decadal land cover change dynamics in bhutan
TRANSCRIPT
lable at ScienceDirect
Journal of Environmental Management xxx (2014) 1e10
Contents lists avai
Journal of Environmental Management
journal homepage: www.elsevier .com/locate/ jenvman
Decadal land cover change dynamics in Bhutan
Hammad Gilani a,b,*, Him Lal Shrestha a,c, M.S.R. Murthy a, Phuntso Phuntso d,Sudip Pradhan a, Birendra Bajracharya a, Basanta Shrestha a
a International Centre for Integrated Mountain Development, GPO Box 3226, Kathmandu, Nepalb School of Geomatics, Liaoning Technical University, 47 Zhonghua Road, Fuxin, Liaoning Province, ChinacDepartment of Environmental Science and Engineering, School of Science, Kathmandu University, Dhulikhel, NepaldDepartment of Forests and Parks Services, Ministry of Agriculture and Forest, Royal Government of Bhutan, Thimphu, Bhutan
a r t i c l e i n f o
Article history:Received 15 July 2013Received in revised form4 February 2014Accepted 15 February 2014Available online xxx
Keywords:Land coverSatellite dataLandsatHarmonized legendLCLU management
* Corresponding author. International Centre for Inment, GPO Box 3226, Kathmandu, Nepal. Tel.: þ977 1
E-mail address: [email protected] (H. Gilani).
http://dx.doi.org/10.1016/j.jenvman.2014.02.0140301-4797/� 2014 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Gilani, H., ehttp://dx.doi.org/10.1016/j.jenvman.2014.02
a b s t r a c t
Land cover (LC) is one of the most important and easily detectable indicators of change in ecosystemservices and livelihood support systems. This paper describes the decadal dynamics in LC changes atnational and sub-national level in Bhutan derived by applying object-based image analysis (OBIA)techniques to 1990, 2000, and 2010 Landsat (30 m spatial resolution) data. Ten LC classes were defined inorder to give a harmonized legend land cover classification system (LCCS). An accuracy of 83% wasachieved for LC-2010 as determined from spot analysis using very high resolution satellite data fromGoogle Earth Pro and limited field verification. At the national level, overall forest increased from 25,558to 26,732 km2 between 1990 and 2010, equivalent to an average annual growth rate of 59 km2/year(0.22%). There was an overall reduction in grassland, shrubland, and barren area, but the observationswere highly dependent on time of acquisition of the satellite data and climatic conditions. The greatestchange from non-forest to forest (277 km2) was in Bumthang district, followed by Wangdue Phodrangand Trashigang, with the least (1 km2) in Tsirang. Forest and scrub forest covers close to 75% of the landarea of Bhutan, and just over half of the total area (51%) has some form of conservation status. This studyindicates that numerous applications and analyses can be carried out to support improved land cover andland use (LCLU) management. It will be possible to replicate this study in the future as comparable newsatellite data is scheduled to become available.
� 2014 Elsevier Ltd. All rights reserved.
1. Introduction
Land cover (LC) change is one of the most important and easilydetectable indicators of change in ecosystem services and liveli-hood support systems. Socioeconomic drivers can induce changesin LC that may disrupt socio-cultural practices and the institutionsassociated with managing natural resources, which in turn in-creases people’s vulnerability to climate change (Agarwal et al.,2002; Lambin and Meyfroidt, 2010). LC assessment and moni-toring of LC dynamics are essential for the sustainable managementof natural resources, environmental protection, and food security(Foley et al., 2005; Jin et al., 2013).
Remotely sensed datasets and geospatial tools provide a uniquepossibility for quantifying changes happening on the earth’s
tegrated Mountain Develop-5003222.
t al., Decadal land cover chang.014
surface, whether through human impact or climate change, overtime (Huang et al., 2009). Multi-temporal LC change analysis andsimulation based on coarse to very high resolution satellite imagesis becoming a well-established technique, recognized among thescientific community and in civil society (Boggs, 2010; Gu et al.,2007; Kumar et al., 2011; Niraula et al., 2013; Townshend et al.,2012; Vogelmann et al., 2012; Xue et al., 2008).
Selection of appropriate satellite images, standard classificationschema, and methods are key challenges for LC change analysis atnational to region level (Herold et al., 2008; Townshend et al., 2012;Xin et al., 2013). Accuracy assessment is necessary to ensure thereliability of the derived LC maps; it can be based on differentsources including ground-based and ancillary information (Foody,2010; Olofsson et al., 2013). Satellite image interpretation andcategorization of features should be easy for others to adopt andreplicate as and when new satellite data become available (Foleyet al., 2005).
The Landsat satellite time series imagery is freely available andhas long been used as a valuable source for monitoring ecosystem
e dynamics in Bhutan, Journal of Environmental Management (2014),
H. Gilani et al. / Journal of Environmental Management xxx (2014) 1e102
change (Vogelmann et al., 2012; Xian et al., 2009), forest cover(Bhattarai et al., 2009; Niraula et al., 2013; Qamer et al., 2012;Townshend et al., 2012), agricultural yields (Abtew and Melesse,2013; Lyle et al., 2013), and urban growth (Yuan et al., 2005).Hansen and Loveland (2012) have reviewed a number of large areaLC change and monitoring products using Landsat imagery. A landcover classification system (LCCS) provides a comprehensive andsystematic classification legend for defined LC features that can berecognized and compared with other descriptions from anywherein the world (Gregorio, 1998). Legends based on LCCS such asGlobCover 2005e06 (Arino et al., 2007) and 2009 (Bontemps et al.,2011) are much appreciated and have been adopted in a number ofsub-national, national, regional, and global LC studies. Object orsegmentation-based image classification is considered to be betterfor change detection than conventional pixel-based satellite imageclassification algorithms (Blaschke, 2010; Duveiller et al., 2008; Ra�siet al., 2011).
In 2007, the Intergovernmental Panel on Climate Change (IPCC)Assessment Report-4 has recognized the Hindu Kush Himalayan(HKH) region as a ‘‘data-deficit area” and although scientists andinstitutions are attempting to fill some of the gaps (Singh et al.,2011), this remains broadly true. Bhutan is a landlocked countryin the eastern Himalayas, rich in natural resources in the form offorest, stunning biodiversity, and water resources. Although anumber of LC maps are available for Bhutan from different sources,as yet no comprehensive, systematic, well-recognized temporalassessment has been carried out. Noord (2010), in a review, showedthat forest cover studies of Bhutan carried out using differentremotely sensed data, methods, definitions, and legends provideddiverse results at the national level.
LC change studies provide information about the past and cur-rent situation and can be used to predict likely future trends, whichis crucial for developing effective plans for natural resource man-agement. The study described here aims to provide basic LC data forBhutan at national and sub-national level, thus filling an important
Fig. 1. Map of s
Please cite this article in press as: Gilani, H., et al., Decadal land cover changhttp://dx.doi.org/10.1016/j.jenvman.2014.02.014
gap. The study used consistent and standard satellite datasets,methodology, and definitions of features. Bhutan has a uniqueposition among developing countries in terms of forest cover andconservation area: forest, including scrub forest, covers approxi-mately 75% of the total land area, and approximately 52% of thetotal land area has some form of conservation status, the greaterpart as a protected area (Noord, 2010). Thus this study focused onanalysing LC change related to forest and protected areas. Due tothe marked altitudinal variation in the country, LC was also ana-lysed at different elevations. An LCCS-based LC scheme wasdeveloped in close consultation with national partners at a work-shop and used in the study. The accuracy of the classified maps wasassessed from reference ground points and high resolution GoogleEarth Pro images. A web-based tool was developed to facilitatedissemination and enable online data analysis.
2. Materials and methods
2.1. Description of study area
Bhutan is a landlocked country in the eastern Himalayas lyingbetween 88�450 and 92�100 E and 26�400 and 28�150 N, with a totalgeographic area of 38,394 km2 (Fig. 1). The elevation ranges from200 masl in the south to more than 7000 masl in the north, and theclimate varies with altitude. The whole country is divided intotwenty districts (local government level- dzongkhags), which arefurther divided into 205 sub-districts (grass root level e gewogs).The population in 2005 was 634,982, giving an overall populationdensity of 16 persons per km2; 69% live in rural areas and 31% inurban areas (NSB, 2005). According to the Department of Forestsand Parks Services, Ministry of Forest and Agriculture, 19,677 km2
of land (51% of the total area) has protected status, 16,396 km2 innine protected areas and 3307 km2 in twelve biological corridors(areas set aside to connect one or more protected areas andconserved and managed for the safe movement of wildlife). Bhutan
tudy area.
e dynamics in Bhutan, Journal of Environmental Management (2014),
Table 1Landsat images used for land cover mapping.
Year Satellite Sensor Data acquisition date Path/Row
1990 Landsat-5 TM 14 Nov, 1990 138/04105 Nov, 1990 139/041
2000 Landsat-7 ETMþ 28 Dec, 2000 137/04120 Nov, 2001 138/04108 Dec, 2005 139/041
2010 Landsat-5 TM 15 Feb, 2010 137/04130 Jan, 2010 137/04206 Feb, 2010 138/04128 Jan, 2010 139/041
H. Gilani et al. / Journal of Environmental Management xxx (2014) 1e10 3
has four major river systems: the Drangme Chhu; the Puna TsangChhu, also called the Sankosh; the Wang Chhu; and the Amo Chhu.
2.2. Data sources
The study used Landsat 30 m spatial resolution ortho-rectifiedand cloud free thematic mapper (TM) and enhanced thematicmapper plus (ETMþ) images for LC mapping and change detection.The images were downloaded from the USGS-EROS archive(Table 1). The whole study area lies within two Landsat images(each scene 185 � 185 km) but additional scenes were processed tocover the edges. All spectral bands were well co-registeredgeometrically but metadata were used for reflectance to radianceconversion, gains, offsets, solar irradiance, solar elevation, andacquisition of date/time given in the image. Normalized differencevegetation index (NDVI), land water mask (LWM), normalized dif-ference snow index (NDSII), and soil-adjusted vegetation index(SAVI) were extracted from the satellite images and played a sig-nificant role in image classification. The satellite image derivedindices helped in the segregation of forest from non-forested fea-tures (NDVI) and classification of water bodies (LWM) in the image.The advanced spaceborne thermal emission and reflection radi-ometer (ASTER) 30 m horizontal resolution global digital elevationmodel (GDEM), with add on products such as slope and aspect, wasused for the topographic information. Base layers with national,district, and protected area boundaries and settlement points in
Table 2LCCS-based legend for land cover mapping.
No LCC code LCC ownlabel
LCC own description
1 21495 Broadleavedforest
Mixed broadleaved species likeoak, maple, birch, Alnus nepalensis,castanopsis
2 21498 Needleleavedforest
Mixed needleleaved species likeConium, spruce, bluepine, fir, chirpine,larch, etc
3 21496//21499 Mixed forest Broadleaved species mixed with coniferspecies usually found in the transitional zo
4 21449 Shrubland Needleleaved shrubs like Juniperus spp.5 21453 Grassland Open grassland like alpine meadow6 10025 Agriculture Vegetable cultivation, irrigated or rainfed
conditions (Practised throughout the regiothe areas are often very small to map as anindividual land use class.)
7 5001 Built-up area All municipal areas8 8001 Water body Perennial flow of water including river bed9 6002-1 Barren area Rocky barren lands and rock outcrops
sometimes associated with sparse trees/scrcover
10 8005 Snow and glacier Under permanent snow and ice cover(more than 9 months per year)
Please cite this article in press as: Gilani, H., et al., Decadal land cover changhttp://dx.doi.org/10.1016/j.jenvman.2014.02.014
geographic information system (GIS) format were used both asbaseline information for the maps and for LC extraction.
2.3. LCCS legend
A land cover classification system (LCCS) is a comprehensive,standardized, a priori classification system designed to meet spe-cific user requirements and created for mapping exercises inde-pendent of the scale or means used to prepare the map. An LCCSdeveloped by the Food and Agriculture Organization (FAO) andUnited Nations Environment Programme (UNEP) with a harmo-nized legend is being applied in global level LC studies (Arino et al.,2007; Gregorio, 1998). As a part of the present study, a nationallevel workshop was conducted in Bhutan in 2010 together with thenational partners with the aim of
� upgrading the existing general legends with inputs derived fromlocal knowledge and a country perspective, and
� preparing a national legend in harmony with the HKH regionallevel legend (Table 2).
Six classes were identified at Level 1 (forest, shrubland, grassland,agriculture, built-up area, others) and ten at Level 2 (broadleavedforest, needleleaved forest, mixed forest, shrubland, grassland, agri-culture, built-uparea,waterbody,barrenarea, snowandglacier). Level1 classes were defined by combining the three classes of forest into asingle ‘forest’ class, and the three last classes into a class of ‘others’.
2.4. Satellite image segmentation and classification
A method called image segmentation is used to locate objectsand boundaries electronically within digital images to facilitateanalysis. Essentially, image segmentation means identifying adja-cent pixels that are similar with respect to some characteristic suchas colour, intensity, or texture, and dividing the image into distinctregions (‘segments’) each containing pixels with similar attributes(a simple example would be identifying the blue and green areas,and the boundaries between them, in an image of blue sky andgrass). Object-based image analysis (OBIA) techniques can then beused to identify individual objects within the image using a com-bination of spectral and spatial characteristics (Blaschke, 2010).
LCC label
Broadleaved Closed to Open TreesFloristic aspect: Oak, Maple, Birch, Alnus nepalensis, Castanopsis, etc
Needleleaved Closed to Open TreesFloristic aspect: Conium, Spruce, Bluepine, Fir, Larch, etc
neBroadleaved Evergreen Closed Trees//Needleleaved Evergreen Closed Trees
Closed to Open Shrubland (Thicket)Herbaceous Closed to Open Vegetation
n;Herbaceous Crop(s)
Built Up Area(s)s Natural Water Bodies
ubBare Soil and/or Other Unconsolidated Material(s)
Snow
e dynamics in Bhutan, Journal of Environmental Management (2014),
Fig. 2. Image classification rules for Level 1 and Level 2 land cover classes.
Table 3Spectral bands of satellite image and add on products for classification.
Imageinputs
Details/formula Interpretation
LandsatBlue (1) Blue bandRed (3) Red band ForestIR (4) Infra-Red band ForestNear-IR (5) Near Infra-Red band Forest and water bodyThermal (6) Thermal band Shadow and cloudDEM ASTER DEM with 30 m
resolutionSnow, shadow, forest andgrassland/shrub
Derived productsNDVI NIR-R/NIR þ R Vegetation and
non-vegetationLWM IR/(Gþ0.0001)*100 Land and waterNDSI (G-IR)/(G þ IR) Snow and othersSAVI (NIR-R)/(NIR þ R þ L))* (1 þ L) Agriculture, grassland and
othersNDWI (NIR e infra-red)/(NIR þ Infra-red) Water bodyBrightness Average luminance of all bands of TM Cloud, riverbed and
H. Gilani et al. / Journal of Environmental Management xxx (2014) 1e104
In this paper, a multi-resolution algorithm was used for seg-mentation of the satellite image, which locally minimized theaverage heterogeneity of image objects for a given resolution(Blaschke, 2010). Image segmentationwas performed and analysedusing various combinations of parameters (scale, shape, andcompactness). In general, the scale-25, shape-0.1, andcompactness-0.5 used in segmentation of the Landsat imagessatisfied the distinct identification of two adjacent features (Myintet al., 2013).
False colour composites (FCC) from the Landsat bands B4, B3,and B2 for red, green, and blue (RGB) colour, respectively, wereused to improve visual interpretation of satellite images and facil-itate identification of LC features. At least 10 segments were usedfor each LC class in order to develop best suitable rule sets forclassification of the 2010 images. Approximately 75% of the totalsegments were used to define rule sets for each LC class (Fig. 2) and25% to assess the accuracy of the defined rules over the entireclassified image. For each class, classification rules were evaluatedby analysing and judging the spectral responses in each spectralband and the geometry of the identified features. The accuracy ofthe segmented imagewas assessed at the entire image level, as wellas at district level. The defined rules had 90% accuracy overall. Anyslight modification in the rules was replicated from one image toanother. Satellite image indices played a significant role in definingthe appropriate ten LC classes as defined by the LCCS (Table 3).
After the final classification map was generated, classified ob-jects with an area smaller than the minimummapping unit (MMU)(1 ha ¼ 9 pixels or 3 � 3 pixels) were fused with the neighbouringLC classes. Following assessment of the accuracy of the class de-scriptions, minor modifications were made to the rule sets. Themodified rule sets were then used for semi-automatic object-basedimage classification of the 1990 and 2000 Landsat images. It waseasy to visually recognize changes in the final stages of the LC mapsand to perform manual editing for further modification.
shadowAltitude Pixel wise value of Elevation Snow, shadow, forest and
grassland/shrubAspect Derivative of DEM with pixel value
0 to 360Identification of foresttypes
Slope Derivative of DEM with pixel value indegree or percentage
Low land high landclasses separation
2.5. Accuracy assessment
Accuracy assessment was performed by compilation of 300systematic points from Google Earth Pro and 119 points located onthe ground. The Google Earth platform offers very high resolution
Please cite this article in press as: Gilani, H., et al., Decadal land cover changhttp://dx.doi.org/10.1016/j.jenvman.2014.02.014
satellite imagery for most places. These images can be used tofurther assess the accuracy of the LC maps generated using imagesegmentation and classification, which can then be used in variousnatural resource applications (Potere, 2008). The overall accuracy ofthe 2010 map was determined using a confusion matrix based onthe 419 reference points, and calculating the kappa value, standarderror, and weighted error, with a 95% confidence interval for kappa.The LC 2010 map was then used as a reference to develop the LCmaps of the earlier years.
2.6. Land cover change analysis
Land change detection using remotely sensed data is an attemptto identify natural and human impacts on earth (Qamer et al.,
e dynamics in Bhutan, Journal of Environmental Management (2014),
Fig. 3. Land cover map for 2010.
H. Gilani et al. / Journal of Environmental Management xxx (2014) 1e10 5
2012). In this study, LC maps at level 1 (6 classes) and level 2 (10classes) for 1990 to 2000 and 2010were evaluated and compared interms of area. A change matrix using level 1 data (6 classes), andcross-tabulated at two different points for 1990e2000, 2000e2010,and 1990e2010, was used to identify changes over the last twodecades. In the change matrix, diagonal values show stability of LCclass, while commission and omission values indicate a shift in areaor percentage between classes (Olofsson et al., 2013). Using thechange matrix, national level “gross change, forest to non-forest”,“gross change, non-forest to forest”, and “net change” wereextracted for each decade interval to analyse the forest situation.These were used to generate a map of the spatial trend in forestcover from 1990 to 2010 which could be linked with ancillary in-formation to help understanding of the changes and support forestresources management in Bhutan. Changes in terms of deforesta-tion/degradation and forest regeneration and re-growth between1990 and 2010 were also identified visually using medium resolu-tion (Landsat) images because it is difficult to identify small patchesin the national level map. The percentage wise LC changes withinand outside conservation areas (protected areas and biologicalcorridors) over the twenty-year period were also calculated sepa-rately. In addition, the distribution of LC 2010 classes over separate1000 m elevation intervals was plotted and analysed.
Table 4Accuracy assessment of land cover 2010.
Broadleavedforest
Needleleavedforest
Mixedforest
Shrubland Grassla
Broadleaved forest 58 3 1 2Needleleaved forest 3 49 2 1Mixed forest 2 1 39 2Shrubland 1 2 4 35 2Grassland 1 3 30Agriculture 1 1 1 2Built-up area 1Water bodyBarren area 1Snow and glacier
Total 65 55 48 44 36Producer’s accuracy (%) 89.23 89.09 81.25 79.55 83.33
Please cite this article in press as: Gilani, H., et al., Decadal land cover changhttp://dx.doi.org/10.1016/j.jenvman.2014.02.014
A web system was developed using ESRI’s ArcGIS JavaScriptapplication programming interface (API). ESRI’s ArcGIS server wasused to publish LC maps and other necessary GIS data layers on theInternet as a map service. LC thematic layers showing changes atboth the national and district levels were published to enable bothpublic visualization and runtime processing.
3. Results
3.1. Land cover change analysis
LC maps covering the whole of Bhutan were developed sepa-rately for 1990, 2000, and 2010 based on the Landsat images(Fig. 3). The overall accuracy of the 2010 LC map, assessed using the419 reference points (300 from Google Earth Pro and 119 groundpoints), was 83%, with a kappa value of 0.81, standard error kappa0.02, 95% confidence interval 0.77e0.85 and weighted kappa 0.90(Table 4).
Three classes of forest were used: “broadleaved”, “needle-leaved”, and “mixed” (broadleaved and needleleaved). All threeclasses showed a progressive small increase from 1990 to 2000 andfrom 2000 to 2010, with an overall increase in forest area from 67 to70% of total land over the whole period (Table 5). All other classes
nd Agriculture Built-uparea
Waterbody
Barrenarea
Snow andglacier
Total User’saccuracy (%)
2 66 87.8855 89.09
2 46 84.781 1 46 76.094 1 39 76.92
28 2 2 1 38 73.682 18 4 25 72.001 32 2 35 91.432 4 32 39 82.05
2 1 27 30 90.00
42 26 36 38 2966.67 69.23 88.89 84.21 93.10
e dynamics in Bhutan, Journal of Environmental Management (2014),
Table 5Land cover 1990, 2000, and 2010 (assessment based on satellite images).
Class level-1 Class level-2 1990 (km2) 1990 (%) 2000 (km2) 2000 (%) 2010 (km2) 2010 (%)
Forest Broadleaved forest 16,510 66.57 16,530 68.88 16,565 69.63Needleleaved forest 8371 9224 9446Mixed forest 677 691 721
Shrubland 4169 10.86 3401 8.86 3869 10.08Grassland 3485 9.08 3418 8.9 2000 5.21Agriculture 1144 2.98 1160 3.02 1200 3.13Built-up area 67 0.17 66 0.17 68 0.18
Others Water body 140 10.34 127 10.17 148 11.79Barren area 1630 1811 1623Snow and glacier 2201 1966 2754
Total 38,394 100 38,394 100 38,394 100
H. Gilani et al. / Journal of Environmental Management xxx (2014) 1e106
showed a small increase in area over the twenty-year period apartfrom “shrubland” and “grassland”, which showed a decrease. Thechange in built-up area was negligible (þ0.01%). The changes in“grassland” and “others” (“water body”, “barren area”, “snow andglacier”) were highly dependent on season and the time of satellitedata acquisition.
The overall forest area increased in the protected areas, bio-logical corridors, and the area outside protected areas and biolog-ical corridors (Table 6). The amount of “shrubland”, “grassland”,and “others” was higher in protected areas than in biological cor-ridors or outside the conservation areas. The area of “agriculture”within conservation areas was much smaller than the area outside.
3.2. Forest cover change analysis
A change matrix was used to identify changes in terms ofdeforestation and reforestation or regeneration. Table 7 shows the“gross change forest to non-forest”, “gross change non-forest toforest”, and “net increase in forest” from 1990 to 2000, 2000 to2010, and 1990 to 2010. There was a small change in “forest to non-forest” from 1990 to 2000, zero between 2000 and 2010, and closeto zero (�2 km2) over the whole period. The change in “non-forestto forest” was more marked with a net increase of 1174 km2 be-tween 1990 and 2010, equivalent to an average annual increase of59 km2 or 0.2%.
Fig. 4 shows the spatial distribution of forest change from 1990to 2010. The change from non-forest to forest (mostly the result ofregeneration through planting) is easy to see and is spread acrossthe country; the deforestation areas are difficult to identify. Fig. 5shows the change “non-forest to forest” at district level. Thegreatest increase was in Bumthang (277 km2), followed by Wang-due Phodrang and Trashigang, and the least in Tsirang (1 km2).
Table 6Percentage total land in different classes within and outside protected areas andbiological corridors.
Class (level 1) Protected areas (%) Biologicalcorridors (%)
Outside protectedareas and biologicalcorridors (%)
1990 2000 2010 1990 2000 2010 1990 2000 2010
Forest 18.44 19.63 19.93 7.14 7.43 7.50 40.99 41.82 42.19Shrubland 7.43 6.61 7.59 0.94 0.61 0.69 2.50 1.64 1.80Grassland 7.04 6.71 4.05 0.30 0.37 0.22 1.74 1.82 0.94Agriculture 0.24 0.24 0.25 0.13 0.13 0.13 2.61 2.65 2.74Built-up area 0.00 0.00 0.00 0.00 0.00 0.00 0.17 0.17 0.17Others 9.56 9.51 10.88 0.12 0.09 0.08 0.66 0.56 0.83
Total 42.71 8.63 48.66
Please cite this article in press as: Gilani, H., et al., Decadal land cover changhttp://dx.doi.org/10.1016/j.jenvman.2014.02.014
Small pockets of land where deforestation and forest regener-ation has taken place over the last two decades were identifiedvisually; Fig. 6 shows typical examples. The LC dynamics in Bhutancan be visualized interactively online together with other GIS layersat http://apps.geoportal.icimod.org/BhutanLandCover/index.html.
3.3. Land cover and elevation
The relationship between LC area and altitude was investigatedby analysing the distribution of LC classes in elevation ranges of1000 m in the 2010 images (Fig. 7). All LC classes except “needle-leaved forest”, “shrubland”, “grassland”, and “snow and glacier” arepresent in the lowest altitude range (1000e2000 masl); “agricul-ture” and “broadleaved forest” extend from the lowest range to4000 masl, “mixed forest” to 5000 masl, and “needleleaved forest”from 3000 to 6000 masl.
4. Discussion
The present study was conducted in order to analyse LC changefrom 1990 to 2010 on a decadal basis using uniform 30 m spatialresolution Landsat images and a consistent LCCS legend. Landsat 8has already been launched and also provides 30 m ground spatialdistance (GSD) data, thus it will be possible to monitor changes atnational and sub-national level after a further decade (in 2020)using the same methodology. An increase was observed in the areaof all three forest classes between 1990 and 2010. The Forest andAgriculture Organization (FAO) Forest Resource Assessment (FRA)reported a constant annual 0.34% increase in Bhutan’s forest from1990e2000, 2000e2005, and 2005e2010, with no deforestation(FAO, 2010). The present study enabledmore detailed identificationof changes in forest to non-forest, non-forest to forest, and overall(Table 7 and Fig. 6). There was some loss of forest area between1990 and 2000, but in the 2010 images, these areas were againidentified as forest (gross change forest to non-forest between 1990and 2010 close to zero) indicating regeneration of the previouslydeforested areas. The gain in forest area far outweighed the lossboth in each decade and over the entire twenty-year period.
Table 7Change in forested area.
Area (km2)
1990e2000 2000e2010 1990e2010
Gross change, forest to non-forest 167 0 2Gross change, non-forest to forest 1055 285 1174Net change 888 285 1172
e dynamics in Bhutan, Journal of Environmental Management (2014),
Fig. 4. Spatial map of changes in forest area between 1990 and 2010.
H. Gilani et al. / Journal of Environmental Management xxx (2014) 1e10 7
A number of factors are thought to have contributed to the in-crease in forest area, including annual plantation campaigns, de-velopments in social forestry, slow population growth rate, highliteracy rate, and use of alternative energy resources. The BhutanMinistry of Agriculture and Forest has taken responsibility forannual plantation of trees throughout the country from 1947 to thepresent. Fig. 8 shows the total plantation (126 km2) across all 20districts from 1987 to 2012 (Data source: MoAF). Buffum et al.(2009) concluded that cattle grazing inside community forests(CFs) decreased significantly, and forest regeneration increasedsignificantly, over the 5 years up to the study date.
The present study indicates that there was virtually no changein built-up areas (0.18%) between 1990 and 2010, which reflectsthe low population growth rate of 1.2% per annum (NSB, 2005).Mather (2007) considered that Bhutan’s forest growth is excep-tional compared to the situation in other Asian countries. Theresults of this study, show that only very small areas have un-dergone degradation or deforestation. However, it is possible that
Fig. 5. Increase in forest area at district level between 1990 and 2010.
Please cite this article in press as: Gilani, H., et al., Decadal land cover changhttp://dx.doi.org/10.1016/j.jenvman.2014.02.014
additional very small-scale local deforestation is occurring in areastoo small to be identified in the study. According to Bhutan’s LandAct, communities must obtain permission from the Department ofForests and Park Services to cut trees for house construction and/or fuelwood, but on occasion people may cut more trees than theyhave permission for. Illegal forest cutting and smuggling also oc-curs occasionally at night in forest areas near to built-up areas(Gyelmo, 2013), which may explain the deforestation area close toParo town (Dorji, 2013) shown in Fig. 6. Natural hazards likelandslides, glacial lake outburst floods (GLOFs), earthquakes, andforest fires can also lead to deforestation, but the impact is usuallyonly temporary. In Bhutan shifting cultivation has been completelybanned since 2007 but is still practised at a very local level (MoAF,2010). However, overall shifting cultivation does not result in aloss of forest cover as the greater part of the cycle comprises theforest fallow period.
Hydropower is the backbone of the Bhutanese economy, withapproximately 5.3 billion units (BU) energy exported to India, from2008 to 2012 (Sharma et al., 2013). Development of hydropowerplants is increasing rapidly in Bhutan, due to the advantages ofgeography and fast flowing rivers. However, hydropower projectspose a threat to flora, fauna, and habitat (RSPN, 2011), as well as tothe surrounding LCLU. The Wang Chhu 570 MW hydropower plantis under construction in Chukha district and will require 50.5 ha ofprivate land downstream. A study by Gaedu College concluded thataround 3420 trees will be affected by the project (Gyeltshen, 2013).
This study used medium resolution images which were suitablefor identification of LC in the ten major classes at national level. Amulti-resolution rule-based OBIA technique was used for the imageclassification, which can be replicated with slight modificationaccording to the spectral responses of different earth features andthe seasonality of the image captured. Commonly accepted seg-mentation parameters were used to delineate objects (Myint et al.,2013). The thresholds used for image classification were deter-mined by analysing the spectral responses, image derived products,shape, patterns, and association of the segmented objects. Forexample, objects with an NDVI value above 0.25 were classified asforest; then rules built around elevation values were used to
e dynamics in Bhutan, Journal of Environmental Management (2014),
Fig. 6. Loss (above) and regeneration (below) of forest between 1990 and 2010.
H. Gilani et al. / Journal of Environmental Management xxx (2014) 1e108
support further classification into sub-categories of forest type.Similarly, spectral band-1 of Landsat was used for snowmapping asit is well established; snowwas identified through a combination ofhigh reflectance in the blue band and an elevation above 3000masl(see Fig. 2 and Table 3). The same approach could be followed usinghigh resolution images to map more detailed classes of LC, and/orto map LC at a sub-national (e.g. district) level. The LCCS allows asmany classes as required to be defined using scientific and/or localindigenous knowledge (Gregorio, 1998).
LC mapping for a specific year can be carried out relatively easilyusing satellite images; but LC change analysis using multi-temporalimages is more demanding (Knorn et al., 2009). Although Landsatimages from different years provide data at the same spatial reso-lution, the variation in spectral response and heterogeneous envi-ronment may lead to problems. These can be minimized though
Fig. 7. Distribution of land cover classes with elevation in 2010.
Please cite this article in press as: Gilani, H., et al., Decadal land cover changhttp://dx.doi.org/10.1016/j.jenvman.2014.02.014
calibration and analytical manipulation. The study area describedhere lies in a mountainous region, and topographic effects in termsof the earth’s curvature, mountain shadow, and clouds representmajor obstacles that need to be taken into account (Itten andMeyer, 1993). An overall accuracy of 83% was achieved with akappa value of 0.81 (average of producer and user accuracies foreach class) (Table 4). Six of the ten major LC classes had producerand user accuracies higher than 80%, with the accuracy for forestclasses close to 87%. Identifying the small areas of agriculture andsettlements in mountain regions using medium resolution satellitedata is quite challenging, and results obtained for LC classes with asmall area of coverage should be treated with caution. These LCclasses could be mapped and analysed more accurately using veryhigh resolution satellite images. The availability of very high
Fig. 8. Area covered by plantation activities in each of the 20 districts in Bhutan(source MoAF).
e dynamics in Bhutan, Journal of Environmental Management (2014),
H. Gilani et al. / Journal of Environmental Management xxx (2014) 1e10 9
resolution satellite images from the Google Earth Pro platform, anddirect information from numerous points on the ground (groundtruthing), enabled a precise assessment of the LC accuracy.
5. Conclusions
The study described here filled a number of data gaps and opensup possibilities for analysis of LCLU change in the eastern Hima-layas. The growth in forest area in Bhutan from 1990 to 2010 wasmuch higher than the loss of forest, mainly due to the annualplantation activities throughout the country. Agriculture isexpanding very slowly due to the limited population growth. Grassand other land use classes depend highly on season and climaticconditions.
This study provides baseline information as well as informationabout the changes in LC over the past two decades. A number ofapplications and analyses can be executed on these three LC layers.The results can be used to support decision making and naturalresource management planning, e.g. by combining populationdensity, administrative areas, and LC-2010map to identify potentialareas for further plantation at district and sub-district level; and inpreparation of Tier 1 greenhouse gas (GHG) inventories. Landsat 8is already functioning and taking satellite images, and has an ex-pected lifetime of 10 years. Thus in 10 years time it will be possibleto validate simulations based on the current data.
Acknowledgements
This paper has been prepared under SERVIR-HIMALAYA, theNational Aeronautics and Space Administration (NASA) funded byUnited States Agency for International Development (USAID). Wewould like to acknowledge contributions from Ms. KuenzangLhamo and Mr. Younten Phuntsho of Bhutan’s Ministry of Agri-culture and Forest. From the International Centre for IntegratedMountain Development (ICIMOD), we thankMr. Deo Raj Gurung forfacilitation, Mr. Kabir Uddin for technical inputs, and our formercolleague Mr. Salman Asif Siddiqui for his work during the initialstages of the study. We are also very grateful for the review andfeedback by Ms. Kezang Lhamo Dorji which helped improve thewhole study. The findings reported stand as scientific study andobservations of the authors and do not necessarily reflect as theviews of ICIMOD.
References
Abtew, W., Melesse, A., 2013. Crop Yield Estimation Using Remote Sensing andSurface Energy Flux Model, Evaporation and Evapotranspiration. Springer,Netherlands, pp. 161e175.
Agarwal, C., Green, G.M., Grove, J.M., Evans, T.P., Schweik, C.M., 2002. A Review andAssessment of Land-use Change Models: Dynamics of Space, Time, and HumanChoice. United States Department of Agriculture, Northeastern research Stationand Indiana University, Pennsylvania: US, p. 67.
Arino, O., Gross, D., Ranera, F., Bourg, L., Leroy, M., Bicheron, P., Latham, J., DiGregorio, A., Brockman, C., Witt, R., Defourny, P., Vancutsem, C., Herold, M.,Sambale, J., Achard, F., Durieux, L., Plummer, S., Weber, J.L., 2007. GlobCover:ESA service for global land cover from MERIS. In: Geoscience and RemoteSensing Symposium, 2007. IGARSS. IEEE International, pp. 2412e2415.
Bhattarai, K., Conway, D., Yousef, M., 2009. Determinants of deforestation in Nepal’sCentral Development Region. J. Environ. Manag. 91, 471e488.
Blaschke, T., 2010. Object based image analysis for remote sensing. ISPRS J. Photo-gram. Remote Sens. 65, 2e16.
Boggs, G.S., 2010. Assessment of SPOT 5 and QuickBird remotely sensed imagery formapping tree cover in savannas. Int. J. Appl. Earth Observ. Geoinform. 12, 217e224.
Bontemps, S., Defourny, P., Van Bogaert, E., Arino, O., Kalogirou, V., Ramos Perez, J.,2011. GlobCover 2009: Products Description and Validation Report. URL: http://ionia1.esrin.esa.int/docs/GLOBCOVER2009_Validation_Report_2.2.
Buffum, B., Gratzer, G., Tenzin, Y., 2009. Forest grazing and natural regeneration in alate successional broadleaved community forest in Bhutan. Mountain Res.Develop. 29, 30e35.
Please cite this article in press as: Gilani, H., et al., Decadal land cover changhttp://dx.doi.org/10.1016/j.jenvman.2014.02.014
Dorji, M., 2013. Rampant Illegal Logging in Paro Threatens Forest Cover, TheBhutanese. The Bhutanese, Thimphu, Bhutan.
Duveiller, G., Defourny, P., Desclée, B., Mayaux, P., 2008. Deforestation in CentralAfrica e estimates at regional national and landscape levels by advanced pro-cessing of systematically distributed Landsat extracts. Remote Sens. Environ.112, 1969e1981.
FAO, 2010. Global Forest Resources Assessment 2010: Country Report Bhutan. Foodand Agriculture Organization of the United Nations, Rome, Italy, p. 56.
Foley, J.A., DeFries, R., Asner, G.P., Barford, C., Bonan, G., Carpenter, S.R., Chapin, F.S.,Coe, M.T., Daily, G.C., Gibbs, H.K., Helkowski, J.H., Holloway, T., Howard, E.A.,Kucharik, C.J., Monfreda, C., Patz, J.A., Prentice, I.C., Ramankutty, N., Snyder, P.K.,2005. Global consequences of land use. Science 309, 570e574.
Foody, G.M., 2010. Assessing the accuracy of land cover change with imperfectground reference data. Remote Sens. Environ. 114, 2271e2285.
Gregorio, D.A., 1998. Land Cover Classification System (LCCS): Classification Con-cepts and User Manual. Food and Agriculture Organization, Rome, Italy, p. 91.
Gu, Y., Brown, J.F., Verdin, J.P., Wardlow, B., 2007. A five-year analysis of MODIS NDVIand NDWI for grassland drought assessment over the central Great Plains of theUnited States. Geophys. Res. Lett. 34.
Gyelmo, D., 2013. Illegal Timber Felling in Decline but Still a Problem. KuenselOn-line, Thimphu, Bhutan.
Gyeltshen, N., 2013. Wangchhu Will Take up 125 Acres of Private Land. Kuense-lOnline, Thimphu, Bhutan.
Hansen, M.C., Loveland, T.R., 2012. A review of large area monitoring of land coverchange using Landsat data. Remote Sens. Environ. 122, 66e74.
Herold, M., Mayaux, P., Woodcock, C.E., Baccini, A., Schmullius, C., 2008. Somechallenges in global land cover mapping: an assessment of agreement andaccuracy in existing 1 km datasets. Remote Sens. Environ. 112, 2538e2556.
Huang, C., Goward, S.N., Schleeweis, K., Thomas, N., Masek, J.G., Zhu, Z., 2009. Dy-namics of national forests assessed using the Landsat record: case studies ineastern United States. Remote Sens. Environ. 113, 1430e1442.
Itten, K.I., Meyer, P., 1993. Geometric and radiometric correction of TM data ofmountainous forested areas. IEEE Trans. Geosci. Remote Sens. 31, 764e770.
Jin, S., Yang, L., Danielson, P., Homer, C., Fry, J., Xian, G., 2013. A comprehensivechange detection method for updating the national land cover database to circa2011. Remote Sens. Environ. 132, 159e175.
Knorn, J., Rabe, A., Radeloff, V.C., Kuemmerle, T., Kozak, J., Hostert, P., 2009. Landcover mapping of large areas using chain classification of neighboring Landsatsatellite images. Remote Sens. Environ. 113, 957e964.
Kumar, U., Kerle, N., Punia, M., Ramachandra, T.V., 2011. Mining land cover infor-mation using multilayer perceptron and decision tree from MODIS data.J. Indian Soc. Remote Sens. 38, 592e603.
Lambin, E.F., Meyfroidt, P., 2010. Land use transitions: socio-ecological feedbackversus socio-economic change. Land Use Policy 27, 108e118.
Lyle, G., Lewis, M., Ostendorf, B., 2013. Testing the temporal ability of Landsat im-agery and precision agriculture technology to provide high resolution historicalestimates of wheat yield at the farm scale. Remote Sens. 5, 1549e1567.
Mather, A.S., 2007. Recent Asian forest transitions in relation to forest-transitiontheory. Int. For. Rev. 9, 491e502.
MoAF, 2010. National Forest Policy of Bhutan. Ministry of Agriculture and Forests,Royal Government of Bhutan, Thimphu, Bhutan, p. 18.
Myint, S.W., Galletti, C.S., Kaplan, S., Kim, W.K., 2013. Object vs. pixel: a systematicevaluation in urban environments. Geocarto Int., 1e22.
Niraula, R.R., Gilani, H., Pokharel, B.K., Qamer, F.M., 2013. Measuring impacts ofcommunity forestry program through repeat photography and satellite remotesensing in the Dolakha district of Nepal. J. Environ. Manag. 126, 20e29.
Noord, V.H., 2010. Feasibility of REDDþ in Bhutan: a Scoping Study. WatershedManagement Division, Department of Forests and Park Services, Ministry ofAgriculture and Forests, Thimphu, Bhutan, p. 49.
NSB, 2005. http://www.nsb.gov.bt/publication/publications.php?id¼2. NationalStatistics Bureau, Bhutan.
Olofsson, P., Foody, G.M., Stehman, S.V., Woodcock, C.E., 2013. Making better use ofaccuracy data in land change studies: estimating accuracy and area andquantifying uncertainty using stratified estimation. Remote Sens. Environ. 129,122e131.
Potere, D., 2008. Horizontal positional accuracy of Google Earth’s high-resolutionimagery archive. Sensors 8, 7973e7981.
Qamer, F., Abbas, S., Saleem, R., Shehzad, K., Ali, H., Gilani, H., 2012. Forest coverchange assessment in conflict-affected areas of northwest Pakistan: the case ofSwat and Shangla districts. J. Mountain Sci. 9, 297e306.
Ra�si, R., Bodart, C., Stibig, H.-J., Eva, H., Beuchle, R., Carboni, S., Simonetti, D.,Achard, F., 2011. An automated approach for segmenting and classifying a largesample of multi-date Landsat imagery for pan-tropical forest monitoring.Remote Sens. Environ. 115, 3659e3669.
RSPN, 2011. Strategic Plan of Royal Society for Protection of Nature 2011e2015.Royal Society for Protection of Nature (RSPN), Thimphu, Bhutan, p. 52.
Sharma, N.K., Tiwari, P.K., Sood, Y.R., 2013. A comprehensive analysis of strategies,policies and development of hydropower in India: special emphasis on smallhydro power. Renew. Sustain. Energy Rev. 18, 460e470.
Singh, P.S., Bassignana-Khadka, I., Karky, S.B., Sharma, E., 2011. Climate Change inthe Hindu Kush-Himalayas: the State of Current Knowledge. InternationalCentre for Integrated Mountain Development, Kathmandu, Nepal, p. 102.
Townshend, J.R., Masek, J.G., Huang, C., Vermote, E.F., Gao, F., Channan, S.,Sexton, J.O., Feng, M., Narasimhan, R., Kim, D., Song, K., Song, D., Song, X.-P.,Noojipady, P., Tan, B., Hansen, M.C., Li, M., Wolfe, R.E., 2012. Global
e dynamics in Bhutan, Journal of Environmental Management (2014),
H. Gilani et al. / Journal of Environmental Management xxx (2014) 1e1010
characterization and monitoring of forest cover using Landsat data: opportu-nities and challenges. Int. J. Digital Earth 5, 373e397.
Vogelmann, J.E., Xian, G., Homer, C., Tolk, B., 2012. Monitoring gradual ecosystemchange using Landsat time series analyses: case studies in selected forest andrangeland ecosystems. Remote Sens. Environ. 122, 92e105.
Xian, G., Homer, C., Fry, J., 2009. Updating the 2001 national land cover databaseland cover classification to 2006 by using Landsat imagery change detectionmethods. Remote Sens. Environ. 113, 1133e1147.
Please cite this article in press as: Gilani, H., et al., Decadal land cover changhttp://dx.doi.org/10.1016/j.jenvman.2014.02.014
Xin, Q., Olofsson, P., Zhu, Z., Tan, B., Woodcock, C.E., 2013. Toward near real-timemonitoring of forest disturbance by fusion of MODIS and Landsat data.Remote Sens. Environ. 135, 234e247.
Xue, Y., Li, Y., Guang, J., Zhang, X., Guo, J., 2008. Small satellite remote sensing andapplications e history, current and future. Int. J. Remote Sens. 29, 4339e4372.
Yuan, F., Sawaya, K.E., Loeffelholz, B., Bauer, M.E., 2005. Land cover classification andchange analysis of the twin cities (Minnesota) metropolitan area by multi-temporal Landsat remote sensing. Remote Sens. Environ. 98, 317e328.
e dynamics in Bhutan, Journal of Environmental Management (2014),