banko jankari.indd 40 Information on human induced disturbance like deforestation has gained much importance particularly in the present global context of climate change and biodiversity conservation challenges. Satellite remote sensing methods have been widely used to map area and patterns of deforestation as well as to analyze the rates of forest cover change (Apan and Peterson, 1998; Franklin et al., 2002; Hall et al., 1991; Mas, 1999; Mas et al., 2004). Forest loss and fragmentation process through habitat loss, fragmentation and isolation of forest patches alter the landscape structure and functions, and ultimately have several ecological effects on ecosystem (Matsushita et al., 2006, McGarigal and Cushman, 2002). Therefore, it is important for deforestation analysis to include spatial dynamics of the forest, landscape which provides information on the temporal change in the patch metrics such as size, number, shape, adjacency and the proximity of patches in a landscape. Previous studies have observed forest cover change in the Terai, a physiographic region of Southern Nepal, which is important from both biodiversity conservation as well as rich forest resources (DFRS, 1999; DoF, 2005; Kandel et al., 2010; Khanal, 2009). This study analyses the spatial and temporal pattern of forest cover change using multi-temporal Landsat Satellite imageries and uses FRAGSTATS, a spatial pattern analysis program to analyze characteristics of landscape fragmentation in Laljhadi Corridor forest of Terai Arc Landscape in western Nepal. Materials and methods Study area Laljhadi forest, a biological corridor lies in Kanchanpur district of the far western region of Nepal. The corridor which encompasses four Village Development Committees (VDCs) (Raikar Bichawa, Baisi Bichawa, Shankarpur and Krishnapur) (Fig. 1), which are located between Dudhwa National Park of India and Shukla Phanta Wild Life Reserve of Nepal. It is a very important corridor as it connects Nepal’s Churia forest in the North to India’s Dudhuwa national park in the South. Geographic positions are 80° 20’E to 80° 33’E and 28° 38’N to 29° 10’N approximately. Using Landsat data for assessing forest cover change and fragmentation in Laljhadi corridor of Kanchanpur district, Nepal R.R Aryal1*, H. L. Shrestha2 and S. Khanal1 The study, carried out at Laljhadi corridor in Kanchanpur district of Nepal, aimed at assessing forest cover change and fragmentation using multi-temporal Landsat data. Post classifi cation change detection was applied on temporal forest cover class datasets obtained by supervised classifi cation technique with maximum likelihood algorithm. The overall change analysis indicated a decreasing trend in forest cover. Statistics on selected landscape metrics were generated to quantify the change in spatial structure resulting from fragmentation. The analysis of the landscape metrics depicted increase in fragmentation over the analysis time period along with progression of deforestation. Key words: Fragmentation, forest cover change, Landsat, Laljhadi corridor 1 Department of Forest Research and Survey, Kathmandu, Nepal 2 International Center for Integrated Mountain Development (ICIMOD), Lalitpur, Nepal * Corresponding author: rajaramaryal@gmail.com 41  Fig. 1: Study area in kanchanpur district, Nepal Data Four Landsat images (Table 1) were downloaded from the United States Geological Service (USGS), Global Land Cover Facility (GLCF) (http:// glcf. umiacs.umd.edu/) web sites on 10 November 2010. All scenes correspond to the peak of the growing season of vegetation, with 9 to 13 years in between acquisitions. Ancillary datasets included 1964 aerial photographs from the Department of Forest Research and Survey , 2010 Rapid Eye image from Forest Resource Assessment Project Nepal as well as 1:25,000 topographic sheets from Survey Department. Landsat Multi Spectral Scanner (MSS) and Thematic Mapper (TM) satellite imageries selected for forest cover change analysis had covered a period of 33 years (1977 - 2010). Table 1: Name and character of the images Satel l i te Type Sensor Date Spatial Resolution (m) Landsat 2 MSS 1977 Jan 23 57 Landsat 5 TM 1990 Oct 23 30 Landsat 5 TM 1999 Nov 09 30 Landsat 5 TM 2010 Oct 30 Image preprocessing Histogram matching was applied so that the histogram of multi-temporal imageries resembles each other and aid classification and change detection. All images were geo-referenced using Survey Department’s Topo Layers and projected to Universe Transverse Mercator (UTM) zone 45. To match the pixel size with Landsat TM, re-sampling of MSS data to 30 m pixel was done using cubic convolution. Image classifi cation The image classifi cation was carried out using ERDAS Imagine 9.2. A supervised classifi cation technique with maximum likelihood algorithm was applied. Landsat images of all the required data were classifi ed into three broad classes: forest cover, water body and others (cultivated land, settlement, barren land). Training samples using ancillary datasets were taken as signature classes for classifi cation. For classifying the image of 1977, 1990, 1999 and 2010, the aerial photo of 1966, topographic map of scale 1: 25000, Rapid Eye image of 2010 were used as reference along with Google Earth. After the classifi cation fi ltering was done fi ltering out patches less than one hectare. After classifi cation in ERDAS 9.2, the classifi cation maps were exported to Arc GIS 9.2 for further processing. Accuracy assessment Typical accuracy assessment approach was applied, which involved verification of the randomly generated locations using reference data. For the quantitative analysis of the image classifi cation, Kappa statistics was applied. Kappa statistics is a measure of agreement between image data and reference data (Jensen, 1996). Overall classification accuracy and Kappa coefficient statistics were derived for each image date as presented in table 2. Table 2: Result of Kappa and accuracy assessment Dates of image Overall kappa coeffi cient Overall classifi cation accuracy 1977 0.9081 95.33% 1990 0.8942 93.33% 1999 0.9439 96.67% 2010 0.8494 91.33% 42   Forest cover change and fragmentation analysis Post-classifi cation techniques are considered to have limitations as comparison of land cover classifi cation does not allow detection of subtle change within land cover categories (Macleod and Congalton, 1998). However, in the present study, since Landsat which is a coarse resolution image, was used focusing on only one forest class with threshold of more than one hectare size, change detection was performed using post-classifi cation comparison method which produced acceptable results. The post classifi cation technique of image differencing was applied on subsequent pairs of the classifi ed single date images so that image difference data were obtained for the three time interval. The classified multi-temporal image with forest and non forest classes were analyzed using spatial statistics of FragStats 3.3 (McGarigal and Marks, 1995) interface inside Patch Analyst 4.2.13 (Rempel et al., 2008) using metrics as listed in table 3. Table 3: List of metrics Landscape Description metrics CA Class area (ha) sum of areas of all patches belonging to a given class. NumP Number of patches for each land use class MPS Mean patch size (ha) MedPS Median patch size (ha) the mpatch size, or 50th percentile. PSCoV Patch size coeffi cient of variance, coeffi cient of variation of patches. PSSD Patch size standard deviation (ha), standard deviation of patch areas. TE Total edge (m) sum of perimeter of forest class ED Edge density (m/ha) amount of edge relative to the landscape area. MPE Mean patch edge (m/patch) average amount of edge per patch (TE / NumP). Results and discussion Forest cover change Landsat MSS Images and TM images of 1977, 1990, 1999 and 2010 were used to produce forest cover maps of the study area (Fig. 2). It was revealed that a considerable amount of forest (4581.72 ha) had decreased during the entire study period. Table 4 depicts the statistics on total forest area and the equivalent percentage of area changed during the time intervals. Highest loss of forest cover (2500 ha) took place between 1999 and 2010. Table 4: Total Forest area in ha, percentage and change Year Forest area(ha) Percent Change area (ha) 1979 25902.42 72.66 - 1990 24573.78 69.26 -1328.64 1999 23866.32 67.26 -707.46 2010 21320.70 60.09 -2545.62 Fig. 2: Forest cover map of the study area in different time period 43  Forest fragmentation analysis The analysis of selected landscape metrics depicted increase in fragmentation over the analysis time period along with progression of deforestation (Table 5). The sum of areas of patches belonging to forest class or the class area (CA) showed trend of deforestation. The number of forest patches (NumP) increased heavily in 1990s and declined sharply in 2010. Field observation as well as the interpretation of secondary data revealed that this may have happened because of many disturbances in forests due to road construction, deforestation and migration in 1990s. By 2010, most of the smaller forest patches had already been converted to other land use classes. Table 5: Selected landscape metrics for forest class Metrics 1979 1990 1999 2010 CA 25902.42 24573.78 23866.32 21320.70 NumP 113 153 95 52 MPS 229.22 160.61 251.22 410.01 MedPS 2.70 2.67 3.54 13.47 PSCoV 597.22 695.68 539.93 411.62 PSSD 1368.99 1117.35 1356.45 1687.72 TE 482975.2 656808.8 524871.1 393510.05 ED 18.64 26.72 21.99 18.46 MPE 4274.12 4292.87 5524.96 7567.50 Mean patch size (MPS) and Median patch size (MedPS) showed overall increment from 1979 to 2010, indicating disappearances of small sized patches. Patch size coeffi cient of variation (PSCoV) increased in 1990 but later decreased in subsequent time slices. Patch size standard deviation (PSSD) which measures absolute variation in patch size and is affected by the average patch size also showed the identical. Total edge (TE) which is sum of perimeter of patches showed decline trend in number of patches, while edge density (ED) remained almost static in the same time interval. However, the average amount of edge per patch, and the mean patch edge (MPE) almost doubled in 2010 as compared to 1979. It is interesting to note that for 2010, MPE is higher which is obtained by dividing total edge by number of patches though while both values are smaller than other time slices. Conclusion The multi-temporal forest cover change analysis revealed the important changes both in the areas of deforestation and fragmentation of forests. A very clear trend toward forest fragmentation was observed with a continuous trend over time of decreasing forest cover as indicated by increase in the number of forest patches and ultimately decreasing following the conversion of smaller patches to other land use types. The use of multi- temporal Landsat images to monitor spatial pattern of forest cover change further supports the potential applicability of Landsat sensor products for monitoring other land use dynamics in Nepalese terrain. However, it is recommended that to understand more detailed dynamics of landscape fragmentation, future research should use higher resolution datasets and encompass wider landscape units rather than part of a small corridor. References Apan, A. A. and Peterson, J. A. 1998. Probing tropical deforestation – the use of GIS and statistical analysis of georeferenced data. Applied Geography 18 (2):137–152. DFRS. 1999. Forest Resources of Nepal 1987 –1998. Department of Forest Research and Survey, Ministry of Forests and Soil Conservation, His Majesty’s Government of Nepal, Forest Resource Information System Project, Kathmandu, Nepal DoF. 2005. Forest Cover Change Analysis of the Terai Districts 1990/ 91-2000/01. Department of Forest, Kathmandu, Nepal. Franklin, S. E., Lavigne, M. B, Wulder, M. A. and Stenhouse, G. B. 2002. Change detection and landscape structure mapping using remote sensing. The Forestry Chronicle 78 (5): 618-625. Hall, F. G., Botkin, D. B., Strebel, D. E., Woods, K. D. and Goetz, S. J. 1991. Large-scale patterns of forest succession as determined by remote sensing. Ecology 72 (2): 628–640. Jensen, J.R. 1996. Introductory Digital Image Processing. New Jersay: Trentice Hall, USA 44   Kandel, C. M., Caetano, M. and Cabral, P. 2010. Forest covers monitoring of Bara district (Nepal) using remote sensing and geographic information systems. Banko Janakari 20 (1): 30-36. Khanal, S. 2009. Change assessment of forest cover in Ghodaghodi lake area in Kailali district of Nepal. Banko Janakari 19 (2): 54-57. Macleod, R. D., and Congalton, R. G. 1998. A quantitative comparison of change-detection algorithms for monitoring eelgrass from remotely sensed data. Photogrammetric Engineering and Remote Sensing 64 (3): 207-216. Mas, J. F. 1999. Monitoring land–cover changes: A comparison of change detection techniques. International Journal of Remote Sensing 20 (1): 139- 152. Mas, J. F., Velázquez, A., Díaz-Gallegos, J. R., Sauced, R. M., Alcántara, C., Bocco, G., Castro, R. Fernandez, T. and Perez- Vega, A. 2004. Assessing land use/cover changes: A nationwide multidate spatial database for Mexico. International Journal of Applied Earth Observations and Geoinformation 5 (4): 249–261. Matsushita, B., Xu, M., and Fukushima, T. 2006. Characterizing the changes in landscape structure in the Lake Kasumigaura Basin, Japan using a high-quality GIS dataset. Landscape and Urban Planning 78 (3): 241–250. McGarigal, K., and Cushman, S. A. 2002. Comparative evaluation of experimental approaches to the study of habitat fragmentation studies. Ecological Applications 12 (2): 335–345. McGarigal, K. and Marks, B. J. 1995. Fragstats: Spatial Pattern Analysis Program for Quantifying Landscape Structure. U. S. Forest Service General Technical Report. Portland, USA. Rempel, R.S., Carr, A.P. and Kaukinen, D. 2008. Patch Grid Extension for ArcMap: Version 4.2. Ontario Ministry of Natural Resources. Ontario, USA.