are used for the landuse and land cover mapping and change detection during the post monsoon per- Fig. 2.Functional flow of methodology used for LULC change detection and transformation analysis.

iod. The producer makes available pre-processed images in geotiff format to the user with the help of producer’s own-designed LPGS_12.5.0 software. It is important to notice that the scene id – LT51430542011043BKT00 acquired on February 02, 2011 is SLC corrected image with less than 10% of cloud density replaced instead of SLC-off image and recently available for user download- able in the USGS-Earth Explorer online portal (http://earthex- plorer.usgs.gov). According to USGS (2004) report on LULC classification, the combination of bands in Landsat TM and ETM+

images are effectively suitable for extraction of various LULC fea- tures, especially from the coastal area. For example, the band com- bination of band 1, band 4 and band 5 are distinctly produced features like water bodies, settlements, and salt marsh in the coastal area. Similarly, the combined image of band 3 and band 4 are classified for cultivable lands, forest, scrub and other natural vegetative cover (Chen et al., 2003). In this analysis, the Landsat ETM+ and TM images with the first five bands such as band 1 (0.45–0.52

l

m), band 2 (0.52–0.60

l

m), band 3 (0.63–0.69

l

^m),

band 4 (0.76–0.90

l

m) and band 5 (1.55–1.75

l

m) are combined into multispectral image for extraction of landuse and land cover feature for the periods of 2000 and 2011.

The combined multispectral images of the year 2000 and 2011 are geometrically corrected using UTM-WGS84 projection and coordinate system with the help of GCPs (Ground Control Points) collected from Survey of India (SOI) published topographical maps (scale 1:25000) and Garmin Etrex-10 handheld GPS field surveyed data. Furthermore, the images are atmospherically corrected to obtain the perfect spectral reflectance values using FLASSH (Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes) model in ENVI 4.5 software (Yuan et al., 2005). Recently several image classification techniques and algorithms have been developed extensively for the landuse and land cover analysis throughout the world. However, each technique is having developed and reviewed for their advantages and disadvantages based on their algorithm, iteration, and processing software. For example, the algorithm based classification methods like support vector machine (SVM), artificial neural network (ANN), Maximum Likeli- hood Classifier (MLC), fuzzy analysis, segmentation and clustering, etc. Among them, the MLC technique depends on a combination of ground samples and personal experience with the study area and is strictly used the field observed training samples of real ground sur- face (Foody et al., 1995; Paola and Schowengerdt 1997; Huang et al., 2002; Foody and Mathur 2004; Yuan et al., 2009; Jayanth et al., 2015). The algorithm based classifiers like SVM, ANN, fuzzy analysis and segmentation are critically analysis multispectral images at sub-pixel level, but the result comprises complexity and bias while classifying medium resolution images acquired especially in the coastal area (Yuan et al., 2009). Jayanth et al.

(2015)have performed image classification using SVM and ANN and ABC algorithm on panchromatic fused LISS-IV data of 2.5 m spatial resolution and they have obtained satisfactory result out of the 3090 total samples. Moreover, they have stated that the SVM and ANN classifier algorithms produces misclassified or over- lapped features while using minimum training samples at 4th order polynomial function. Subsequently, the accuracy assessment statistics of kappa coefficient for these algorithms are estimated in the descending order like SVM (0.765) and ANN (0.645).Weller et al. (2006)have stated that the ANN algorithm is a potential clas- sification tool that mainly used for dinoflagellate cyst images.

Though, number of advanced techniques available for LULC change analysis, the MLC classifier is used per-pixel signature files of train- ing samples collected from the real ground using GPS based Spec- trometer survey and used as knowledge engine for clustering the maximum likelihood digital numbers (DN) of pixels in the image (Foody et al., 1997; Afify 2011; El-Asmar et al., 2013; Iqbal and

Khan, 2014; Butt et al., 2015a). Based on the view of above con- cepts, the MLC classification technique is selected for analysis of medium resolution image to extract the LULC feature extraction in the study area. Moreover, the primary objective of this study is to estimate the decadal changes and their transformations of LULC features under the USGS-LULC Level II Classification System from the Landsat ETM+ and TM image for the periods of 2000 and 2011 using MLC classifier algorithm. However, many research- ers have used MLC classifier of supervised classification technique for LULC feature extraction from the Landsat TM and ETM+ image and achieved relatively high accurate results up to 95% (Butt et al., 2015b). For example,Sabins (1987), Rawat et al. (2013), Rawat and Kumar (2015) and Hazarika et al. (2015) and other researchers have employed the MLC classifier to analysis ETM+ images and achieved highly satisfactory results in land use/land cover change studies (Aspinall and Hill, 1997; Gibson and Power, 2000; Foody, 2002; Bruzzone et al., 2004; Joshi et al., 2011). The MLC algorithm is derived from the Bayes theorem, which states that the a posteri- ori distributionP(i|x), i.e., the probability that a pixel with feature vector x ^{belongs to class i, and it is expressed as} Pðij

x

Þ ¼Pð

x

jiÞPðiÞ=Pð

x

Þ, where P(x|i) is the likelihood function, P(i) is the a priori information, i.e., the probability that class i occurs in the study area and P(x) is the probability that x ^is observed (Ahmad 2012). In this analysis, the MLC algorithm is exe- cuted on Landsat ETM+ and TM images acquired on 2000 and 2011 for extracting the LULC features using ERDAS Imagine 2014 soft- ware. The MLC algorithm is used the spectral signature compo- nents of the 120 training samples (10 samples from each feature class) collected from the GPS based Spectrometer (HandHeld 2 hand-held spectroradiometer, ASD Inc) survey on the real ground surface. Using the surveyed data, the training sets are prepared from the clustered pixels on the images at pixel ratio of 1:10.

Before starting classification, it is significantly executed the feature space technique on training sets for correcting the misclassified or overlapped pixels and it is recoded subject to the corresponding feature class (Verbyla and Hammond, 1995; Kaliraj and Chandrasekar, 2012). Availability of satisfactory spectral signature of the sample class is ensuring the minimal confusion in clustering of pixels during feature extraction processes (Baraldi et al., 2005;

Carlotto, 2009; Butt et al., 2015a). Mathematically, the MLC algo- rithm is performed clustering of maximum likelihood pixels belongs to a particular class repeatedly till end of iteration based on the multivariate normal distribution values of pixels (represen- tative pixels) in the training sets using probability density function (Jensen, 1996; Gibson and Power, 2000; Lu et al., 2004; Onur et al., 2009; El Asmar and Hereher, 2011). Spectral signatures for the respective land cover types derived from the satellite imagery were recorded by using the pixels enclosed by these polygons. A satis- factory spectral signature is the one ensuring that there is ‘minimal confusion’ among the land covers to be mapped (El Gammal et al., 2010). The MLC algorithm produces clustered signature of training sets as a knowledge engineer used to classify the pixels of images based on the maximum probability of belonging to a particular class (Richards and Jia, 2006). Subsequently, the clustered signa- tures of training sets are iterating to cluster the maximum likeli- hood pixels distributed over an image and grouped into separate cluster by the mean vector and the covariance matrices. It is per- formed by the spectral distance method that calculates spectral distance between the measurement vector in the representative pixels and the mean vector in each signature for classifying the multiple classes based on the Euclidean distance (Atkinson et al., 1997). The clustered image consists of different feature classes is introduced to nearest neighbour resembling (kernel window size 33) analysis for replacing the disintegrated pixels to the most common pixel of feature class. The result of classified images is

categorized into twelve feature classes based on the USGS-LULC Classification System at Level II category. The Level II system is probably comprises major LULC features for compilation and map- ping at regional scale. Therefore, among the four level (Level I – IV) category of this system, the Level II classification system frame- work is considered more suitable for categorizing the LULC fea- tures from the classified image. In this Level II system, the clear definition of LULC features helps to differentiate the many classes belongs to common LULC family and is intentionally provides attri- butes of the LULC features (Rawat et al. 2013). For example, based on distinct LULC feature definition, the agriculture land (Level I) is classified to cropland and plantation area in the study area using Level II system. The result of LULC map (1:10000) is cross- verified with the state level landuse and land covermap (scale 1:

50000) for 2011–2012 published by Bhuvan (Indian Geoportal website) and the result shows the spatial distribution of LULC fea- tures are relatively same as the referenced map, however statistical measurement of both maps are not cross-verified in this case.

Finally, the LULC layers for the two years 2000 and 2011 are sepa- rately prepared for geo-database with the attributes include class name, location, areal extent and perimeter using ArcGIS 10.2 soft- ware. Furthermore, these classified images are involved for accu- racy assessment using post field verification technique.

3.2. Landuse and land cover change detection analysis

Change detection assessment of LULC features provides infor- mation sources for monitoring the trend of changes of an area over time. In this study, the change detection analysis is performed on the classified images for the year 2000 and 2011. Based on the review of many researchers, the post-classification comparison method is adopted for change detection analysis because of this strictly performs pixel-to-pixel comparison of classified images of the area acquired on two different times (Lo and Shipman, 1990;

Kressler and Steinnocher, 1996; Abuelgasim et al. 1999; Petit and Lambin, 2001; Coppin et al., 2004; Sylla et al.,2012; Rawat et al., 2013; Hazarika et al., 2015). In this study, the post-classification comparison technique encompasses the change detection matrix is used to estimate the changes in landuse and land cover features using ENVI 4.5 software. Change detection matrix executes geo- metrically referenced pixel (xi1, yj1) of a particular feature class in classified image of 2000 and the corresponding pixels (xi2, yj2) of the feature class in classified image of 2011 to identify the differ- ences in per pixel units within a particular feature class as well as the group of pixels of the same feature classes. The comparative analysis of feature classes allows estimating differences in area and perimeter of the landuse classes (Jensen, 1996; Skelsey et al., 2003;

Richards and Jia, 2006). In which, the matrix is arranged with a dimension of 1212 rows and columns to calculate the differ- ences by subtracting pixel (in rows (fr)) of the classified image in 2000 from the pixel (in column (fc)) of classified image in 2011.

The change detection statistics used to compile a matrix tabulation is expressed as (final state (fr) – initial state (fc)/initial state (fc)).

This statistics produce the pixel variations in a particular class cor- responding to both images indicate the changes in the state of an object and phenomena over time (Atkinson et al., 1997; Lu et al., 2004; Joshi et al., 2011). In addition to that, the magnitude of change (K) is calculated for estimating the degree of expansion or reduction of area (in percentile) in a particular feature class and the equation is expressed asK= (fi/i)100. Where, the K refers magnitude of change (percentile) f is classified image of ini- tial time and i reference data (classified image of recent time). In which, the outcome of negative value refers loss of area of LULC feature, while the positive value indicates gain of area of LULC fea- ture. The vectorized layers of LULC features are intersected to determine the rate of transition of a particular LULC feature to

one or another more states (Verbyla and Hammond, 1995;

Baraldi et al., 2005; El Gammal et al., 2010; Amin and Fazal, 2012). It is performed by assigning the values of differences in cross-tabulation assigned feature classes in rows and columns to calculate transition rate over time. Hence, the results of LULC change and transformation are used as primary information source for understanding the trends in landuse and land cover changes along the coastal region.

4. Results and discussion

The South West coast of Kanyakumari district in Tamil Nadu, India is reported to be facing serious environmental challenges in the form of shoreline erosion, loss of landforms, decreasing of sed- iment load, seawater intrusion and aquifer contamination due to rapid changes in landuse and land cover induced by climatic changes, sea level rise, population growth and other human encroachment activities (ICMAM, 2002; Hentry et al., 2010).

Assessment of LULC changes and transformation in the study area is extracted from the Landsat ETM+ images (30 m) for the period of 2000 and 2011 using the Maximum Likelihood Classifier (MLC) algorithm of supervised image classification technique in ERDAS Imagine software. Thus, the result describes the different LULC fea- ture classes based on the Level-II category of USGS-LULC classifica- tion table with overall classification accuracies of 81.16% and 77.52% and overall Kappa coeffient statistical values of 0.83 and 0.76 respectively. The information of LULC changes and transfor- mation is an essential source for coastal vulnerability assessment and coastal resources management in the coastal area.

4.1. Assessment of landuse and land cover change between 2000 and 2011

Fig. 3A and B shows the spatial distribution of the major lan- duse and land cover classes in the study area for the year 2000 and 2011. The areal extent of these LULC features in square kilome- ter and area coverage in percent for the two different periods is shown inTable 1; and its corresponding values of graphical repre- sentation is shown inFig. 4A and B. The different types of the LULC features such as beachface landforms (including beaches, sand dunes and associated other nearshore features), dune vegetation, plantation, cultivable land, fallow land, barren land, settlement (or) built-ups, river, freshwater bodies, saltwater bodies, saltpan and beach mining are experienced significant changes during past decade in the various parts of the coastal area.

4.1.1. Beachface land cover

Beachface land cover comprises sandy beaches, embryo dunes, foredunes, sand dunes and associated landforms within the near- shore area. The features of this class are experienced with high rate of changes over the time due to the marine and coastal processes and other anthropogenic activities. The total area of this class is estimated as 6.38 km²in 2000 and 4.62 km²in 2011; it is equal to 2.18% and 1.58% of the total study area respectively (Fig. 4).

The different features of beachface land cover are significantly reduced in the various parts and it is evidently noticed negative changes at the rate of 1.76 km²during 2000–2011 (Fig. 5). A high rate of change is observed in the Kanyakumari, Kovalam, Pallam, Periyakadu, Pillaithoppu, Muttam, Manavalakurichi, Mandaikadu, Midalam, and Inayamputhenthurai coastal areas. The sandy bea- ches in the Kovalam, Pallam, Periyakadu, Pillaithoppu, and Inayam- puthenthurai coastal zones are significantly decreased on the down-drift side of the groins and the seaward side of settlements.

Choudhary et al. (2013)andMisra and Balaji (2015)have stated that the coastal structures along the eastern coast of India mainly

Fig. 3.Spatial distribution of Landuse and land cover in the study area; a) LULC status in 2000, b) LULC status in 2011 (source: Landsat ETM+ images).

disturb the shoaling, refraction and diffraction of wave processes and cause accretion on up-drift side and erosion on the down- drift side of the coast. The same phenomena induce severe erosion on the shoreline on the down-drift side of the study area causes altering the morphological structure of landforms (Kaliraj et al., 2012). In the rocky coast of Kanyakumari, Kovalam, Muttam, and Colachel, the cliff erosional activities are reduced the area in the headlands due to high-energy wave actions. the decreasing of sed- iment flux causes shoreline erosion in the middle-western parts due to the construction of coastal structures that intervening the river discharges and littoral sediments (Cherian et al., 2012;

Mujabar and Chandrasekar, 2013). However, in the high-energy wave dominated coastal zones, the beachface landforms are evi- dently decreased in areal extent due to the cumulative effect of waves and currents.

4.1.2. Dune vegetative cover

The foredunes and sand dunes comprise various types of vege- tative covers including shrubs, bushes, grasses, and woody species like casuarinas, coconuts and others species in the coastal area.

Dune vegetative cover can protect dunes and beaches from ero- sional processes of wind, waves and storm surge and recruiting of sand to the eroded beaches (Nayak, 2002; Yagoub and Kolan, 2006; Santhiya et al., 2010; Hereher, 2011; Misra et al., 2013).

The dune vegetative cover is reduced in various parts of the study area from 4.32 km²to 3.36 km²(Fig. 4) during 2000–2011 with a total loss in area as 0.96 km²(Fig. 5). Removal of dune vegetative cover along the foredunes could be due to the Tsunami waves occurred on 26th, December 2004 (Chandrasekar, 2005; Hentry et al., 2010). The embryo dunes of grasses and bushes are disap- peared in the different parts in the Chothavilai, Sanguthurai, Puthenthurai and Rajakkamangalamthurai coastal areas due to wave erosional processes, wind induced saltation and human encroachment activities. Meanwhile, the casuarinas tree covers known as woody fences are deforested or uprooted along the fore- dunes and backshore dunes in the Rajakkamangalamthurai, Pal- lam, Manavalakurichi, Mandaikadu and Midalam coastal areas due to placer mining and expansion of settlements and built-ups.

Decreasing of dune vegetative cover in the various parts causes coastal instability and increases rate of erosion vulnerability to the coastal region.

4.1.3. Plantation and woody vegetative cover

The plantation along the coastal area refers to garden consisting of a cultivated woody species without undergrowth like shrubs, bushes and other natural vegetative covers, while the natural woody vegetative cover refers to naturally distributed woody spe-

cies in the coastal plains, sand dunes and barren lands (Brown et al., 1999; Small and Nicholls, 2003; Zhao et al., 2006). The coastal plantations are one of the most dominant landuse covering coconut trees, orchards, and other woody species. The plantation along the coastal area plays an important role to reduce the dam- ages and protect coastal ecosystem, human lives by acting as pro- tective shield during extreme natural events (Yagoub and Kolan, 2006; Misra and Balaji, 2015). The spatial extend of this landuse is estimated as 97.07 km² (33.24%) in 2000 and decreased to 94.15 km² (32.24%) in 2011 with a loss of 2.29 km² (Figs. 4 and 5). Plantations and woody vegetative cover across the different places of the coastal belt were uprooted and damaged by the Tsu- nami calamity occurred on 26th December 2004 (Chandrasekar, 2005; Kaliraj et al., 2012). However, reducing the areal extent in coastal vegetative causes beach subsidence in the coastal area.

4.1.4. Cultivable land

The cultivable lands are predominantly distributed along the riverbanks of alluvial plain, flood plain and coastal plain in the var- ious parts of the study area. It is primarily used for cultivating paddy, banana, flower, sugarcane and vegetables to sustain human life. The areal extent of this landuse shows a significant decreasing trend from 101.37 km² (34.72%) to 96.87 km² (33.17%) during 2000–2011 (Figs. 4 and 5). During these periods, there is a reduc- tion of area of 4.50 km²in cultivable land due to land degradation, urbanization, encroachment for settlement and real estate activi- ties. Due to land degradation and soil infertility, the cultivable lands are vastly converted into fallow and become barren land in the coastal region (Yang et al., 2015). In the Kanyakumari, Kova- lam, South Thamaraikulam, Pallam, Rajakkamangalamthurai, Mut- tam, Colachel, Midalam, Inayamputhenthurai and Thengapattinam coastal zones, the marginal area of cultivable landuse are con- verted into settlements causes increasing of pressure on the coastal aquifer causes groundwater contamination and seawater intrusion into the inland aquifers.

4.1.5. Fallow land

Fallow lands are mainly associated with agricultural lands which temporarily uncultivated for one or more season but not less than one year or unable to sustain a crop beyond planting and emergence for one or more production seasons (Ruthenberg et al., 1980; Chaurasia et al., 1996; Chauhan and Nayak, 2005;

Haque et al., 2008). This is due to the effect of land salinity, chem- ically affected soil fertility, land degradation, and runoff, etc (Li et al., 2010). The total area of this land is estimated as 47.97 km² (16.43%) in 2000 and it is reduced to 34.28 km²(11.74%) in 2011 (Fig. 4). During these periods, the area of 13.69 km²of fallow land is tracked for various purposes like built-ups, aquaculture, saltpan Table 1

Area and amount of change in different coastal land use and land cover features in the study area during 2000–2011.

Sl.

No.

Landuse and Land cover feature

2001 2011 Area in changes during 2001–

2011^*(km²)

% of area in changes during 2001–

2011^* Area

(km²)

% of distribution

Area (km²)

% of distribution

1 Beachface land cover 6.38 2.18 4.62 1.58 1.76 0.60

2 Dune vegetation 4.32 1.48 3.36 1.15 0.96 0.33

3 Plantation 97.07 33.24 94.15 32.24 2.92 1.00

4 Cultivable land 101.37 34.72 96.87 33.17 4.50 1.54

5 Fallow land 47.97 16.43 34.28 11.74 13.69 4.69

6 Settlement/Built-ups 17.54 6.01 39.22 13.43 21.68 7.42

7 Barren land 5.06 1.73 5.59 1.91 0.53 0.18

8 River 1.92 0.66 1.28 0.44 0.64 0.22

9 Water bodies 3.3 1.13 3.12 1.07 0.18 0.06

10 Saltwater bodies 3.43 1.17 3.20 1.10 0.23 0.08

11 Saltpan 3.21 1.10 5.01 1.72 1.80 0.62

12 Beach mining 0.43 0.15 1.30 0.45 0.87 0.30

*Negative value refers to decrease of area and positive value refers to increase of area in a LULC features.

and other human needs (Figs. 5 and 6). In the Pazhayar estuary, the larger area of fallow lands in the Kanyakumari, South Thamaraiku- lam, Manakudi and Pallam coastal areas are converted into saltpan, marshy lands and settlements, whereas, in the western parts of the Colachel and Thengapattinam coastal areas, the fallow lands are rapidly transformed to built-ups and other infrastructures.

4.1.6. Barren land cover

Barren land cover is the second dominantly increasing land cover in the study area. This is mainly associated with stony waste and bare soil with or without shrubs, scrubs and thorny bushes.

The spatial coverage of barren land in 2000 is 5.06 km²that corre- sponds to 1.73%, it is raised to 5.59 km² in 2011, and covering Fig. 4.Percentage of area in LULC features; a) % of distribution in 2000, b) % of distribution in 2011.

1.91% of the total study area (Figs. 4 and 5). The Terisand dunes of Kanyakumari-Kovalam coastal tracks are noticed a larger area of barren land cover since long periods, but it is recently occupied

for resorts and built-ups (Chandrasekar et al., 2013). However, a large scale of fallow lands are changed to barren land cover in the marginal areas of Kanyakumari, South Thamaraikulam, Pallam, Fig. 5.Net changes in area of LULC feature classes during 2000–2011.

Fig. 6.Net loss and Net gain of area in LULC feature classes during 2000–2011.

Murungavilai, Muttam, Colachel, Kurumpanai, and Mel Midalam coastal settlements. This is due to result of different factors such as poor rainfall, land salinity, soil infertility, land degradation, shifted cultivation system, inaccessible and logging activities (Hentry et al., 2010). The increasing trend of barren land is clearly attributed to decreasing of soil fertility, groundwater availability and agricultural productivity and soil fertility along the study area.

4.1.7. River and water bodies

The drainage system constitutes rivers, tributaries, and water bodies in various parts of the study area. There are three major riv- ers namely Thamirabarani, Valliyar and Pazhayar along with their tributaries are flowing southerly from the Western Ghats. The spa- tial coverage of rivers and streams are denoted as 1.92 km²in 2000 and it is reduced to 1.28 km²in 2011 (Fig. 5). In the western part, the Thamirabarani river course is slightly occupied for built-ups and agriculture activities, whereas the Pazhayar riverbanks are sig- nificantly encroached for saltpan and aquaculture purposes. The Valliyar riverbanks in the middle-western parts are disturbed by placer mining and urbanization activities (Kaliraj and Chandrasekar, 2012). Moreover, the water bodies throughout the study area are reduced in area from 3.30 km²to 3.12 km²during 2000–2011 due to human encroachment for built-ups and agricul- ture activities (Figs. 5 and 6). A reduction of area in water bodies reduces water storage capacity and groundwater level that induces the cultivable land into fallow land, and long-term fallowness leads to increases of barren land cover in the study area (Hentry et al., 2010). Moreover, due to human encroachment activities, the water bodies are decreased in capacity of storage water causes over exploitation of groundwater during summer that directly increases pressure on seawater intrusion to the coastal freshwater aquifers (Kaliraj et al., 2014).

4.1.8. Saltpan and saltwater bodies

Saltpans and saltwater bodies are mainly distributed in and around the estuaries and backwater in the study area. Saltpans are extensively increased in areal extent from 3.21 km² to 5.01 km²in the Pazhayar estuary near Manakudi and Thamaraiku- lam settlements and in the backwater of the Kovalam and Colachel coastal areas. The saltwater bodies along the backwater and estu- aries are encroached for urban or settlement expansion activities and forming the saltpan and hence it is decreased in area from 3.43 km² to 3.20 km². Meanwhile, a few freshwater pools in the backshore parts of Kovalam, Rajakkamangalamthurai, Murungavi- lai, Manavalakurichi, Kodimunai, and Thengapattinam coastal areas are changed into saltwater bodies due to increase of salinity by the result of seawater intrusion (Hentry et al., 2010). In which, the Manavalakurichi-Mandaikadu coastal track is found a few number of pits with saltwater due to the result of high tide waves and storm surges that stored seawater on the abandoned placer mining quarries (Kaliraj et al., 2014). It is noticed that the increase of area in saltwater bodies and saltpan causes groundwater con- tamination through free movement of seawater to the freshwater aquifers in the coastal as well as in the inland regions.

4.1.9. Beach mining and quarries

In the study area, the beach mining pits are noticed in the Manavalakurichi-Mandaikadu coastal stretches with the areal extent of 0.43 km²in 2000 and 1.30 km²in 2011 due to removal of sand from beaches and dunes for placer mining activities (Figs. 5 and 6). A few abandoned placer mining pits are mainly found along the Manavalakurichi-Mandaikadu coastal track, in which, the salt- water is being logged during high tide waves and storm surges.

Removal of sand due to placer mining is a pathway for intruding seawater to the inland aquifers (Kaliraj et al., 2014). Besides, some parts of granite quarries are noticed near Inayamputhenthurai and

Thengapattinam coastal settlements and a few abandoned mining pits are noticed in the marginal areas of the coastal upland in Kova- lam, Muttam, Colachel and Inayamputhenthurai due to removal of Terisand materials for construction of built-ups and infrastructures.

4.1.10. Settlements and Built-ups

Settlements and built-ups refers to the land areas of human habitation covered by buildings, road networks, rails, infrastruc- tures and other utilities in association with livelihood activities (Adeel, 2010). The SW coast is one of the important tourist spots in the Tamil Nadu state. The tourist places such as Kanyakumari, Kovalam, Chothavilai, Sanguthurai, Muttam, and Colachel coastal areas are experienced rapid growth of population during 2000–

2011. The settlement areas are directly proportional to the growth of population in these areas and it increases predominantly due to better access of daily human needs (Alphan, 2003; Ohri and Poonam, 2012). The aerial extent of settlements and built-ups is 17.54 km² in 2000 and it covers 6.01% of the total study area (Fig. 4). Due to extending of settlements in rural as well as urban areas, this is increased to 39.22 km²corresponding to 13.43% of the total area in 2011 (Table 1). This highlight the 7.42% of the area in settlements and built-ups is developed during the past decade on beachface landforms, cultivable land, fallow land and barren land by encroachment of natural land cover features (Figs. 4 and 5). In the eastern parts, the sand dune complexes and coastal plains are converted into built-ups, factories and resorts; whereas, in the middle-western parts near Midalam, Colachel and Muttam coastal areas, the fertile agricultural land, fallow land and few parcels of barren lands are utilized for the expansion of settlement areas. It is noticed that the encroachment of coastal land cover for settle- ments and built-ups in a large scale produce negative impact on coastal ecosystem and increases rate of vulnerability to the coastal region.

4.2. Landuse and land cover transformation between 2000 and 2011

Landuse and land cover transformation refers to the process of change dynamics in land features from one form to another during periodical times due to natural or human induced transformation activities (Richards, 1990; Amin and Fazal, 2012). Land features are experiencing continuous state of transformation from their existing form into another form depending on local environmental suitability and local people requirements (Lambin et al., 2003;

Kumar and Jayappa, 2009; Leh et al., 2013). Table 2 shows the result of change detection matrix of LULC transformations, which denotes the amount of area in LULC features transferred from pre-existing state to another one or more states in the study area during 2000–2011. Fig. 7 shows the graphical representation of LULC feature transformation over time. During these periods, the total area of 45.90 km²in various landuse and land cover are trans- formed or interchanged into another one or more forms. In which, the major LULC features such as cultivable land (2.23 km²), fallow land (15.32 km²), barren land (2.14 km²) plantations and other woody vegetations (1.05 km²) are presently occupied for settle- ments and built-ups in large scale and hence the spatial coverage of settlement and built-ups is predominantly increased from 17.54 km²to 39.22 km²during 2000–2011 (Figs. 5 and 6). In the Kanyakumari, Kovalam, Chothavilai, Sanguthurai, Muttam coastal zones, the area of 1.24 km²in beachface land cover include sandy beaches, dunes, plains and uplands are largely encroached for built-ups and other infrastructures due to rapid growth of popula- tions. In these areas, it is observed that the expansion of area in set- tlements and built-ups are directly proportional to the population growth that produces a severe vulnerability to the coastal

resources (ICMAM, 2002). Whereas, the some parts of this land cover (0.63 km²) along the Manavalakurichi-Mandaikadu coastal tracks are changed into mining pits with or without saltwater due to removal of sands from beaches and dunes for placer mining

activities. Similarly, the area of 0.21 km²in this land cover is trans- formed into saltwater bodies in the Kovalam, Rajakkaman- galamthurai, Murungavilai, Manavalakurichi and Thengapattinam coastal areas due to high-energy waves, tides and storm surges.

Table 2

The change detection matrix showing the area of Landuse and Land cover transformations in the study area during 2000 – 2011.

(Note: the value noted in shaded box (diagonally) refers to the total area of LULC feature in 2000; the column shows the amount of area in a LULC feature transferred into another land use/cover during 2000–2011; whereas, the row shows the amount of area in a LULC features captured (contributed) from one or more another land use/cover).

Fig. 7.Transformation of LULC feature class into one or more another states during 2000–2011.

The dune vegetative cover along the embryo dunes, foreshore dunes, and backshore dunes are severely damaged due to high- energy wave erosion, storm surge, wind induced saltation and other human encroachment activities (Kaliraj et al., 2012). In the middle-eastern parts, the area of 0.12 km²in dune vegetative cover is removed and the area becomes reformed with various beachface landforms. Whereas, the area of 0.42 km²in this land features are converted into settlements and built-ups in the Pallam, Puthenthu- rai, Murungavilai, Rajakkamangalamthurai and Midalam coastal areas. Meanwhile, the area of 0.42 km² in dune vegetative cover is removed due to placer mining activities in the Manavalakurichi-Mandaikadu coastal stretches, and the area of 0.32 km² is transformed into shallow depth saltwater bodies in various parts of the study area. In the study area, the spatial cover- age of plantations and other natural woody vegetative cover is transformed to different land uses such as built-ups, cultivable land, fallow and barren land cover in the nearshore, backshore and coastal plain areas (Hentry et al., 2010). In which, some parcels of coconut plantation in the coastal plains and alluvial plains are significantly interchanged to cultivable land (1.12 km²) and fallow land (1.31 km²) in the South Thamaraikulam, Murungavilai, Mida- lam, and Inayamputhenthurai coastal areas (Fig. 7). Whereas, the area of 1.05 km²in plantation is converted into built-ups and other infrastructures in the Pallam, Rajakkamangalamthurai, Periyakadu, Pillaithoppu, Colachel due to expansion of urban and rural settle- ments towards nearshore and backshore areas. The area of 0.21 km²in plantations and woody vegetative cover in the various parts are presently transformed into barren land cover because of land degradation due to increasing land salinity and it could be the result of Tsunami waves in 2004 (Chandrasekar et al., 2013).

In the study area, it is really depreciated that the area of 8.82 km² in fertile cultivable lands are transformed into other forms like barren, fallows, saltpan, and built-ups during 2000–

2011. During these periods, the cultivable land spread over the coastal plains and alluvial plains are converted into fallow land (4.35 km²) and barren land (1.05 km²) respectively, whereas, the area of 2.23 km²around the marginal parts of settlements are con- verted into built-ups and other infrastructures. However, some parcels of cultivable lands (0.42 km²) along the coastal belt are interchanged with plantations and other woody vegetative cover, which plays an important role to protect coastal landforms from Tsunami, storm surge and other natural hazards (Hentry et al., 2010; Chandrasekar et al., 2013).

In this study area, it is significantly noted that the one third of fallow lands (15.32 km²) are converted into settlements and built-ups, and it is directly reflects the rapid growth of population in the southern coastal belt. The fallow lands around the major coastal settlements namely Kanyakumari, Rajakkamangalamthu- rai, Muttam, and Colachel are utilized for built-ups and infrastruc- tures due to rapid urban expansion activities. Meanwhile, the fallow lands are increasing in area due to conversion of cultivable land and deforestation of plantation and woody vegetation in the various parts of the study area. Interestingly, the fallow lands in some parts of the river course are transformed into cultivable land (1.82 km²) and plantations (0.53 km²) in the middle-eastern and western parts of the study area. Unfortunately, the area of 1.55 km²in fallow land along the coastal belts is further degraded to barren land cover due to increase of salinity and soil erosion. A few parcel of fallow lands are converted into saltpan and saltwater bodies in the Pazhayar estuary near South Thamaraikulam and Manakudi coastal areas. It is increasing pressure on aquifer vulner- ability to groundwater contamination and seawater intrusion in the middle-east parts of the coast (Kaliraj et al., 2014). Similarly, the area of 2.14 km²in barren land cover is converted into built- ups, resorts and other infrastructures in the Kanyakumari, Kova- lam, Muttam, Colachel and Midalam coastal areas. Fortunately,

the area of 1.06 km²and 0.32 km²in this land cover is transformed to cultivable land and coconut plantation respectively in the middle-eastern and middle-western parts of the study area. How- ever, some parcels of this land cover are converted into saltpan in the Colachel and Manakudi coastal areas.

In the Manavalakurichi-Mandaikadu coastal tracks, the aban- doned placer mining pits (00.15 km²) are changed into saltwater pools due to entering of seawater during high tides and storm surges. Similarly, a few saltwater bodies around the backwater and estuaries near Kovalam, Manakudi, South Thamaraikulam and Colachel coastal areas are converted into saltpan (1.21 km²) and barren land cover (0.24 km²) due to anthropogenic activities.

Moreover, the low-laying areas along the estuaries are transformed to barren lands due to non-availability of tidal water or backwater (Chandramohan et al., 1993; Mujabar and Chandrasekar, 2012).

Ahmed (2011) has pointed out the increase of pressure on the coastal area due to natural and human activities leads dynamic changes in coastal landuse and land cover. In the study area, it is clearly observed that the various LULC features are dynamically transformed to different land uses due to natural and human activ- ities based on the local environmental suitability and requirement of the local people. However, most of the land cover features are transformed into built-ups and settlements without considering their negative impacts on the coastal systems and processes that results increasing of vulnerability to the coastal landforms.

4.3. Accuracy assessment

Accuracy assessment is an integral process of feature extraction from the classified images. It highlights the possible sources of errors in a classified image, thus enhancing the quality of informa- tion derived from the data (Lea and Curtis, 2010).Bradley (2009) has pointed out that no image classification is said to be complete unless its accuracy has been assessed. Therefore, it is critical to ensuring the accuracy of the information used in the decision pro- cess through statistical analysis based on the pre and post-field ground truth verification and available collateral information (Foody, 2010; Huang and Hsieh, 2012). The standard method is used to assess the accuracy of classified image is confusion matrix or simply known as error matrix (Story and Congalton, 1986;

Biging et al., 1998; Zhang et al., 2000; Foody, 2002; Onur et al., 2009; Mujabar and Chandrasekar, 2013). This method allows the comparison of pixels of classified image to the corresponding pix- els of referenced image for which the class is known (Smith et al., 2003; Richards and Jia, 2006; Szuster et al., 2011). The confusion matrix compiles pixels of agreement and disagreement by compar- ing the location and class of each ground truth pixel with the cor- responding location and class in the classification image. It is acxc matrix (cis the number of classes) assigned by columns and rows, the elements of which indicate the number of pixels in the testing data. In this matrix, the columns depict the number of pixels in a class of the reference data, and the rows show the number of pixels in a class of the classified image (SCGE, 2011). From this error matrix, the accuracy assessment is implemented using three differ- ent scales namely producer’s accuracy, user’s accuracy and overall accuracy based on the rule of error of commission and error of omission (Coppin and Bauer, 1996; Boschetti et al., 2004;

Carlotto, 2009). In which, the producer’s accuracy is defined as total number of correctly classified units (pixels) in a particular class (xy) divided by total number of class (xy) units (pixels) iden- tified in reference data (Bradley, 2009; Mohammady et al., 2015).

Whereas, user’s accuracy is defined as the correctly classified class (xy) divided by total number of units (pixels) classified as class (xy). The overall accuracy is the sum of all correctly classified units (pixels) divided by total number of units (pixels) of all classes (Congalton and Green, 1999; Lu and Weng, 2007; Li and Zhou,

2009). The Kappa coefficient is another measurement in image classification technique used to measure the extent of classifica- tion accuracy for diagonal elements as well as for all the elements.

In the confusion matrix, the pixels are calculated for coefficient values based on the difference between the actual agreement in the error matrix and the chance agreement in the total row and columns (Coppin and Bauer, 1996; Yang et al., 2001; Viera and Garrett, 2005; Foody, 2010).

For the present study, the 120 samples (10 samples per class) are derived for accuracy assessment of all LULC feature class of the classified images for the year 2000 and 2011. The post field ver- ification of LULC features is performed using 120 ground truth points observed with the help of handheld GPS (global positioning system) instrument (Garmin GPSMAP 76Cx). These reference points are overlaid on the classified images to check the accuracy of feature classification and the results shows that the 89% accu- racy in post field verification of LULC features. Moreover, the con- fusion error matrix is executed comparative analysis between reference points (120 ground truth points derived from GPS based field survey) and the corresponding pixels of LULC features of clas- sified images to calculate the producer’s accuracy, user’s accuracy and Kappa coefficients for all LULC classes along with overall accu- racy and overall Kappa coefficients. The statistical information of accuracy assessment is shown inTable 3. The result shows the overall accuracy is estimated as 81.16% and 77.52% for the classi- fied images of the year 2000 and 2011 respectively. The overall Kappa coefficient value of the classified images of the year 2000 and 2011 is 0.86 and 0.76. The derived values indicate the accept- able accuracy of the classified LULC features for landuse and land cover studies in the coastal region. Furthermore, the accuracy of extracted feature classes are randomly verified using reference points collected from GPS surveyed data for beach landforms; Sur- vey of India published topographical map for relief and drainage;

NRSC (National Remote sensing Centre) published LULC map for crop lands and plantations; and high resolution image of Google Earth for settlements and built-ups. The ground truth verification reveals that location and distribution of croplands, dune vegeta- tions, sandy beaches and built-ups are demarcated with relative high accuracy than other classes. Settlements and built-ups are recorded with producer accuracy of 56.99% 2000 and 69.16% for the year of 2000 and 2011 respectively. It is mainly due to demol- ished built-ups after tsunami on 26th December 2004. The change of built-ups has also produces significant variation in relative accu- racy of Kappa coefficient values from 0.67 to 0.76 for the decade.

Between the year 2000 and 2011, the river and beach mining areas are noted the lower Kappa coefficient values such as 0.63 and 0.62 due to regular activities of surface runoff and beach mining that

changing the landforms dynamically. However, some of LULC fea- tures like saltwater bodies and beach mining are classified with bias due to overlapping of pixels and the overlapped pixels are carefully recoded with the help of reference points. Overall obser- vation of accuracy shows the reliability of LULC features in terms of location, area and spatial distribution in the study area.

5. Conclusion

Decadal changes of LULC features reflect the socio-economic conditions in the area and a key driver indicating the hazards and vulnerability due to natural and anthropogenic activities. GIS and remote sensing provides an effective platform for assessment of LULC changes and transformations over time. In the study area, it is observed that area in beachface landforms, plantations, cul- tivable lands, fallows, and barren lands are converted into settle- ments and built-ups and it is increased twice in spatial extent from 2000 to 2011 due to human encroachment and urban expan- sion activities. Decreasing of area in these land covers affects the sediment load that causes shoreline erosion along the various parts of the coastal regions. The coastal structures like groins, revet- ments, seawalls are evidently decreasing the areal extent of the beaches on the down-drift due to backwashing of more sediment by rip currents that leads morphological changes landforms and dynamics of shoreline throughout the year. It is noticed that the plantations and dune vegetative cover are uprooted or severely damaged during the year 2004 due to the effect of Tsunami occurred on 26th December 2004. During the last decades, the cul- tivable land and fallow land are encroached for built-ups due to rapid growth of populations and this causing landform degrada- tion, loss of biodiversity, seawater intrusion and pollution of groundwater in the coastal region. Similarly, the saltpan and salt- water bodies are increased in area due to placer mining and other anthropogenic activities. Significantly, the LULC features are dynamically transformed to another land uses due to requirement of the local people. For example, the cultivable lands are converted to built-ups due to human encroachment activities, whereas the natural processes as land degradation and surface runoff are trans- formed the cultivable land into fallow land and barren results loss of soil fertility and agriculture productivity in the study area.

Unfortunately, most of the LULC features are transformed to built-ups and settlements and other land uses without considering their negative impacts on the coastal systems and processes with increasing rate of vulnerability to the coastal landforms. Unregu- lated LULC features shifting activities is alarming the environmen- tal issue in the coastal zones and severely violated the government regulations and policies. This study provides primary information

Table 3

Accuracy assessment of classified Landsat ETM+ images for 2000 and 2011.

LULC Class Name Producer’s accuracy (%) User’s accuracy (%) Kappa coefficient

2000 2011 2000 2011 2000 2011

Beachface land cover 86.80 72.35 82.60 78.21 0.77 0.68

Dune vegetation 78.16 67.54 66.24 65.68 0.85 0.79

Plantation 69.21 81.65 68.71 76.22 0.83 0.64

Cultivable land 85.49 69.45 76.39 65.00 0.89 0.82

Fallow land 77.55 88.42 92.42 85.21 0.79 0.73

Settlement/Built-ups 56.99 69.19 72.46 64.85 0.67 0.76

Barren land 94.87 86.22 89.32 67.43 0.94 0.89

River 89.45 91.05 92.88 90.55 0.75 0.63

Water bodies 93.44 96.28 89.23 84.24 0.91 0.87

Saltwater bodies 82.14 83.56 78.52 67.22 0.73 0.80

Saltpan 94.56 98.88 93.54 89.15 0.93 0.90

Beach mining 85.45 76.95 71.65 96.52 0.84 0.62

Overall accuracy 81.16% 77.52 %

Overall Kappa coefficient 0.83 0.76