Spatio-temporal Variation in Landuse/cover Dynamics in Shahzad River Basin Uttar Pradesh, India: A Geospatial Approach

Landuse/cover is an important component that reflects the interaction between environment and human activities. Landuse/cover pattern is an outcome of natural and socio-economic factors and their utilization by man in space and time. The present study makes an attempt to monitor landcover dynamics in Shahzad river basin, a rainfed basin in Lalitpur district of Uttar Pradesh (India) using remote sensing and GIS technique. Digital Elevation Model (DEM) was prepared using Shuttle Radar Topography Mission (SRTM) data, the lowest and highest elevations encountered in the basin are 280 m and 495 m above MSL respectively. The higher elevations are encountered in southern most parts of the basin whereas lower elevations are found in the north. The general slope is from south to north, as defined by the course of the Shahzad river. Two data set viz. IRS-P6 LISS III data of 2005 and IRSP6 LISS III of 2015 have been analyzed through visual interpretation technique. Visual interpretation technique was used to identify the various landuse/cover categories. Landuse/cover maps of 2005 and 2015 derived from satellite data were digitized in Arc GIS environment. Editing and topology building was carried out using Arc GIS 10 and area under each category of landuse/cover was computed in km2 as well as in percentage. An attempt was made to estimate and quantify the overall change as well as transitional change in landuse/cover classes over a decade. A comparison of 2005 and 2015 data analysis suggests that the area of water body has significantly increased from 19.53 km 2 in 2005 to 34.85 km 2 in 2015, i.e. 15.32 km 2 (1.39%), the area of uncultivated land has decreased from 352.81 km 2 (32.61%) in 2005 to 337.80 km 2 (30.7%) in 2015, showing 15.01 km 2 (1.91%) decrease. The area of cultivated land has increased from 464.78 km 2 (42.76%) to 473.06 km 2 (43%), i.e. 8.28 km 2 (0.24%). However, vegetation in the watershed has reduced, i.e. under open forest and dense forest has reduced by 11.37 km 2 (1.04%) and 13.27 km 2 (1.21%) respectively during 2005-2015. Open scrub, stone quarry, built-up land, exposed rock and wasteland have also reported change in their respective areas. Open scrub has slightly increased from 73.63 km 2 (6.69%) to 76.23 km 2 (6.93%), whereas stone quarry has increased from 1.78 km 2 (0.16%) to 3.53 km 2 (0.32%), which suggests expansion in quarrying activity. Change matrix analysis indicates that cultivated land, uncultivated land, open scrub and wasteland are the most unstable categories which have interchanged into different landuse/cover during 2005-2015. Moreover it also indicates, cultivated land (38.42 ha), uncultivated land (51 ha) and open scrub (59.64 ha) has been converted into settlement area.


Introduction
Land use and land cover (LULC) are terminologically different terms but are often used interchangeably. Land cover refers to the physical characteristics of the earth including area captured under vegetation, water, soil, landforms, topography etc. whereas landuse refers to the pattern in which land resources have been utilized by the humans. Different driving forces acting upon the surface of earth lead to the formation or modification of land use/cover categories. The land use/cover pattern of a region is an outcome of natural and socio-economic factors and their utilization by man in time and space (Rawat and Kumar, 2015). Hence, assessing the land use/ cover and their dynamic nature is not only important for estimating the area of a land used under different categories but in turn helps in evaluating the potential of natural resources, watershed management, socio-economic planning and other development sectors.
Since land use/cover are paired complementary to each other thus change in land use affects the land cover and vice versa. Change in land cover does not primarily indicate deterioration of land but it may be driven by the anthropogenic activities and can also be triggered by natural processes there by affecting the biospheric and climatic cycle. Since natural resources are dynamic in nature, thus land use/cover changes are important elements for monitoring, evaluating, protecting and planning for earth resources (Rawat et al., 2013). LULC classification and change detection is one of the most reliable methods to monitor the land dynamics. Change detection of LULC is generally carried out using multitemporal data sets to assess changes in the landscape witnessed in the country, especially at watershed or basin level to provide more accurate information about landuse/cover dynamics for proper management and conservation of natural resources (Gibson and Power, 2000).
Remote sensing and GIS have emerged as one of the most powerful tools for classifying, mapping, monitoring, evaluating and assessing the natural resources in less time, with low cost and better accuracy. Satellite image pixcels can be classified in LULC categories either by automatic extraction or by visual interpretation method (Meinel and Neubert, 2000). Automatic classification is based on extrapolating the celebrated patches of homogeneous color and texture on the satellite image, thereby providing a meaningful categorization of LULC (Dronova et al., 2011;Chen et al., 2012). Further the pixel-based analysis may lead to the 'salt-and-pepper' effect in mapping heterogeneous landscapes (Zhang et al., 2014). It is difficult to determine the accuracy of the result produced by the automatic classification, even sometimes it fails to recognize difference in "use" of given surface land (Fallati et al., 2017). On the other hand, visual interpretation is a slow and tiresome method as compared to the automatic supervised classification but can be rationally used when analyst is familiar with the field and research area is comparatively small. Visual interpretation has advantage over automatic classification since the former can identify dynamic changes more accurately and hence represents the process of landuse/cover change more effectively. Therefore this interpretation technique is frequently used in practical approach like in rural development, smart city project, flood mapping etc. (Zhang et al., 2014). Vector layer in form of polygons are used to divide the image into the parcels, these parcels are classified into desired classes, thus avoids inter-class spectral variation (Apline et al., 1999;Apline and Smith, 2008).
In the recent past the advancement of research has led to more emphasis on development at watershed level, as the traits and allocation of natural resources depends upon the natural boundary. Many workers consider watershed as a basic unit for the identification, estimation, development and prioritization of natural resources. Adinarayana et al. (1995) used integrated approach to prioritize watershed on the basis of sediment yield index in Western Ghats, Rahaman et al. (2015) prioritized sub watershed on morphometric basis using Fuzzy Analytical Hierarchy Process in Kallar watershed, Welde (2016) conducted the study on prioritization of watershed for land and water management in Tekeze dam, Ethiopia. Watershed, besides being a naturally occurring hydrological unit also carries a unique socio-ecological aspect which plays an important role in determining the ecological, food and social security and provision of life support services to local communities. Several attempts have been made in the recent past to quantify the change detection by applying remote sensing and GIS techniques. Kibret et al. (2016) carried out a study in South Central Ethiopia, to assess the LULC changes during four decades using multi temporal satellite data. Fallati et al. (2017) have carried out LULC analysis in the Republic of Maldives using remote-sensing data through visual interpretation technique in conjunction with socioeconomic data.
The present study makes an attempt to assess the LULC changes and probable causes leading to the changes in the Shahzad watershed, Lalitpur district, Bundelkhand, India from 2005 to 2015. The region of Bundelkhand have complex geological and physiographical landscape varying from Bundelkhand Plains, Bundelkhand upland and vindhyan Plateaus. The study area has remained socioeconomically backward with moderate drought frequency but high intensity, and faces the metrological drought. Hence, monitoring the resources and assessing the cause of changes become a prime focus especially in the agricultural dominant watershed. The major objectives of the study are to (i) identify and delineate different LULC categories under the area for 2005 and 2015 (ii) Verify the accuracy of the data in order to get more reliable and accurate results (iii) quantify the amount of overall change in LULC categories both in area and percentage as well as to monitor the LULC transformation from one category to another and (iv) ascertain the possible causes of change. Several studies have been carried out throughout the globe on the change detection analysis by using remote sensing data (Palaniyandi and Nagarathinam, 1997;Rogan and Chen, 2004;Güler et al., 2007;Chunxiao et al., 2008;Song et al., 2009;Prakasam, 2010;Nagarajan and Poongothai, 2011;Liu et al., 2012 andGarai andNarayana, 2018).

Study Area
The Shahzad river basin is located in Lalitpur district of Bundelkhand region in Uttar Pradesh (India) and bounded by 24 ○ 25' N to 25 ○ 03'N latitudes and 78 ○ 17' E to 78 ○ 39' E longitudes with a geographical area of about 1100 km 2 . Shahzad river (a tributary of Jamini river) flows from south to north through the center of the watershed bifurcating it into East and West. The basin experiences sub-humid climate with average annual rainfall of 806 mm. Topography is represented by plains, pleateau, hills and ridges where elevation values range from 280m to 495m above mean sea level. Geologically, the area is dominated by Bundelkhand Granatoid Complex (BGC) comprising of gneisses, schist and granites which are underlained by arechean formation. North East-South West (NE-SW) trending quartz reefs and basic dykes traverse the BGC. Thick pile of quaternary sediments of older alluvium comprising of Banda alluvium overlies on the BGC formation. Sandstone of Kaimur group (Upper Vindhyan) is exposed in the south (GSI, 2008). The basin is mainly drained by Shahzad river and its tributaries. The dendritic to sub-dendritic drainage pattern dominates the watershed, however there are variation in drainage pattern at local level, where main river divides the entire watershed into eastern and western halves. Two reservoirs namely Shahzad reservoir and Govindsagar reservoir are the characteristics feature in the north and south respectively.
The watershed is mainly defined agricultural land where the main landuse is cultivation which is by and large rainfed, however irrigation from ground water at places is through tube wells. Major crops grown in the area are Sorghum, Pigeon pea, Black gram, Green gram, Barley, Mustard etc. Over exploitation of ground water has not only resulted in lowering of water level but has also caused failure of wells.

Data Base Preparation
IRS P6 LISS III false color composite (FCC) of 2005 and 2015 February, with band combination 2,3,4 and spatial resolution of 23.5m were used for landuse/cover mapping and classification ( Figure 2). The data for same month has been used to reduce the reflectance conflict and seasonal variation for the classification of landuse/cover. The satellite data pertaining to study area was obtained from National Remote Sensing Centre (NRSC), Hyderabad. The data set were imported to ERDAS imagine version 14 to extract the desired form of image used for classification purpose. The layer stack option in raster tool box was used to generate standard false color composite (FCC). Image was geo-referenced and was projected in Universal Transverse Mercator (UTM) projection, taking World Geodetic System (WGS84) as the datum. The study area lies in UTM Zone 44 N. The georeferenced boundary of Shahzad watershed as Area of Interest (AOI) was superimposed on satellite image to subset the image and to extract the study area.

Landuse/Cover Delineation and Analysis
Visual interpretation technique was followed for change detection of land use/cover using ArcGIS 10 software, since the authors are familiar with the area being classified. Interpretation keys were developed on the basis of field knowledge and literature. To enable storage of data associated with each LULC, a data base structure was created prior to visual interpretation of IRS FCC images. Based on spectral characteristics of landuse/cover classes, the interpretation keys were developed (Table 1) and on-screen digitization for various landuse/cover classes was created in form of polygons with unique ids. Ground truthing was carried in key areas to verify the spectral signature. Twelve landuse/cover classes were identified in the study area viz. (i) cultivated land (ii) uncultivated land (iii) dense forest (iv) open forest (v) open scrub (vi) waterbody (vii) dry waterbody (viii) exposed rock (ix) stone quarry (x) industrial land (xi) built up land and (xii) wasteland.   ) for 2005 and 2015 was created using the observed and the classified landuse/cover map of each class for accuracy assessment. Accuracy parameter viz. producer's accuracy (a measure of omission error), user's accuracy (a measure of commission error), overall accuracy and kappa statistics were estimated (Thakkar et al., 2017). Producer's accuracy is a ratio of number of correctly classified pixels to the number of training pixels (the column total) used in particular category whereas User's accuracy is obtained by dividing the number of correctly classified pixel by total number of pixel in that category (the row total). The overall accuracy is computed by dividing the number of correctly classified pixel by total number of reference pixel. Kappa analysis is a discrete multivariate technique used in accuracy assessments (Jensen 1996). Kappa analysis yields a Khat statistic (an estimate of Kappa) that is a measure of agreement or accuracy (Congalton 1991).
Where, r = number of rows in the error matrix x ii = number of observation in row i and column i (on the major diagonal) x i+ = total of observations in row i (shown as marginal total to right of the matrix) x +1 = total of observation in column i (shows as marginal total at bottom of the matrix) N = total number of observation included in matrix

Landuse/Cover Change Detection and Analysis
Post classification technique was applied to the generated temporal landuse/cover maps to perform the landuse/cover change detection. Vector layer of 2005 and 2015 were overlaid to produce change information inform of "from" to "to" classes. Cross tabulation of two different decadal data of classified image were compared to determine the qualitative and quantitative aspect of the change for the period 2005-2015. Change matrix was produced in GIS environment. Quantitative areal data of the overall landuse/cover changes as well as gains and losses in each category between 2005 and 2015 were then compiled. Open scrub occupies an area of 73.63 km 2 (6.69%), which is wide spread throughout the whole basin.

Landuse/Cover Status
Open forest though present as patches throughout the basin has an area of 24.99 km 2 (2.27%), whereas dense forest occupies an area of 65.07 km 2 (5.91%) and is confined to southern and northwestern parts of the basin. The exposed rock terrain is largely confined to the southern part of the basin and occupies an area of 0.56 km 2 (0.05%). Although the settlements/built-up land are present throughout the basin, but are largely confined to the central part of the basin in and around Lalitpur

IJARSG-An Open Access Journal (ISSN 2320 -0243)
town and occupies an area of about 15.06 km 2 (1.36%). Wasteland occupies an area of 77.64 km 2 (7.05%) and is found in the northern and western parts, and is associated with the open scrub, cultivated and uncultivated lands. The other land use/land cover category viz; industrial land and stone quarry occupy areas of about 3.87 km 2 (0.31%) and 1.78 km 2 (0.16%) respectively. The quarrying activity is reported from southern part of the Shahzad watershed, where sandstone is being mined for use as a building stone. Figure 3 (a) presents land use/land cover map derived from LISS III FCC data of 2005.
The other dominant LULC categories are wasteland and open scrub with areas of 83.53 km 2 (7.59%) and 76.23 km 2 (6.93%) respectively. Natural vegetative cover i.e. dense forest covers 51.80 km 2 (4.70%) area whereas open forest occupies 13.62 km 2 (1.23%) area. Though the area under waterbody has increased to 34.85 km 2 (3.16%) but an increase in area of about 2.07 km 2 (0.18%) is also observed in dry waterbody, compared to 2005. The other land use/land cover categories viz; industrial land and stone quarry, occupy areas of about 4.69 km 2 and 3.53 km 2 respectively. Table 2 presents details of area and extent of change under each category of LULC in the Shahzad watershed during 2005-2015. Figure 3 (b) presents LULC map derived from 2015 IRS FCC data.

Accuracy Assessment
Accuracy assessment is an important factor in classification. One of the most commonly followed method is the preparation of a classification error matrix (confusion matrix/contingency table). Error matrices compare, on a category-by-category basis, the relationship between known reference data (ground truth) and the corresponding result of a classification (Lillesand et al., 2004).
The error matrix shows the overall accuracy of 83.2% and 84.6% followed by Kappa coefficient of 0.80 and 0.82 in classified images of 2005 and 2015 respectively. Both producer and user accuracy of individual classes ranges from 65.21% to 96.11% and 71.42% to 100% in 2005 (Table 3) whereas for 2015, producer and user accuracies turned out to be 71.42% to 100% and 73.33% to 96.07%, respectively (Table 4).

Surface Water and Landuse Pattern
Agriculture in Shahzad basin is rainfed and depends upon the rainfall but it has been observed that area around the surface water body especially environ dams have shown positive change over the decade. Figure 6 shows the water spread in the reservoir during 2005 and 2015 which has resulted into conversion of uncultivated lands into cultivated land thereby indicating good irrigation facilities and water availability.

Conclusion
Multi temporal satellite imagery and GIS application have proved to be cost effective and accurate method to quantify the spatial and temporal phenomenon of LULC dynamics which would have not been possible with the conventional mapping. From the study it is concluded that LULC is dynamic in nature, thus interconversion of one LULC class to another at different part of the watershed in same time period is frequent depending upon the local variation like water availability, geology, livestock stress etc. e.g. in some part of the study area, 15. open forest and dense forest has been converted into open scrub. Areal extent of dry waterbody has increased due to decline rainfall and utilization of surface water but overall extent of waterbody has also increased as water from Rajghat reservoir (built on the Betwa river) fed to Govindsagar reservoir for irrigation purpose. Settlement has increased due to expansion of Lalitpur town over a decade. The present study highlights the capacity of remote sensing coupled with GIS in analyzing the change dynamics of LULC. A study like this will open a new spectrum for planning, managing and utilizing the available natural and land resource at watershed level.