Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Land use classification and land use change analysis using satellite images in Lombok Island, Indonesia

Land use classification and land use change analysis using satellite images in Lombok Island,... FOREST SCIENCE AND TECHNOLOGY, 2016 VOL. 12, NO. 4, 183191 http://dx.doi.org/10.1080/21580103.2016.1147498 Land use classification and land use change analysis using satellite images in Lombok Island, Indonesia Cheolmin Kim Division of Forest Industry Research, National Institute of Forest Science, Seoul, Republic of Korea ABSTRACT ARTICLE HISTORY Received 19 August 2015 The objective of this study was to classify land use and land cover status and to identify land use Accepted 25 January 2016 changes, especially of deforestation and forest degradation in the past 20 years in Lombok Island using satellite imageries to support REDDC program implementation. Medium scale Landsat MSS, TM, KEYWORDS and ETMC data from 1990 to 2000 at 5 year intervals were used to extract information on land use Change detection; and land use changes. A land use/cover classification system was established, based on the six broad deforestation; forest land use categories of IPCC Good Practice Guidelines. For land use classification, a supervised degradation; land use classification method was applied, and a “binary change mask applied to date 2” algorithm was used classification; Lombok for land use change detection. As of 2010, cropland dominates the land cover of Lombok, comprising 61.4% of total area. Forest is the second dominant land cover class, covering c. 118,365 ha or about 25.8% of the land. Shrubland occupies 7.5% of land area. Forested land in 1990 was estimated at c. 156,900 ha or 34% of the total land area of Lombok. Subsequently, forested land has decreased by 47,363 ha over the past 20 years. This means that, since 1990, 28.6% of forest has been converted to non-forest land use, mostly presumed to be cropland and shrubland. Introduction Coarse spatial resolution optical sensors have been useful for mapping vegetation at the global, continental scale, because Land use and/or land cover is the result of human uses of of large coverage scanning and high frequency in data acqui- land and the interactions of global climate changes on the sition (Langner et al. 2007). Medium resolution satellites, Earth’s surface. Land use and land cover play a major role in such as Landsat TM, have been most frequently used for veg- the carbon cycle by acting as a source and sink of carbon. etation mapping. Mid-resolution satellite images are a practi- Deforestation, afforestation, and re-growth of forest cause cal and effective primary data source especially for REDD the release and sequestering of carbon, thereby affecting monitoring to identify deforestation (Hiepe and Kanamaru atmospheric CO concentrations and increasing the green- 2008). High resolution satellite data are used for validation of house effect (Asner et al. 2005). Regular monitoring and small areas from the results of coarser resolution analysis. assessment of land use and land cover change is therefore Remote sensing image classification is a complex process critical for understanding the extent and impact of such which involves many steps, including the determination of a anthropogenic and natural changes on the Earth at local, land cover classification system, collection of data sources, regional, or global scales (Potapov et al. 2008). selection of a classification algorithm, extraction of thematic Remotely sensed data have been widely used to classify information, and accuracy assessment (Jensen 2005; Lu and land cover and to provide estimates of its corresponding Weng 2007). Technical progress in image classification has area. Remote sensing combined with ground measurements been achieved since the 1990s and a great deal of research have played a key role in determining, with confidence, the has been conducted to classify land cover and monitor forest loss of forest cover since the 1990s (DeFries et al. 2006; loss, especially for tropical forest vegetation (Tucker et al. GOFC-GOLD 2009). The strength of remote sensing is in its 1985; Woodcock et al. 1994; Foody et al. 1996; Hansen et al. ability to provide spatially explicit information and repeated 1996; Kartawinata et al. 2001; Tottrup 2004;Lu 2005). coverage of large areas, especially remote areas that are diffi- For REDD baseline setting, the most appropriate dataset is cult to access otherwise (Lillesand and Kiefer 1999). medium resolution satellite data, such as Landsat TM imag- A variety of satellite data sources are used in classifying ery. With global coverage, the regularly acquired largest his- land use and establishing historical trends of forest changes, torical archive and freely available space-based Earth especially for deforestation and forest degradation (Rose- observations, Landsat imagery is preferred for monitoring nqvist et al. 2003; DeFries et al. 2006; Gibbs et al. 2007). tropical forests in developing countries (Vieira et al. 2003; Developments in sensor technology have allowed the acquisi- Salovaara et al. 2005; Kumar et al. 2010; Li et al. 2011; Pota- tion of a various range of scales ranging from coarse spatial pov et al. 2012; Zhuravleva et al. 2013). These datasets serve a resolution of 1 km (e.g., NOAA AVHRR, MODIS) to key role in establishing regional historical deforestation rates, medium spatial resolution of about 20 m to 30 m (e.g., Land- which is critical for REDD implementation to reduce emis- sat TM, ETMC, SPOT HRV), as well as high resolution of sions from deforestation and forest degradation. Several less than 5 m (e.g., Ikonos, QuickBird, LIDAR, and others). CONTACT Cheolmin Kim [email protected] © 2016 Korean Forest Society 184 C. KIM studies have reported forest loss and change detections in Table 1. Satellite datasets used for the study. Indonesia using remote sensing data. Margono et al. (2012) Acquisition date used Landsat time-series datasets from 1990 to 2010 to quan- Satellite Spatial Primary No image resolution data Supplementary Remarks tify the extent and change of primary forest in Sumatra, Indo- 1 Landsat-4 MSS 80 m 08/08/1987 03/12/1991 nesia. Hansen et al. (2009) have demonstrated a synoptic 2 Landsat-5 TM 30 m 05/26/1995 monitoring of national-scale forest clearing within Indonesia 3 Landsat-7 ETMC 30 m 08/19/2000 05/18/2001 4 Landsat-7 ETMC 30 m 05/13/2005 05/16/2006 SLC-off, gap-filled aggregating Landsat imagery for change interpretation and 5 Landsat-7 ETMC 30 m 03/24/2010 10/15/2009 〃 more coarse data (MODIS and AVHRR) for stratifying Indo- nesia into low, medium, and high change categories. Broichet al. (2011) also examined the use of time-series Land- sat and MODIS imagery for quantifying forest loss in Suma- Satellite images tra and Kalimantan from 2000 to 2005. For peatland Medium scale Landsat MSS, TM, and ETMC satellite imag- degradation and development in Indonesia, Miettinen and eries were used for this study. The study area, Lombok Island, Liew (2010) analyzed Landsat and SPOT satellite images and is located at the position of Path 116/Row 66 of the Landsat revealed that there had been a remarkable reduction and deg- Worldwide Reference System (WRS). Landsat time-series radation of peat swamp in the islands of Sumatra and data from 1990 to 2010 with 5 year intervals were selected for Borneo. extracting information on land use and land cover changes The objective of this study is to analyze land use and land on Lombok Island. The images were downloaded from the cover changes using satellite imageries, especially those US Geological Survey National Center for Earth Resources related to deforestation and forest degradation in the past Observation and Science through the GLOVIS data portal 20 years on Lombok Island. Deforestation is defined as the (http://glovis.usgs.gov). The image files are downloadable in direct human-induced conversion of forest land to non-forest Landsat Level 1 Data Products that standard radiometric and land, including the long-term or permanent loss of forest geometric correction was processed. As each band file is pro- cover (UNFCCC 2009). Forest degradation is a reduction of vided unlayered in GeoTIFF output format, the downloaded the canopy cover or stocking within the forest (FAO 2000; band files were layer stacked in ERDAS Imagine for analysis Defries et al. 2006). Based on the findings of this study, future (USGS 2014). Landsat datasets used for the study are listed in trends of forest changes and projected amount of carbon Table 1. emissions will be established, which are essential for develop- In humid tropical forest environments, such as Indonesia ing a potential REDDC implementation. and other tropical countries, cloud cover is a major problem in working with optical remotely sensed data (Asner 2001; Hansen et al. 2008; Margono et al. 2012). In this study, one or two additional subsidiary images were collected near the Materials and methods date of each time-series image to remove the clouded area. The scenes were combined and the regions with clouds and Study area shadows were substituted by the supplementary datasets to Lombok is an island in West Nusa Tenggara (NTB) province, create an improved image. The time sequential image com- eastern Indonesia (Figure 1). It is a roughly circular island, posites were nominally centered for 1990, 1995, 2000, 2005, with a “tail” (the Sekotong Peninsula) to the southwest. It is and 2010. approximately 70 km across with a total land area of 4738 km . The island’s topography is dominated by Mt. Rinjani, Determination of land use classification system which is located in the central-northern part of the island and rises to 3726 m, making it the second highest volcano in All land classes of interest must be selected and carefully Indonesia. Northern and western Lombok has lower popula- defined to classify remotely sensed data successfully into land tion density with higher forested areas than other areas of the use and land cover categories in the survey area. This requires island. Annual precipitation varies greatly by geographical the use of a classification scheme containing taxonomically location, ranging from 400 mm in the eastern and southern clear definitions of classes. Classes in the system should nor- areas to 4250 mm in the western and northern parts of the mally be mutually exclusive, exhaustive, and hierarchical island. (Jensen 2005). IPCC Good Practice Guidance suggested six Figure 1. Location of Lombok Island in West Nusa Tenggara Province, Indonesia. FOREST SCIENCE AND TECHNOLOGY 185 broad categories for representing land areas within a country: Settlement forest; cropland; grassland; wetland; settlements; and other Settlement comprises all developed land, including areas of land (IPCC 2003). Based on these land use frames, all coun- human habitation and transportation infrastructure. tries are recommended to estimate carbon stocks and emis- sions and removals of greenhouse gases, and to eventually Other report these under the United Nations Framework Conven- This class includes bare soil, rock, ice, and all unmanaged tion on Climate Change (UNFCCC). land areas that do not fall into any of the previous classes. In this study, the land areas are classified as forest (primary forest, secondary forest), shrubland, cropland (paddy field, Image classification and change detection dryland cultivation), coconut plantation, upland grassland, wetland, settlements, and other, through direct fieldwork and For image classification, a supervised classification method by referencing preceding reports (Jaya et al. 2011;Korindo was principally used. Supervised classification usually 2012). Each class is considered sufficiently representative and requires a priori knowledge about the region, where ground includes all land area within Lombok Island, reducing possible truth data are collected for each land use class. After super- overlaps and omissions as far as practicable. The characteris- vised classification, post-classification sorting was performed tics of each land use category are described below. to improve classification results incorporating “if-then” rules (Hutchinson 1982; Cibula and Nyquist 1987; Janssen et al. Primary forest 1990) with slope and elevation data. Classified images were Forest includes all land with woody vegetation consistent then sieved, clumped, and filtered before yielding a final out- with thresholds used to define forest land in the country; put. All image processing activities were performed in land area more than 0.5 ha with trees higher than 5 m and a ERDAS Imagine 9.1. canopy of more than 10% (FAO 2010). Forest land in Lom- In pursuance of multi-date land use change detection, a bok is further sub-divided into primary forest and secondary “binary change mask applied to date 2” algorithm was con- forest. Primary forest in this study is defined as mature or ducted (Jensen et al. 1993; Jensen 2005). This method uses intact forest, where the standing stocks have almost reached two image datasets (i.e., Date 1 and Date 2 data). A tradi- stability. The forest is generally of native tree species, there tional classification is first performed using the Date 1 image. are no clear indications of human interventions, and the eco- Next, one of the bands from both dates of imagery is placed logical processes are not significantly disturbed. in a new dataset. The two-band dataset is analyzed using image differencing to produce a change image file. Then a Secondary forest threshold value is selected to identify areas of “change” and Secondary forest is regenerated forest that has been disturbed “no-change” pixels in the change image file. The change by human activities or natural disasters. Secondary forest image is then recoded into a binary mask file consisting of may include a natural forest with timber extraction, retaining areas that have changed between the two dates. The change artificial gaps in the canopy to 50%60%. Agroforestry and mask is then overlaid onto Date 2 of the analysis and only community forests belong to this kind of forest. those pixels that are detected as having changed are classified in the Date 2 imagery. A traditional post-classification com- Shrubland parison can be applied to yield “from-to” change information Shrubland refers to land with woody vegetation where the from this method. In this study we used the 1995 image as a dominant woody elements are shrubs, bushes, and young base and classified beforehand using a Maximum Likelihood generation trees, generally less than 5 m in height. The latter supervised classification method. Change detection using a appears usually after forest clear-cutting activities without binary change mask was then applied to the 2000 image, crop cultivation. This land cover type can therefore be con- identifying the “change” area and producing a classification sidered as degraded forest land. map of 2000 with change information between 1995 and 2000. This process was subsequently applied to 2005, 2010, Cropland and, again, to the 1990 image (Figure 2). Cropland is arable and tillage land, including rice fields and dryland cultivation areas. Cropland includes land covered Accuracy assessment with temporary crops followed by harvest and a period of bare soil or fallow. Coconut plantations are considered a sub- An error matrix was created for accuracy assessments which category of cropland in Lombok since they have been estab- guarantee the quality of the information derived from lished for estate crop production. remotely sensed data. It is performed by comparing the result created by remote sensing analysis to a reference or ground Upland grassland truth data for selected sample points (Congalton 1991; Foody Upland grassland is an area with herbaceous plant types, but 2002). A random number generator was used to yield ran- without crop cultivation. Trees and shrubs can be present but dom x, y coordinates within the study area. All locations cover is less than 10%. Upland grassland usually appears were then visited in the field or evaluated using Google Earth around the upper elevations of Mt. Rinjani. map service system. An error matrix usually provides detailed assessment of the agreement between the classified results Wetland and reference data, with the information of how the misclas- This class includes areas and lands that are covered or satu- sification happened. For accuracy evaluation, overall classifi- rated by water for all or part of the year. Wetland includes cation accuracy and Kappa coefficient were calculated from reservoirs, rivers, lakes, and streams, either natural or the error matrix. Overall classification accuracy was com- constructed. puted as the total number of correctly classified pixels divided 186 C. KIM Figure 2. Diagram of multi-date image classification and change detection. by the total number of sample points. Meanwhile, Kappa Primary and secondary forests are relatively well distin- coefficient is a measure of overall statistical agreement of an guished in Landsat imagery. On the Landsat TM false color error matrix, which takes non-diagonal elements into composite image, the tones of primary forests appear dark account. Kappa analysis is recognized as a powerful method reddish brown compared to secondary forests, which usually for evaluating a single error matrix for it indicates the proba- show a redder and smoother texture than mature forests. Pri- bility of correct classification after removing the probability mary forests in Lombok are mainly distributed in the remote of accidentally correct classification (Smits et al. 1999; Foody and hilly areas around Mt. Rinjani, while secondary forests 2004). are found at low altitudes near roads and settlements. When attempting to identify agricultural croplands, the results may vary considerably depending on the date of image Results and discussion acquisition, because crops grow and are harvested according to seasonal and annual phenological cycles. Lombok is a Land use classification tropical island with two seasons, a rainy season that begins in Land use and land cover of Lombok Island from 1990 to 2010 November and ceases in March the following year, and a dry are summarized in Table 2. This table is a result of land use season which lasts from the end of the rainy season in March classification of Landsat satellite images. The areas are to October. In sufficient rainfall areas such as western Lom- arranged by year and by land use sub-categories. As of 2010, bok, rice is cultivated in paddy fields from December until cropland (dryland agriculture, paddy field, and estate crop the following July. Rice is often intercropped with cassava, altogether) dominates the land cover of this region, compris- beans, and vegetables. However, in other areas plants that do ing 61.4% of the total area. Forest (primary and secondary) is not require much water, such as corn, peanuts, and tobacco, the second dominant land cover class, covering approxi- are cultivated even in dry season. Such areas are therefore mately 118,369 ha or about 25.8% of the land. Shrubland, classified as paddy fields or dryland agriculture in a satellite occupying 7.5% of the land area, appears around the transi- image depending on the date or season of observation. Dur- tion zone between forested and non-forested lands or along ing the growing season, paddy fields exhibit a pink color in the edge of the Mt. Rinjani crater. Because of their similar Landsat false color images, while dryland cultivation shows spectral reflectance signatures, it was difficult to definitely light brown colors, often leading to confusion with shrub- differentiate shrubland from dryland agriculture on Landsat land. Therefore, comparing the area of paddy field to dryland images. Land use classification maps from 1990 to 2010 are agriculture in Lombok is insignificant, while the sum is shown in Figure 3. implicative. Estate crops, most of which are coconut palm plantations on Lombok Island, constituted 39,119 ha or 8.5% of the total Table 2. Land use classification of Lombok Island from 1990 to 2010 (ha). land area of Lombok. In tropical and subtropical regions Class name 1990 1995 2000 2005 2010 coconut palm is common and provides many necessities for Primary forest 66,433.4 54,880.7 53,139.5 51,114.4 51,110.6 Secondary forest 99,299.1 105,064.2 77,452.3 69,752.1 67,258.0 local livelihoods, such as food, fiber, timber, and fuel. Coco- Shrubland 14,119.2 12,767.3 33,626.5 42,051.6 34,418.6 nut is usually found from sea level to 150 m, but grows up to Dryland agriculture 154,337.2 145,704.5 171,472.2 165,500.1 175,844.4 Paddy field 54,010.5 62,834.2 63,822.5 66,286.6 66,213.5 600 m in elevation near the equator (Chan and Elevitch Estate crop 52,957.7 53,067.6 36,975.5 39,263.1 39,119.3 2006). In Indonesia it is illegal to plant coconut palm trees Grassland 4,382.9 7,682.8 6,314.9 7,161.4 7,158.0 within designated forest areas, so they are generally estab- Wetland 3,439.7 3,329.9 3,328.9 3,346.2 3,346.6 Settlement 3,073.2 7,940.8 7,384.9 8,666.5 8,663.4 lished on private land, either in pure stands or mixed with Other 6,154.0 4,935.1 4,689.8 5,064.6 5,074.1 other tree crops. Coconut is an agricultural estate tree crop Total 458,207.0 458,207.0 458,207.0 458,206.5 458,206.5 like oil palm. Both are woody perennial plants having a more FOREST SCIENCE AND TECHNOLOGY 187 Figure 3. Land use classification of Lombok from 1990 to 2010. or less definite crown, consistent with the threshold for defi- relatively short and cover is sparse, so shrubland exhibits nition of forest, so they are sometimes included within the light red on the Landsat TM false color composite image. forest plantation category (FAO 2001, 2007). From Landsat Classifying shrubland as a distinct subclass in this study is imagery, this class exhibits a light orange color, but may based upon the local context of land use changes. Food and appear similar to secondary forest. In Lombok, most coconut Agriculture Organization (FAO) guidelines generalize land trees are distributed along the coastal areas and often on the cover to forest, other wooded land, and other land uses in slopes of lowland hills. monitoring the world’s forests through the Forest Resources Shrubland is a type of wooded land area covered with Assessment Program (FAO 2010). Here, shrubland is catego- shrubs and intermixed with sprouts, saplings, or bushes. This rized as a subclass of other wooded land, which refers to land class often occurs after forest clear-cutting or appears around not classified as “forest” with a crown cover of 5%10% the edges of the volcanic crater. Shrublands are also found in of trees able to reach a height of 5 m at maturity, or with a arid and semi-arid regions of eastern Lombok, occasionally combined cover of shrubs, bushes, and trees more than 10%. in forest transition areas. These areas are often mixed with The definition of forest by the UNFCCC and FAO include croplands, forest, or other land uses. Vegetation height is areas that are temporarily unstocked as a result of human 188 C. KIM Table 3. Error matrix of land use classification of Lombok Island. errors were due to confusion between secondary forest and Reference data coconut plantation and the low differentiating ability Row between shrubland and dryland agriculture. PF SF SL DA PD EC GL W ST O total Classification Primary forest (PF) 22 5 1 28 Secondary forest (SF) 33 3 2 8 1 47 Land use change and forest loss Shrubland (SL) 2 12 5 19 Dryland agriculture (DA) 1 11 101 5 4 3 125 Forest loss and deforestation rate of Lombok Island from Paddy field (PD) 5 7 22 3 1 38 Estate crop (EC) 1 2 5 20 5 22 1990 to 2000 at 5 year intervals is shown in Figure 4, illustrat- Grassland (GL) 1 2 4 7 ing a steady decrease of forest during the study period. In Wetland (W) 21 21 1990, forested land including primary forest and secondary Settlement (ST) 1 5 6 Other (O) 2 1 3 forest was estimated at c. 165,732 ha or 36% of total land area Column total 22 43 35 121 29 35 4 22 15 1 327 on Lombok. Since then, forested land has decreased by 47,363 ha over the past 20 years, which means that 28.6% of interventions such as harvesting or natural causes, and which 1990 forest has changed to other land uses. The rate of forest are expected to regenerate or return to forest within several loss for the entire study period was 2358.2 ha/y. The highest years. In this context, shrubland classified through image annual rate of deforestation recorded was 3.67% during the interpretation in this study may include certain areas that are period 1995 to 2000, slowing to 0.41% in recent years. There “temporarily unstocked” due to clear-cutting or overexploita- are no obvious reasons for the drastic forest decline and tion (i.e., they are assumed to be non-forest one time but change between the study periods. Presumably, the temporal expected to regenerate and could be included as forest in the rate and spatial extent of forest loss was largely affected by near future). In this study we inferred such unstocked land timber extraction, expansion of agricultural land and urban area as forest degradation for the present time, since these development, and weak governance institutions (Holmes areas suffer structural and functional changes that reduce 2002, Curran et al. 2004). biomass and the capacity of the forest to provide goods and A change matrix for the time period between 1990 and services. However, there remain some difficulties in differen- 2010 was produced by post-classification comparison from tiating such future forests from genuine arid or alpine shrub- the classification results, which yield “from-to” change infor- land in Landsat satellite images. mation identifying where, and how much, change has The results of the accuracy assessment shows that overall occurred (Table 4). As seen in the matrix table, 81.7% of land accuracy was 73.4% and Kappa coefficient was 66.8% covers remained unchanged between the years, since the val- (Table 3). From this table it could be inferred that the major ues reported along the diagonal express the unchanged area. Figure 4. Forest area and deforestation rate of Lombok from 1990 to 2010 at 5 year intervals. Table 4. Land cover transition matrix in Lombok, 1990 to 2010 (ha). Forest Shrubland Cropland Grassland Wetland Settlement Other Total 1990 Forest 113,291.6 22,420.6 25,749.0 3,154.9 21.7 112.2 982.5 165,732.5 Shrubland 1,054.0 5,608.9 5,992.5 894.6 6.0 114.5 448.8 14,119.2 Cropland 3,370.9 4,810.5 245,739.9 770.6 477.1 5,173.2 963.2 261,305.5 Grassland 228.7 1,351.7 413.6 2,302.7 0.3 0.6 85.3 4,382.9 Wetland 13.6 4.4 405.9 0.2 2,437.4 220.7 357.5 3,439.7 Settlement 1.0 1.2 414.3 0.2 7.9 2,627.0 21.7 3,073.2 Other 408.8 221.4 2,462.1 34.7 396.2 415.2 2,215.6 6,154.0 Total 118,368.6 34,418.6 281,177.2 7,158.0 3,346.6 8,663.4 5,074.6 458,207.0 FOREST SCIENCE AND TECHNOLOGY 189 Figure 5. Forest degradation at 5 year intervals in Lombok. It is seen that the increased land areas of c. 19,872 ha in crop- 84.7% of forest degradation took place in primary forest, land mostly came from forest and shrubland classes. Simi- probably attributable to illegal logging (Margonao et al. larly, it is inferred that forests were significantly converted to 2012). It is also notable that vegetation recovery was observed shrubland and cropland during the past 20 years. over the 1691 ha of degraded forest during the period 2005 to When we simply define forest degradation as an area of 2010. However, the recovery does not imply that no logging unstocked forest land (either transitory or persistent) with has occurred during the time, because a number of evidences the secondary forests that have experienced a transition from of illegal logging are still found in the forests. primary forest due to overexploitation or natural causes, The overall pattern of deforestation and forest degradation 22.5% or 37,327 ha of forest in total were degraded over the during 1990 to 2010 is illustrated in Figure 6. The Figure past 20 years. Forest degradation at 5 year intervals is pre- depicts that the forested area in the southwestern parts of the sented in Figure 5. Among them, 54.0% or 20,143 ha of forest island experienced wider expanses of deforestation, while degradation had occurred from 1995 to 2000. Another northern parts around Mt. Rinjani show a pattern of 11,266 ha of forest was degraded by 1995. During this period, unplanned frontier deforestation and forest degradation Figure 6. Deforestation and forest degradation of Lombok Island from 1990 to 2010. 190 C. KIM along the edges of existing forest. This seems to imply that sustainable forest management, and enhancement of forest there was no intensive commercial logging or expansion of carbon stocks. oil palm estates causing large-scale forest breakdowns on the island. Rather, agricultural land expansion and small-scale Disclosure statement forest conversion were the main drivers of forest and land use changes. No potential conflict of interest was reported by the author. Conclusion References Total forest carbon stocks and changes, the major concerns Asner GP. 2001. Cloud cover in Landsat observations of the Brazilian for REDDC, are determined by two important factors: the Amazon. Int J Remote Sens. 22:38553862. total forest area; and the carbon stock densities per hectare of Asner GP, Knapp DE, Broadbest E, Oliviera P, Keller M, Silva J. 2005. forest. Classifying land use categories and identifying land Selective logging in the Brazilian Amazon. Science 310:480482. Broich M, Hansen MC, Potapov P, Adusei B, Lindquist E, Stehman S. use changes are therefore the first and principal step for 2011. Time-series analysis of multi-resolution optical imagery for implementing a REDDC program in any country. In this quantifying forest cover loss in Sumatran and Kalimantan, Indone- context, the purpose of this study was to analyze land use sia. Int J Appl Earth Obs Geoinf. 13:277291. and land cover changes using satellite imageries, especially Chan E, Elevitch CR. 2006. Species profiles for pacific island agroforestry those related to deforestation and forest degradation in the  Cocos nucifera (coconut). Hawaii: Permanent Agriculture Resources. past 20 years on Lombok Island, Indonesia. Cibula WG, Nyquist MO. 1987.Use of topographic and climatological For land use classification, IPCC (2003) suggests adopting models in a geographical data base to improve Landsat MSS classifi- six land use categories: forest land; cropland; grassland; set- cation for Olympic National Park. PE&RS 53:12791287. tlements; wetlands; and other land. We further classified for- Congalton RG. 1991. A review of assessing the accuracy of classification ests into two major sub-categories, primary forest and of remotely sensed data. Remote Sens Environ. 37:3546. Curran LM, Trigg SN, McDonald AK, Astiani D, Hardiono YM, Siregar secondary forest, distinguishing these categories from shrub- P, Caniago I, Kasischke E. 2004. Lowland forest loss in protected land, which often occurs after forest clear-cutting and in for- areas of Indonesian Borneo. Science 303:10001002. est transition areas. Classifying shrubland as a distinct sub- DeFries R, Achard F, Brown S, Herold M, Murdiyarso D, Schlamadinger class is unique to this study. All land area of Lombok Island B, Souza C. 2006. Reducing greenhouse gas emissions from defores- was classified in accordance with these suggested land use tation in developing countries: considerations for monitoring and measuring. GOFC-GOLD report 26. Rome, Italy: Global Terrestrial categories by analyzing Landsat imagery, and a land use Observing System. change matrix from 1990 to 2010 was developed. [FAO] Food and Agriculture Organization of United Nations. 2000.On We categorized shrubland as a separate land use class, and definitions of forest and forest change. Forest Resources Assessment noted the conversion from primary and secondary forest Programme Working Paper 33. FRA 2000. Rome, Italy: FAO. types to shrubland as “forest degradation”, identifying this [FAO] Food and Agriculture Organization of United Nations. 2001. Forest plantation resources, FAO data-sets 1980, 1990, 1995 and change in both spatial and temporal dimensions. However, 2000. FAO Working Paper FP/14. Rome, Italy: FAO. the precise detection of forest degradation was difficult [FAO] Food and Agriculture Organization of United Nations. 2007. through remote sensing analysis when it progresses tempo- Definitional issues related to reducing emissions from deforestation rarily and easily confused with dryland cultivation areas. in developing countries. Forest and Climate Change Working Paper As of 2010, cropland dominates the land cover of this 5. Rome, Italy: FAO. [FAO] Food and Agriculture Organization of United Nations. 2010. region, comprising 61.4% of the total area. Forest is the sec- Global Forest Resources Assessment 2010, main report. FAO ond dominant land cover class, covering c. 118,369 ha or Forestry Paper 163. Rome, Italy: FAO. about 25.8% of the land. Shrubland occupied 7.5% of the Foody GM. 2002. Status of land cover classification accuracy assessment. land area. In 1990, forest was estimated at c. 165,732 ha or Remote Sens Environ. 80:185201. 36% of total land area on Lombok. Since then, forested land Foody GM. 2004. Thematic map comparison: evaluating the statistical significance of difference in classification accuracy. PE&RS has decreased by 47,363 ha over the past 20 years, which 70:627633 means that 28.6% of 1990 forest has been deforested. There Foody GM, Palubinskas G. Lucas RM, Curran PJ, Honzak M. 1996. are no obvious reasons for these drastic forest declines and Identifying terrestrial carbon sinks: classification of successional changes between the study periods. Presumably, the temporal stages in regenerating tropical forest from Landsat TM data. Remote rate and spatial extent of forest loss was largely affected by Sens Environ. 55:205216. Gibbs HK, Brown S, Niels JO, Foley JA. 2007. Monitoring and estimat- timber extraction, expansion of agricultural land and urban ing tropical forest carbon stocks: making REDD a reality. Environ development, and weak governance institutions. Res Lett 2:045023. The overall pattern of deforestation and forest degradation [GOFC-GOLD] Global Observation of Forest and Land Cover Dynam- during 1990 to 2010 was illustrated as well. Figure 6 shows ics. 2009. A sourcebook of methods and procedures for monitoring that the forested area in the southwestern parts of the island and reporting anthropogenic greenhouse gas emissions and remov- als caused by deforestation, gains and losses of carbon stocks in for- experienced wider expanses of deforestation, while the forests ests remaining forests, and forestation. Alberta: GOFC-GOLD around Mt. Rinjani declined along the forest edges, charac- Project Office, Natural Resources Canada. terized by a pattern of unplanned frontier deforestation and Hansen MC, Dubayah R, Defries R. 1996. Classification trees: an alter- forest degradation presumably arose due to agricultural land native to traditional land cover classifiers. Int J Remote Sens. expansion. 17:10751081. Hansen MC, Roy D, Lindquist E, Justice CO, Alstaat A. 2008. A method Based on the findings of this study, future trends of for integrating MODIS and Landsat data for systematic monitoring forest changes and projected carbon emissions will be of forest cover and change in the Congo Basin. Remote Sens Envi- established for REDDC implementation, along with the ron. 112:24952513. development of strategies for reducing deforestation and Hansen MC, Stehman SV, Potapov PV, Pittman KW, Margono AB, forest degradation, additional dimensions of conservation, Stolle F. 2009. Quantifying changes in the rates of forest clearing in FOREST SCIENCE AND TECHNOLOGY 191 Indonesia from 1990 to 2005 using remotely sensed data. Environ deforestation and forest degradation in Sumatra (Indonesia) using Res Lett 4:034001 Landsat time series data sets from 1990 to 2010. Environ Res Lett Hiepe C, Kanamaru H. 2008. Review of literature on monitoring to sup- 7:034010. port REDD. UN-REDD Programme MRV Working Paper 3. Avail- Miettinen J, Liew SC. 2010. Degradation and development of peatlands able from: http://www.unredd.net/ in peninsular Malaysia and in the islands of Sumatra and Borneo Holmes D. 2002. Where have all the forests gone? Washington: The since 1990. Land Degrad Dev. 21:285296. World Bank. Potapov PV, Yaroshenko A, Turubanova S, Dubinin M, Laestadius L, Hutchinson CF. 1982. Techniques for combining Landsat and ancillary Thies C, Aksenov D, Egorov A, Yesipova Y, Glushkov I, et al. 2008. data for digital classification improvement. PE&RS 48:3340. Mapping the world’s intact forest landscapes by remote sensing. Ecol [IPCC] Intergovernmental Panel on Climate Change. 2003. Good Prac- Soc. 13:51. tice Guidance for land use, land-use change and forestry. Kanagawa, Potapov PV, Turubanova S, Hansen MC, Adusei B, Broich M, Altstatt Japan: National Greenhouse Gas Inventories Programme. A, Mane L, Justice CO. 2012. Quantifying forest cover loss in Demo- Janssen LF, Jaarsma J, Van der Linden E. 1990. Integrating topographic cratic Republic of the Congo, 2000-2010, with Landsat ETMC data. data with remote sensing for land-cover classification. PE&RS Remote Sens Environ. 122:106116. 56:15031506. Rosenqvist A, Milne A, Lucas R, Imhoff M, Dobson C. 2003. A review of Jaya INS, Massijaya Y, Puspaningsih N, Purnama ES, Wijaya PA, remote sensing technology in support of Kyoto Protocol. Environ Sci Wulandri R. 2011. Analysis on land-use change using remote sens- Policy 6:441455. ing and GIS technology. Bogor, Indonesia: Lombok REDD project Salovaara KJ, Thessler S, Malik RN, Tuomisto H. 2005. Classification of (KIPCCF). Amazonian primary rain forest vegetation using Landsat ETMC sat- Jensen JR, Cowen DJ, Narumalani S, Althausen JD, Weatherbee O. 1993. ellite imagery. Remote Sens Environ. 97:3951. An evaluation of Coastwatch change detection protocol in South Smits PC, Dellepiane SG, Schowengerdt RA. 1999. Quality assessment of Carolina. PE&RS 59:10391046. image classification algorithms for land-cover mapping: a review and Jensen JR. 2005. Introductory digital image processing: a remote sensing a proposal for a cost-based approach. Int J Remote Sens. perspective. 3rd ed. Bergen (NJ): Pearson Prentice Hall. 20:14611486. Kartawinata K, Riswan S, Gintings AN, Puspitojati T. 2001. An overview Tottrup C. 2004. Improving tropical forest mapping using multi-date of post-extraction secondary forest in Indonesia. J Trop For Sci. Landsat TM data and pre-classification image smoothing. Int J 13:621638. Remote Sens. 25:717730. Korindo. 2012. REDD regional economic development model and Tucker CJ, Townshend JRG, Goff TE. 1985. African land-cover classifi- methodology. Final report on Korea-Indonesian joint project for cation using satellite data. Science 227:369375. adaptation and mitigation of climate changes in forestry. Jakarta, [UNFCCC] United Nations Framework Convention on Climate Change. Indonesia: Korindo. 2009. Decision 4/CP. 15. Monitoring guidance for activities relating to Kumar P, Rani M, Pandey PC, Majumdar A, Nathawat MS. 2010. Moni- reducing emissions from deforestation and forest degradation and the toring of deforestation and forest degradation using remote sensing role of conservation, sustainable management of forests and enhance- and GIS: a case study of Ranchi in Jharkhand (India). Rep Opinion ment of forest carbon stocks in developing countries. Available from: 2:1420. http://unfccc.int/methods/lulucf/items/6917.php Langner A, Miettinen J, Siegert F. 2007. Land cover change 2002-2005 in USGS [Internet]. 2014. Landsat processing details. United States Geo- Borneo and the role of fire derived from MODIS imagery. Glob logical Survey; [updated 2014 Dec 8]. Available from: http://landsat. Change Biol. 13:23292340. usgs.gov/Landsat_Processing_Details.php Lillesand TM, Kiefer RW. 1999. Remote sensing and image interpreta- Vieira ICG, Almeida AS, Davidson EA, Stone TA, Carvalho CJR, Guer- tion. New York: John Wiley and Sons. rero JB. 2003. Classifying successional forests using Landsat spectral Li G, Lu D, Moran E, Hetrick S. 2011. Land-cover classification on a properties and ecological characteristics in eastern Amazonia. moist tropical region of Brazil with Landsat Thematic Mapper imag- Remote Sens Environ. 87:470481. ery. Int J Remote Sens. 32:82078230. Woodcock CE, Collins JB, Gopal S, Jakabhazy VD, Li X, Macomber S, Lu D. 2005. Integration of vegetation inventory data and Landsat TM Ryherd S, Harward VJ, Levitan J, Wu Y, Warbington R. 1994. Map- image for vegetation classification in the western Brazilian Amazon. ping forest vegetation using Landsat TM imagery and a canopy For Ecol Manage. 213:369383. reflectance model. Remote Sens Environ. 50:240254. Lu D, Weng Q. 2007. A survey of image classification methods and tech- Zhuravleva I, Turubanova S, Potapov PV, Hansen MC, Tyukavina niques for improving classification performance. Int J Remote Sens. A, Minnemeyer S, Laporte N, Goetz S, Verbelen F, Thies C. 28:823870. 2013. Satellite-based primary forest degradation assessment in Margono BA, Turubanova S, Zhuravleva I, Potapov PV, Tyukavina A, theDemocraticRepublicofthe Congo, 2000-2010.Environ Res Caccini A, Goetz S, Hansen MC. 2012. Mapping and monitoring Lett 8:024034. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Forest Science and Technology Taylor & Francis

Land use classification and land use change analysis using satellite images in Lombok Island, Indonesia

Kim, Cheolmin
Forest Science and Technology , Volume 12 (4): 9 – Oct 1, 2016
Download PDF
9 pages

Loading next page...
 
/lp/taylor-francis/land-use-classification-and-land-use-change-analysis-using-satellite-1zFav5aPQl

References (49)

Publisher
Taylor & Francis
Copyright
© 2016 Korean Forest Society
ISSN
2158-0715
eISSN
2158-0103
DOI
10.1080/21580103.2016.1147498
Publisher site
See Article on Publisher Site

Abstract

FOREST SCIENCE AND TECHNOLOGY, 2016 VOL. 12, NO. 4, 183191 http://dx.doi.org/10.1080/21580103.2016.1147498 Land use classification and land use change analysis using satellite images in Lombok Island, Indonesia Cheolmin Kim Division of Forest Industry Research, National Institute of Forest Science, Seoul, Republic of Korea ABSTRACT ARTICLE HISTORY Received 19 August 2015 The objective of this study was to classify land use and land cover status and to identify land use Accepted 25 January 2016 changes, especially of deforestation and forest degradation in the past 20 years in Lombok Island using satellite imageries to support REDDC program implementation. Medium scale Landsat MSS, TM, KEYWORDS and ETMC data from 1990 to 2000 at 5 year intervals were used to extract information on land use Change detection; and land use changes. A land use/cover classification system was established, based on the six broad deforestation; forest land use categories of IPCC Good Practice Guidelines. For land use classification, a supervised degradation; land use classification method was applied, and a “binary change mask applied to date 2” algorithm was used classification; Lombok for land use change detection. As of 2010, cropland dominates the land cover of Lombok, comprising 61.4% of total area. Forest is the second dominant land cover class, covering c. 118,365 ha or about 25.8% of the land. Shrubland occupies 7.5% of land area. Forested land in 1990 was estimated at c. 156,900 ha or 34% of the total land area of Lombok. Subsequently, forested land has decreased by 47,363 ha over the past 20 years. This means that, since 1990, 28.6% of forest has been converted to non-forest land use, mostly presumed to be cropland and shrubland. Introduction Coarse spatial resolution optical sensors have been useful for mapping vegetation at the global, continental scale, because Land use and/or land cover is the result of human uses of of large coverage scanning and high frequency in data acqui- land and the interactions of global climate changes on the sition (Langner et al. 2007). Medium resolution satellites, Earth’s surface. Land use and land cover play a major role in such as Landsat TM, have been most frequently used for veg- the carbon cycle by acting as a source and sink of carbon. etation mapping. Mid-resolution satellite images are a practi- Deforestation, afforestation, and re-growth of forest cause cal and effective primary data source especially for REDD the release and sequestering of carbon, thereby affecting monitoring to identify deforestation (Hiepe and Kanamaru atmospheric CO concentrations and increasing the green- 2008). High resolution satellite data are used for validation of house effect (Asner et al. 2005). Regular monitoring and small areas from the results of coarser resolution analysis. assessment of land use and land cover change is therefore Remote sensing image classification is a complex process critical for understanding the extent and impact of such which involves many steps, including the determination of a anthropogenic and natural changes on the Earth at local, land cover classification system, collection of data sources, regional, or global scales (Potapov et al. 2008). selection of a classification algorithm, extraction of thematic Remotely sensed data have been widely used to classify information, and accuracy assessment (Jensen 2005; Lu and land cover and to provide estimates of its corresponding Weng 2007). Technical progress in image classification has area. Remote sensing combined with ground measurements been achieved since the 1990s and a great deal of research have played a key role in determining, with confidence, the has been conducted to classify land cover and monitor forest loss of forest cover since the 1990s (DeFries et al. 2006; loss, especially for tropical forest vegetation (Tucker et al. GOFC-GOLD 2009). The strength of remote sensing is in its 1985; Woodcock et al. 1994; Foody et al. 1996; Hansen et al. ability to provide spatially explicit information and repeated 1996; Kartawinata et al. 2001; Tottrup 2004;Lu 2005). coverage of large areas, especially remote areas that are diffi- For REDD baseline setting, the most appropriate dataset is cult to access otherwise (Lillesand and Kiefer 1999). medium resolution satellite data, such as Landsat TM imag- A variety of satellite data sources are used in classifying ery. With global coverage, the regularly acquired largest his- land use and establishing historical trends of forest changes, torical archive and freely available space-based Earth especially for deforestation and forest degradation (Rose- observations, Landsat imagery is preferred for monitoring nqvist et al. 2003; DeFries et al. 2006; Gibbs et al. 2007). tropical forests in developing countries (Vieira et al. 2003; Developments in sensor technology have allowed the acquisi- Salovaara et al. 2005; Kumar et al. 2010; Li et al. 2011; Pota- tion of a various range of scales ranging from coarse spatial pov et al. 2012; Zhuravleva et al. 2013). These datasets serve a resolution of 1 km (e.g., NOAA AVHRR, MODIS) to key role in establishing regional historical deforestation rates, medium spatial resolution of about 20 m to 30 m (e.g., Land- which is critical for REDD implementation to reduce emis- sat TM, ETMC, SPOT HRV), as well as high resolution of sions from deforestation and forest degradation. Several less than 5 m (e.g., Ikonos, QuickBird, LIDAR, and others). CONTACT Cheolmin Kim [email protected] © 2016 Korean Forest Society 184 C. KIM studies have reported forest loss and change detections in Table 1. Satellite datasets used for the study. Indonesia using remote sensing data. Margono et al. (2012) Acquisition date used Landsat time-series datasets from 1990 to 2010 to quan- Satellite Spatial Primary No image resolution data Supplementary Remarks tify the extent and change of primary forest in Sumatra, Indo- 1 Landsat-4 MSS 80 m 08/08/1987 03/12/1991 nesia. Hansen et al. (2009) have demonstrated a synoptic 2 Landsat-5 TM 30 m 05/26/1995 monitoring of national-scale forest clearing within Indonesia 3 Landsat-7 ETMC 30 m 08/19/2000 05/18/2001 4 Landsat-7 ETMC 30 m 05/13/2005 05/16/2006 SLC-off, gap-filled aggregating Landsat imagery for change interpretation and 5 Landsat-7 ETMC 30 m 03/24/2010 10/15/2009 〃 more coarse data (MODIS and AVHRR) for stratifying Indo- nesia into low, medium, and high change categories. Broichet al. (2011) also examined the use of time-series Land- sat and MODIS imagery for quantifying forest loss in Suma- Satellite images tra and Kalimantan from 2000 to 2005. For peatland Medium scale Landsat MSS, TM, and ETMC satellite imag- degradation and development in Indonesia, Miettinen and eries were used for this study. The study area, Lombok Island, Liew (2010) analyzed Landsat and SPOT satellite images and is located at the position of Path 116/Row 66 of the Landsat revealed that there had been a remarkable reduction and deg- Worldwide Reference System (WRS). Landsat time-series radation of peat swamp in the islands of Sumatra and data from 1990 to 2010 with 5 year intervals were selected for Borneo. extracting information on land use and land cover changes The objective of this study is to analyze land use and land on Lombok Island. The images were downloaded from the cover changes using satellite imageries, especially those US Geological Survey National Center for Earth Resources related to deforestation and forest degradation in the past Observation and Science through the GLOVIS data portal 20 years on Lombok Island. Deforestation is defined as the (http://glovis.usgs.gov). The image files are downloadable in direct human-induced conversion of forest land to non-forest Landsat Level 1 Data Products that standard radiometric and land, including the long-term or permanent loss of forest geometric correction was processed. As each band file is pro- cover (UNFCCC 2009). Forest degradation is a reduction of vided unlayered in GeoTIFF output format, the downloaded the canopy cover or stocking within the forest (FAO 2000; band files were layer stacked in ERDAS Imagine for analysis Defries et al. 2006). Based on the findings of this study, future (USGS 2014). Landsat datasets used for the study are listed in trends of forest changes and projected amount of carbon Table 1. emissions will be established, which are essential for develop- In humid tropical forest environments, such as Indonesia ing a potential REDDC implementation. and other tropical countries, cloud cover is a major problem in working with optical remotely sensed data (Asner 2001; Hansen et al. 2008; Margono et al. 2012). In this study, one or two additional subsidiary images were collected near the Materials and methods date of each time-series image to remove the clouded area. The scenes were combined and the regions with clouds and Study area shadows were substituted by the supplementary datasets to Lombok is an island in West Nusa Tenggara (NTB) province, create an improved image. The time sequential image com- eastern Indonesia (Figure 1). It is a roughly circular island, posites were nominally centered for 1990, 1995, 2000, 2005, with a “tail” (the Sekotong Peninsula) to the southwest. It is and 2010. approximately 70 km across with a total land area of 4738 km . The island’s topography is dominated by Mt. Rinjani, Determination of land use classification system which is located in the central-northern part of the island and rises to 3726 m, making it the second highest volcano in All land classes of interest must be selected and carefully Indonesia. Northern and western Lombok has lower popula- defined to classify remotely sensed data successfully into land tion density with higher forested areas than other areas of the use and land cover categories in the survey area. This requires island. Annual precipitation varies greatly by geographical the use of a classification scheme containing taxonomically location, ranging from 400 mm in the eastern and southern clear definitions of classes. Classes in the system should nor- areas to 4250 mm in the western and northern parts of the mally be mutually exclusive, exhaustive, and hierarchical island. (Jensen 2005). IPCC Good Practice Guidance suggested six Figure 1. Location of Lombok Island in West Nusa Tenggara Province, Indonesia. FOREST SCIENCE AND TECHNOLOGY 185 broad categories for representing land areas within a country: Settlement forest; cropland; grassland; wetland; settlements; and other Settlement comprises all developed land, including areas of land (IPCC 2003). Based on these land use frames, all coun- human habitation and transportation infrastructure. tries are recommended to estimate carbon stocks and emis- sions and removals of greenhouse gases, and to eventually Other report these under the United Nations Framework Conven- This class includes bare soil, rock, ice, and all unmanaged tion on Climate Change (UNFCCC). land areas that do not fall into any of the previous classes. In this study, the land areas are classified as forest (primary forest, secondary forest), shrubland, cropland (paddy field, Image classification and change detection dryland cultivation), coconut plantation, upland grassland, wetland, settlements, and other, through direct fieldwork and For image classification, a supervised classification method by referencing preceding reports (Jaya et al. 2011;Korindo was principally used. Supervised classification usually 2012). Each class is considered sufficiently representative and requires a priori knowledge about the region, where ground includes all land area within Lombok Island, reducing possible truth data are collected for each land use class. After super- overlaps and omissions as far as practicable. The characteris- vised classification, post-classification sorting was performed tics of each land use category are described below. to improve classification results incorporating “if-then” rules (Hutchinson 1982; Cibula and Nyquist 1987; Janssen et al. Primary forest 1990) with slope and elevation data. Classified images were Forest includes all land with woody vegetation consistent then sieved, clumped, and filtered before yielding a final out- with thresholds used to define forest land in the country; put. All image processing activities were performed in land area more than 0.5 ha with trees higher than 5 m and a ERDAS Imagine 9.1. canopy of more than 10% (FAO 2010). Forest land in Lom- In pursuance of multi-date land use change detection, a bok is further sub-divided into primary forest and secondary “binary change mask applied to date 2” algorithm was con- forest. Primary forest in this study is defined as mature or ducted (Jensen et al. 1993; Jensen 2005). This method uses intact forest, where the standing stocks have almost reached two image datasets (i.e., Date 1 and Date 2 data). A tradi- stability. The forest is generally of native tree species, there tional classification is first performed using the Date 1 image. are no clear indications of human interventions, and the eco- Next, one of the bands from both dates of imagery is placed logical processes are not significantly disturbed. in a new dataset. The two-band dataset is analyzed using image differencing to produce a change image file. Then a Secondary forest threshold value is selected to identify areas of “change” and Secondary forest is regenerated forest that has been disturbed “no-change” pixels in the change image file. The change by human activities or natural disasters. Secondary forest image is then recoded into a binary mask file consisting of may include a natural forest with timber extraction, retaining areas that have changed between the two dates. The change artificial gaps in the canopy to 50%60%. Agroforestry and mask is then overlaid onto Date 2 of the analysis and only community forests belong to this kind of forest. those pixels that are detected as having changed are classified in the Date 2 imagery. A traditional post-classification com- Shrubland parison can be applied to yield “from-to” change information Shrubland refers to land with woody vegetation where the from this method. In this study we used the 1995 image as a dominant woody elements are shrubs, bushes, and young base and classified beforehand using a Maximum Likelihood generation trees, generally less than 5 m in height. The latter supervised classification method. Change detection using a appears usually after forest clear-cutting activities without binary change mask was then applied to the 2000 image, crop cultivation. This land cover type can therefore be con- identifying the “change” area and producing a classification sidered as degraded forest land. map of 2000 with change information between 1995 and 2000. This process was subsequently applied to 2005, 2010, Cropland and, again, to the 1990 image (Figure 2). Cropland is arable and tillage land, including rice fields and dryland cultivation areas. Cropland includes land covered Accuracy assessment with temporary crops followed by harvest and a period of bare soil or fallow. Coconut plantations are considered a sub- An error matrix was created for accuracy assessments which category of cropland in Lombok since they have been estab- guarantee the quality of the information derived from lished for estate crop production. remotely sensed data. It is performed by comparing the result created by remote sensing analysis to a reference or ground Upland grassland truth data for selected sample points (Congalton 1991; Foody Upland grassland is an area with herbaceous plant types, but 2002). A random number generator was used to yield ran- without crop cultivation. Trees and shrubs can be present but dom x, y coordinates within the study area. All locations cover is less than 10%. Upland grassland usually appears were then visited in the field or evaluated using Google Earth around the upper elevations of Mt. Rinjani. map service system. An error matrix usually provides detailed assessment of the agreement between the classified results Wetland and reference data, with the information of how the misclas- This class includes areas and lands that are covered or satu- sification happened. For accuracy evaluation, overall classifi- rated by water for all or part of the year. Wetland includes cation accuracy and Kappa coefficient were calculated from reservoirs, rivers, lakes, and streams, either natural or the error matrix. Overall classification accuracy was com- constructed. puted as the total number of correctly classified pixels divided 186 C. KIM Figure 2. Diagram of multi-date image classification and change detection. by the total number of sample points. Meanwhile, Kappa Primary and secondary forests are relatively well distin- coefficient is a measure of overall statistical agreement of an guished in Landsat imagery. On the Landsat TM false color error matrix, which takes non-diagonal elements into composite image, the tones of primary forests appear dark account. Kappa analysis is recognized as a powerful method reddish brown compared to secondary forests, which usually for evaluating a single error matrix for it indicates the proba- show a redder and smoother texture than mature forests. Pri- bility of correct classification after removing the probability mary forests in Lombok are mainly distributed in the remote of accidentally correct classification (Smits et al. 1999; Foody and hilly areas around Mt. Rinjani, while secondary forests 2004). are found at low altitudes near roads and settlements. When attempting to identify agricultural croplands, the results may vary considerably depending on the date of image Results and discussion acquisition, because crops grow and are harvested according to seasonal and annual phenological cycles. Lombok is a Land use classification tropical island with two seasons, a rainy season that begins in Land use and land cover of Lombok Island from 1990 to 2010 November and ceases in March the following year, and a dry are summarized in Table 2. This table is a result of land use season which lasts from the end of the rainy season in March classification of Landsat satellite images. The areas are to October. In sufficient rainfall areas such as western Lom- arranged by year and by land use sub-categories. As of 2010, bok, rice is cultivated in paddy fields from December until cropland (dryland agriculture, paddy field, and estate crop the following July. Rice is often intercropped with cassava, altogether) dominates the land cover of this region, compris- beans, and vegetables. However, in other areas plants that do ing 61.4% of the total area. Forest (primary and secondary) is not require much water, such as corn, peanuts, and tobacco, the second dominant land cover class, covering approxi- are cultivated even in dry season. Such areas are therefore mately 118,369 ha or about 25.8% of the land. Shrubland, classified as paddy fields or dryland agriculture in a satellite occupying 7.5% of the land area, appears around the transi- image depending on the date or season of observation. Dur- tion zone between forested and non-forested lands or along ing the growing season, paddy fields exhibit a pink color in the edge of the Mt. Rinjani crater. Because of their similar Landsat false color images, while dryland cultivation shows spectral reflectance signatures, it was difficult to definitely light brown colors, often leading to confusion with shrub- differentiate shrubland from dryland agriculture on Landsat land. Therefore, comparing the area of paddy field to dryland images. Land use classification maps from 1990 to 2010 are agriculture in Lombok is insignificant, while the sum is shown in Figure 3. implicative. Estate crops, most of which are coconut palm plantations on Lombok Island, constituted 39,119 ha or 8.5% of the total Table 2. Land use classification of Lombok Island from 1990 to 2010 (ha). land area of Lombok. In tropical and subtropical regions Class name 1990 1995 2000 2005 2010 coconut palm is common and provides many necessities for Primary forest 66,433.4 54,880.7 53,139.5 51,114.4 51,110.6 Secondary forest 99,299.1 105,064.2 77,452.3 69,752.1 67,258.0 local livelihoods, such as food, fiber, timber, and fuel. Coco- Shrubland 14,119.2 12,767.3 33,626.5 42,051.6 34,418.6 nut is usually found from sea level to 150 m, but grows up to Dryland agriculture 154,337.2 145,704.5 171,472.2 165,500.1 175,844.4 Paddy field 54,010.5 62,834.2 63,822.5 66,286.6 66,213.5 600 m in elevation near the equator (Chan and Elevitch Estate crop 52,957.7 53,067.6 36,975.5 39,263.1 39,119.3 2006). In Indonesia it is illegal to plant coconut palm trees Grassland 4,382.9 7,682.8 6,314.9 7,161.4 7,158.0 within designated forest areas, so they are generally estab- Wetland 3,439.7 3,329.9 3,328.9 3,346.2 3,346.6 Settlement 3,073.2 7,940.8 7,384.9 8,666.5 8,663.4 lished on private land, either in pure stands or mixed with Other 6,154.0 4,935.1 4,689.8 5,064.6 5,074.1 other tree crops. Coconut is an agricultural estate tree crop Total 458,207.0 458,207.0 458,207.0 458,206.5 458,206.5 like oil palm. Both are woody perennial plants having a more FOREST SCIENCE AND TECHNOLOGY 187 Figure 3. Land use classification of Lombok from 1990 to 2010. or less definite crown, consistent with the threshold for defi- relatively short and cover is sparse, so shrubland exhibits nition of forest, so they are sometimes included within the light red on the Landsat TM false color composite image. forest plantation category (FAO 2001, 2007). From Landsat Classifying shrubland as a distinct subclass in this study is imagery, this class exhibits a light orange color, but may based upon the local context of land use changes. Food and appear similar to secondary forest. In Lombok, most coconut Agriculture Organization (FAO) guidelines generalize land trees are distributed along the coastal areas and often on the cover to forest, other wooded land, and other land uses in slopes of lowland hills. monitoring the world’s forests through the Forest Resources Shrubland is a type of wooded land area covered with Assessment Program (FAO 2010). Here, shrubland is catego- shrubs and intermixed with sprouts, saplings, or bushes. This rized as a subclass of other wooded land, which refers to land class often occurs after forest clear-cutting or appears around not classified as “forest” with a crown cover of 5%10% the edges of the volcanic crater. Shrublands are also found in of trees able to reach a height of 5 m at maturity, or with a arid and semi-arid regions of eastern Lombok, occasionally combined cover of shrubs, bushes, and trees more than 10%. in forest transition areas. These areas are often mixed with The definition of forest by the UNFCCC and FAO include croplands, forest, or other land uses. Vegetation height is areas that are temporarily unstocked as a result of human 188 C. KIM Table 3. Error matrix of land use classification of Lombok Island. errors were due to confusion between secondary forest and Reference data coconut plantation and the low differentiating ability Row between shrubland and dryland agriculture. PF SF SL DA PD EC GL W ST O total Classification Primary forest (PF) 22 5 1 28 Secondary forest (SF) 33 3 2 8 1 47 Land use change and forest loss Shrubland (SL) 2 12 5 19 Dryland agriculture (DA) 1 11 101 5 4 3 125 Forest loss and deforestation rate of Lombok Island from Paddy field (PD) 5 7 22 3 1 38 Estate crop (EC) 1 2 5 20 5 22 1990 to 2000 at 5 year intervals is shown in Figure 4, illustrat- Grassland (GL) 1 2 4 7 ing a steady decrease of forest during the study period. In Wetland (W) 21 21 1990, forested land including primary forest and secondary Settlement (ST) 1 5 6 Other (O) 2 1 3 forest was estimated at c. 165,732 ha or 36% of total land area Column total 22 43 35 121 29 35 4 22 15 1 327 on Lombok. Since then, forested land has decreased by 47,363 ha over the past 20 years, which means that 28.6% of interventions such as harvesting or natural causes, and which 1990 forest has changed to other land uses. The rate of forest are expected to regenerate or return to forest within several loss for the entire study period was 2358.2 ha/y. The highest years. In this context, shrubland classified through image annual rate of deforestation recorded was 3.67% during the interpretation in this study may include certain areas that are period 1995 to 2000, slowing to 0.41% in recent years. There “temporarily unstocked” due to clear-cutting or overexploita- are no obvious reasons for the drastic forest decline and tion (i.e., they are assumed to be non-forest one time but change between the study periods. Presumably, the temporal expected to regenerate and could be included as forest in the rate and spatial extent of forest loss was largely affected by near future). In this study we inferred such unstocked land timber extraction, expansion of agricultural land and urban area as forest degradation for the present time, since these development, and weak governance institutions (Holmes areas suffer structural and functional changes that reduce 2002, Curran et al. 2004). biomass and the capacity of the forest to provide goods and A change matrix for the time period between 1990 and services. However, there remain some difficulties in differen- 2010 was produced by post-classification comparison from tiating such future forests from genuine arid or alpine shrub- the classification results, which yield “from-to” change infor- land in Landsat satellite images. mation identifying where, and how much, change has The results of the accuracy assessment shows that overall occurred (Table 4). As seen in the matrix table, 81.7% of land accuracy was 73.4% and Kappa coefficient was 66.8% covers remained unchanged between the years, since the val- (Table 3). From this table it could be inferred that the major ues reported along the diagonal express the unchanged area. Figure 4. Forest area and deforestation rate of Lombok from 1990 to 2010 at 5 year intervals. Table 4. Land cover transition matrix in Lombok, 1990 to 2010 (ha). Forest Shrubland Cropland Grassland Wetland Settlement Other Total 1990 Forest 113,291.6 22,420.6 25,749.0 3,154.9 21.7 112.2 982.5 165,732.5 Shrubland 1,054.0 5,608.9 5,992.5 894.6 6.0 114.5 448.8 14,119.2 Cropland 3,370.9 4,810.5 245,739.9 770.6 477.1 5,173.2 963.2 261,305.5 Grassland 228.7 1,351.7 413.6 2,302.7 0.3 0.6 85.3 4,382.9 Wetland 13.6 4.4 405.9 0.2 2,437.4 220.7 357.5 3,439.7 Settlement 1.0 1.2 414.3 0.2 7.9 2,627.0 21.7 3,073.2 Other 408.8 221.4 2,462.1 34.7 396.2 415.2 2,215.6 6,154.0 Total 118,368.6 34,418.6 281,177.2 7,158.0 3,346.6 8,663.4 5,074.6 458,207.0 FOREST SCIENCE AND TECHNOLOGY 189 Figure 5. Forest degradation at 5 year intervals in Lombok. It is seen that the increased land areas of c. 19,872 ha in crop- 84.7% of forest degradation took place in primary forest, land mostly came from forest and shrubland classes. Simi- probably attributable to illegal logging (Margonao et al. larly, it is inferred that forests were significantly converted to 2012). It is also notable that vegetation recovery was observed shrubland and cropland during the past 20 years. over the 1691 ha of degraded forest during the period 2005 to When we simply define forest degradation as an area of 2010. However, the recovery does not imply that no logging unstocked forest land (either transitory or persistent) with has occurred during the time, because a number of evidences the secondary forests that have experienced a transition from of illegal logging are still found in the forests. primary forest due to overexploitation or natural causes, The overall pattern of deforestation and forest degradation 22.5% or 37,327 ha of forest in total were degraded over the during 1990 to 2010 is illustrated in Figure 6. The Figure past 20 years. Forest degradation at 5 year intervals is pre- depicts that the forested area in the southwestern parts of the sented in Figure 5. Among them, 54.0% or 20,143 ha of forest island experienced wider expanses of deforestation, while degradation had occurred from 1995 to 2000. Another northern parts around Mt. Rinjani show a pattern of 11,266 ha of forest was degraded by 1995. During this period, unplanned frontier deforestation and forest degradation Figure 6. Deforestation and forest degradation of Lombok Island from 1990 to 2010. 190 C. KIM along the edges of existing forest. This seems to imply that sustainable forest management, and enhancement of forest there was no intensive commercial logging or expansion of carbon stocks. oil palm estates causing large-scale forest breakdowns on the island. Rather, agricultural land expansion and small-scale Disclosure statement forest conversion were the main drivers of forest and land use changes. No potential conflict of interest was reported by the author. Conclusion References Total forest carbon stocks and changes, the major concerns Asner GP. 2001. Cloud cover in Landsat observations of the Brazilian for REDDC, are determined by two important factors: the Amazon. Int J Remote Sens. 22:38553862. total forest area; and the carbon stock densities per hectare of Asner GP, Knapp DE, Broadbest E, Oliviera P, Keller M, Silva J. 2005. forest. Classifying land use categories and identifying land Selective logging in the Brazilian Amazon. Science 310:480482. Broich M, Hansen MC, Potapov P, Adusei B, Lindquist E, Stehman S. use changes are therefore the first and principal step for 2011. Time-series analysis of multi-resolution optical imagery for implementing a REDDC program in any country. In this quantifying forest cover loss in Sumatran and Kalimantan, Indone- context, the purpose of this study was to analyze land use sia. Int J Appl Earth Obs Geoinf. 13:277291. and land cover changes using satellite imageries, especially Chan E, Elevitch CR. 2006. Species profiles for pacific island agroforestry those related to deforestation and forest degradation in the  Cocos nucifera (coconut). Hawaii: Permanent Agriculture Resources. past 20 years on Lombok Island, Indonesia. Cibula WG, Nyquist MO. 1987.Use of topographic and climatological For land use classification, IPCC (2003) suggests adopting models in a geographical data base to improve Landsat MSS classifi- six land use categories: forest land; cropland; grassland; set- cation for Olympic National Park. PE&RS 53:12791287. tlements; wetlands; and other land. We further classified for- Congalton RG. 1991. A review of assessing the accuracy of classification ests into two major sub-categories, primary forest and of remotely sensed data. Remote Sens Environ. 37:3546. Curran LM, Trigg SN, McDonald AK, Astiani D, Hardiono YM, Siregar secondary forest, distinguishing these categories from shrub- P, Caniago I, Kasischke E. 2004. Lowland forest loss in protected land, which often occurs after forest clear-cutting and in for- areas of Indonesian Borneo. Science 303:10001002. est transition areas. Classifying shrubland as a distinct sub- DeFries R, Achard F, Brown S, Herold M, Murdiyarso D, Schlamadinger class is unique to this study. All land area of Lombok Island B, Souza C. 2006. Reducing greenhouse gas emissions from defores- was classified in accordance with these suggested land use tation in developing countries: considerations for monitoring and measuring. GOFC-GOLD report 26. Rome, Italy: Global Terrestrial categories by analyzing Landsat imagery, and a land use Observing System. change matrix from 1990 to 2010 was developed. [FAO] Food and Agriculture Organization of United Nations. 2000.On We categorized shrubland as a separate land use class, and definitions of forest and forest change. Forest Resources Assessment noted the conversion from primary and secondary forest Programme Working Paper 33. FRA 2000. Rome, Italy: FAO. types to shrubland as “forest degradation”, identifying this [FAO] Food and Agriculture Organization of United Nations. 2001. Forest plantation resources, FAO data-sets 1980, 1990, 1995 and change in both spatial and temporal dimensions. However, 2000. FAO Working Paper FP/14. Rome, Italy: FAO. the precise detection of forest degradation was difficult [FAO] Food and Agriculture Organization of United Nations. 2007. through remote sensing analysis when it progresses tempo- Definitional issues related to reducing emissions from deforestation rarily and easily confused with dryland cultivation areas. in developing countries. Forest and Climate Change Working Paper As of 2010, cropland dominates the land cover of this 5. Rome, Italy: FAO. [FAO] Food and Agriculture Organization of United Nations. 2010. region, comprising 61.4% of the total area. Forest is the sec- Global Forest Resources Assessment 2010, main report. FAO ond dominant land cover class, covering c. 118,369 ha or Forestry Paper 163. Rome, Italy: FAO. about 25.8% of the land. Shrubland occupied 7.5% of the Foody GM. 2002. Status of land cover classification accuracy assessment. land area. In 1990, forest was estimated at c. 165,732 ha or Remote Sens Environ. 80:185201. 36% of total land area on Lombok. Since then, forested land Foody GM. 2004. Thematic map comparison: evaluating the statistical significance of difference in classification accuracy. PE&RS has decreased by 47,363 ha over the past 20 years, which 70:627633 means that 28.6% of 1990 forest has been deforested. There Foody GM, Palubinskas G. Lucas RM, Curran PJ, Honzak M. 1996. are no obvious reasons for these drastic forest declines and Identifying terrestrial carbon sinks: classification of successional changes between the study periods. Presumably, the temporal stages in regenerating tropical forest from Landsat TM data. Remote rate and spatial extent of forest loss was largely affected by Sens Environ. 55:205216. Gibbs HK, Brown S, Niels JO, Foley JA. 2007. Monitoring and estimat- timber extraction, expansion of agricultural land and urban ing tropical forest carbon stocks: making REDD a reality. Environ development, and weak governance institutions. Res Lett 2:045023. The overall pattern of deforestation and forest degradation [GOFC-GOLD] Global Observation of Forest and Land Cover Dynam- during 1990 to 2010 was illustrated as well. Figure 6 shows ics. 2009. A sourcebook of methods and procedures for monitoring that the forested area in the southwestern parts of the island and reporting anthropogenic greenhouse gas emissions and remov- als caused by deforestation, gains and losses of carbon stocks in for- experienced wider expanses of deforestation, while the forests ests remaining forests, and forestation. Alberta: GOFC-GOLD around Mt. Rinjani declined along the forest edges, charac- Project Office, Natural Resources Canada. terized by a pattern of unplanned frontier deforestation and Hansen MC, Dubayah R, Defries R. 1996. Classification trees: an alter- forest degradation presumably arose due to agricultural land native to traditional land cover classifiers. Int J Remote Sens. expansion. 17:10751081. Hansen MC, Roy D, Lindquist E, Justice CO, Alstaat A. 2008. A method Based on the findings of this study, future trends of for integrating MODIS and Landsat data for systematic monitoring forest changes and projected carbon emissions will be of forest cover and change in the Congo Basin. Remote Sens Envi- established for REDDC implementation, along with the ron. 112:24952513. development of strategies for reducing deforestation and Hansen MC, Stehman SV, Potapov PV, Pittman KW, Margono AB, forest degradation, additional dimensions of conservation, Stolle F. 2009. Quantifying changes in the rates of forest clearing in FOREST SCIENCE AND TECHNOLOGY 191 Indonesia from 1990 to 2005 using remotely sensed data. Environ deforestation and forest degradation in Sumatra (Indonesia) using Res Lett 4:034001 Landsat time series data sets from 1990 to 2010. Environ Res Lett Hiepe C, Kanamaru H. 2008. Review of literature on monitoring to sup- 7:034010. port REDD. UN-REDD Programme MRV Working Paper 3. Avail- Miettinen J, Liew SC. 2010. Degradation and development of peatlands able from: http://www.unredd.net/ in peninsular Malaysia and in the islands of Sumatra and Borneo Holmes D. 2002. Where have all the forests gone? Washington: The since 1990. Land Degrad Dev. 21:285296. World Bank. Potapov PV, Yaroshenko A, Turubanova S, Dubinin M, Laestadius L, Hutchinson CF. 1982. Techniques for combining Landsat and ancillary Thies C, Aksenov D, Egorov A, Yesipova Y, Glushkov I, et al. 2008. data for digital classification improvement. PE&RS 48:3340. Mapping the world’s intact forest landscapes by remote sensing. Ecol [IPCC] Intergovernmental Panel on Climate Change. 2003. Good Prac- Soc. 13:51. tice Guidance for land use, land-use change and forestry. Kanagawa, Potapov PV, Turubanova S, Hansen MC, Adusei B, Broich M, Altstatt Japan: National Greenhouse Gas Inventories Programme. A, Mane L, Justice CO. 2012. Quantifying forest cover loss in Demo- Janssen LF, Jaarsma J, Van der Linden E. 1990. Integrating topographic cratic Republic of the Congo, 2000-2010, with Landsat ETMC data. data with remote sensing for land-cover classification. PE&RS Remote Sens Environ. 122:106116. 56:15031506. Rosenqvist A, Milne A, Lucas R, Imhoff M, Dobson C. 2003. A review of Jaya INS, Massijaya Y, Puspaningsih N, Purnama ES, Wijaya PA, remote sensing technology in support of Kyoto Protocol. Environ Sci Wulandri R. 2011. Analysis on land-use change using remote sens- Policy 6:441455. ing and GIS technology. Bogor, Indonesia: Lombok REDD project Salovaara KJ, Thessler S, Malik RN, Tuomisto H. 2005. Classification of (KIPCCF). Amazonian primary rain forest vegetation using Landsat ETMC sat- Jensen JR, Cowen DJ, Narumalani S, Althausen JD, Weatherbee O. 1993. ellite imagery. Remote Sens Environ. 97:3951. An evaluation of Coastwatch change detection protocol in South Smits PC, Dellepiane SG, Schowengerdt RA. 1999. Quality assessment of Carolina. PE&RS 59:10391046. image classification algorithms for land-cover mapping: a review and Jensen JR. 2005. Introductory digital image processing: a remote sensing a proposal for a cost-based approach. Int J Remote Sens. perspective. 3rd ed. Bergen (NJ): Pearson Prentice Hall. 20:14611486. Kartawinata K, Riswan S, Gintings AN, Puspitojati T. 2001. An overview Tottrup C. 2004. Improving tropical forest mapping using multi-date of post-extraction secondary forest in Indonesia. J Trop For Sci. Landsat TM data and pre-classification image smoothing. Int J 13:621638. Remote Sens. 25:717730. Korindo. 2012. REDD regional economic development model and Tucker CJ, Townshend JRG, Goff TE. 1985. African land-cover classifi- methodology. Final report on Korea-Indonesian joint project for cation using satellite data. Science 227:369375. adaptation and mitigation of climate changes in forestry. Jakarta, [UNFCCC] United Nations Framework Convention on Climate Change. Indonesia: Korindo. 2009. Decision 4/CP. 15. Monitoring guidance for activities relating to Kumar P, Rani M, Pandey PC, Majumdar A, Nathawat MS. 2010. Moni- reducing emissions from deforestation and forest degradation and the toring of deforestation and forest degradation using remote sensing role of conservation, sustainable management of forests and enhance- and GIS: a case study of Ranchi in Jharkhand (India). Rep Opinion ment of forest carbon stocks in developing countries. Available from: 2:1420. http://unfccc.int/methods/lulucf/items/6917.php Langner A, Miettinen J, Siegert F. 2007. Land cover change 2002-2005 in USGS [Internet]. 2014. Landsat processing details. United States Geo- Borneo and the role of fire derived from MODIS imagery. Glob logical Survey; [updated 2014 Dec 8]. Available from: http://landsat. Change Biol. 13:23292340. usgs.gov/Landsat_Processing_Details.php Lillesand TM, Kiefer RW. 1999. Remote sensing and image interpreta- Vieira ICG, Almeida AS, Davidson EA, Stone TA, Carvalho CJR, Guer- tion. New York: John Wiley and Sons. rero JB. 2003. Classifying successional forests using Landsat spectral Li G, Lu D, Moran E, Hetrick S. 2011. Land-cover classification on a properties and ecological characteristics in eastern Amazonia. moist tropical region of Brazil with Landsat Thematic Mapper imag- Remote Sens Environ. 87:470481. ery. Int J Remote Sens. 32:82078230. Woodcock CE, Collins JB, Gopal S, Jakabhazy VD, Li X, Macomber S, Lu D. 2005. Integration of vegetation inventory data and Landsat TM Ryherd S, Harward VJ, Levitan J, Wu Y, Warbington R. 1994. Map- image for vegetation classification in the western Brazilian Amazon. ping forest vegetation using Landsat TM imagery and a canopy For Ecol Manage. 213:369383. reflectance model. Remote Sens Environ. 50:240254. Lu D, Weng Q. 2007. A survey of image classification methods and tech- Zhuravleva I, Turubanova S, Potapov PV, Hansen MC, Tyukavina niques for improving classification performance. Int J Remote Sens. A, Minnemeyer S, Laporte N, Goetz S, Verbelen F, Thies C. 28:823870. 2013. Satellite-based primary forest degradation assessment in Margono BA, Turubanova S, Zhuravleva I, Potapov PV, Tyukavina A, theDemocraticRepublicofthe Congo, 2000-2010.Environ Res Caccini A, Goetz S, Hansen MC. 2012. Mapping and monitoring Lett 8:024034.

Journal

Forest Science and Technology – Taylor & Francis

Published: Oct 1, 2016

Keywords: Change detection; deforestation; forest degradation; land use classification; Lombok

There are no references for this article.