Scholarly article on topic 'Identifying Change Trajectory over the Sumatra's Forestlands Using Moderate Image Resolution Imagery'

Identifying Change Trajectory over the Sumatra's Forestlands Using Moderate Image Resolution Imagery Academic research paper on "Earth and related environmental sciences"

CC BY-NC-ND
0
0
Share paper
Keywords
{"forest cover change" / "vegetation pattern change" / MODIS / Sumatra}

Abstract of research paper on Earth and related environmental sciences, author of scientific article — Yudi Setiawan, M. Irfansyah Lubis, Sri Malahayati Yusuf, Lilik Budi Prasetyo

Abstract Land use land cover change (LULCC) is the key to a wide range of regional and global environmental issues including land degradation, loss of biodiversity, climate change, food security, human vulnerability and environmental sustainability. Recently, the awareness on the LULCC has increased considerably in tropical regions, especially in related to the change in forestlands and other highly vulnerable areas such as the tropical peat swamp area. The high rate of forest conversion as well as agricultural development in inappropriate areas of Indonesia, particularly in Sumatra, has attracted international environmental concern as scientists further demonstrate the relationship between these changes and global climate change, biodiversity and other ecosystem services. In this study, we applied spatial analysis to investigate the feasibility in using long-term satellite datasets for detecting and quantifying the forest cover change in Sumatra. Then, we systematically identified the change trajectory of those changes. The results indicated several types of change such as agricultural development, forest cover change, either by forest fire or human factors, and temporary change in forest plantation area. Each change category has specific change mechanisms or processes, which then reveals the specific spatial model for each type.

Academic research paper on topic "Identifying Change Trajectory over the Sumatra's Forestlands Using Moderate Image Resolution Imagery"

CrossMark

Available online at www.sciencedirect.com

ScienceDirect

Procedia Environmental Sciences 24 (2015) 189 - 198

The 1st International Symposium on LAPAN-IPB Satellite for Food Security and Environmental

Monitoring

Identifying change trajectory over the Sumatra's forestlands using moderate image resolution imagery

Yudi Setiawana*, M. Irfansyah Lubisa, Sri Malahayati Yusuf, Lilik Budi Prasetyob

aCenter for Environmental Research, Bogor Agriculutral University, Kampus IPB Darmaga, Bogor, 16680, Indonesia bDepartment of Forest Resources Conservation and Ecotourism, Bogor Agriculutral University, Kampus IPB Darmaga, Bogor, Indonesia

Abstract

Land use land cover change (LULCC) is the key to a wide range of regional and global environmental issues including land degradation, loss of biodiversity, climate change, food security, human vulnerability and environmental sustainability. Recently, the awareness on the LULCC has increased considerably in tropical regions, especially in related to the change in forestlands and other highly vulnerable areas such as the tropical peat swamp area. The high rate of forest conversion as well as agricultural development in inappropriate areas of Indonesia, particularly in Sumatra, has attracted international environmental concern as scientists further demonstrate the relationship between these changes and global climate change, biodiversity and other ecosystem services. In this study, we applied spatial analysis to investigate the feasibility in using long-term satellite datasets for detecting and quantifying the forest cover change in Sumatra. Then, we systematically identified the change trajectory of those changes. The results indicated several types of change such as agricultural development, forest cover change, either by forest fire or human factors, and temporary change in forest plantation area. Each change category has specific change mechanisms or processes, which then reveals the specific spatial model for each type. © 2015TheAuthors.PublishedbyElsevierB.V Thisis an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Selection and peer-review under responsibility of the LISAT-FSEM Symposium Committee Keywords: forest cover change; vegetation pattern change; MODIS, Sumatra

* Corresponding author. Tel.:+62-251-8621262; fax: +62-251-8622134. E-mail address: setiawan.yudi@apps.ipb.ac.id

1878-0296 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Selection and peer-review under responsibility of the LISAT-FSEM Symposium Committee doi: 10.1016/j.proenv.2015.03.025

1. Introduction

Information on land use land cover (LULC) and their changes are essential for many aspects of global environmental research. Understanding the role of land-use in regional and global environmental changes requires a historical reconstruction of past change conversions and projection of future trends [1]. Land use change is the result of the complex interactions between human and biophysical-environment, which affect a wide range of temporal and spatial scales. Such changes can be measured and analyzed, including determination of their driving factors; however, it is difficult to aggregate the change at a big island scale (regional scale), e.g. Sumatra Island, Indonesia. At that regional-scale, the aggregation level of the data analysis is related to the simplification of the variability over geographic and socio-economic conditions [2]. Meanwhile, developing the spatial model which can identify critical areas of land use changes and provide insight on the changes pattern is a challenge in the modeling of regional land use changes.

Land use changes, especially the changes in forestlands, are driven by many and diversified underlying processes [3, 4]. Anthropogenic drivers, e.g. demographic and socio-economic conditions, and biophysical constraints such as: soil, climate and topography, determine the spatial pattern of land use [5, 6], and their relationships are often used in the spatial models of deforestation [7, 8]. The spatial integration of socio-economic and biophysical-environmental data, which provides a mechanism to explore the relationship between human activities, landscape conditions and forest cover changes, is an important step to improve understanding of the change in forestland and its future role [9].

In more complex Sumatra's forestlands, forest cover change is the result of many, non-linear, interactions between socio-economic and cultural conditions, biophysical constraints and land use history. Many of the biophysical processes related to this change can be recognized through a change in the long-term vegetation dynamics pattern [10].

This study examines the feasibility of using long-term satellite datasets for detecting and quantifying the forest cover change in Sumatra. then, the change trajectory of those changes will be identified systematically.

2. Methodology

2.1. Study Area

This study encompassed all of the main island of Sumatra which is located on the southern rim of the Indonesian archipelago (upper left corner: 5,613°N 90,164°E; lower right corner: 5,859°S 105,501°E) and has an area of 473,481 km2 with a current population of almost 50 million (53 million administratively, as Riau Islands and Bangka-Belitung Islands are included). The island is administratively divided into 8 provinces, namely Aceh, North Sumatra, West Sumatra, Riau, Jambi, Bengkulu, South Sumatra and Lampung (Figure 1).

On Sumatra, rainfall distribution is strongly influenced by the Barisan Mountains from Aceh to Lampung. These lie very close to the west coast and separate a narrow and wet coastal strip to the west from broad and relatively dry plains to the east; they also enclose several inter-montane basins lying in pronounced rain-shadows. Mean annual rainfall on Sumatra is mostly less than 3,000 mm, falling below 1,500 mm on the northeast coast and below 2,000 mm in several inter-montane basins [11].

2.2. Data

a. Satellite Image

The MODIS vegetation indices (MODIS VI) have some advantages in providing basic information related to vegetation dynamics on the land surface [12]. Vegetation index (VI) is commonly used to measure reliable spatial and temporal inter-comparison of terrestrial photosynthetic activity, in other words, it is a measure of vegetation greenness. We collected the MODIS-based Vegetation Index product composited over 16-day intervals from January 2001 to December 2012, which is provided by the U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center, available on https://lpdaac.usgs.gov/.

Fig.1. Study site

The MODIS VI composited 16-day product called "MODIS MOD13Q1" is available for alternatively packaged MODIS data and addressing a specific need for real-time monitoring and historical trend analysis. In this dataset, to select the best VI value over 16-day intervals, the maximum value composited (MVC) algorithm is applied to the surface reflectance data and incorporated with data of band quality, negative surface reflectance, cloud mask, view angle, and sun angle. There are 23 composites produced each year covering the clear-sky land surface, totally 276 time series during 12 years (2001-2012).

Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) are commonly used to measure reliable spatial and temporal inter-comparisons of terrestrial photosynthetic activity and canopy structural variations [13]. Whereas the NDVI is chlorophyll sensitive, the EVI is more responsive to canopy structural variations, including leaf area index (LAI), canopy type, plant physiognomy, and canopy architecture [14]. The EVI was developed to optimize the vegetation signal with improved sensitivity in high biomass regions and improved vegetation monitoring through a de-coupling of the canopy background signal and a reduction in atmosphere influences [13]. The MODIS EVI is embedded in the MOD13Q1 product. The MODIS Land Discipline Group [15] has developed the EVI for use with MODIS data following this equation below:

EVI = G-

' Pred

Pnir + C\Pred - C2Pblue + L

where, p nir, p red and p blue are the remote sensing reflectances in the NIR, red and blue, respectively, L is a soil adjustment factor, and C1 and C2 describe the use of the blue band in correction of the red band for atmospheric aerosol scattering. The coefficients, C1, C2, and L, are empirically determined as 6.0, 7.5, and 1.0, respectively. G is a gain factor set to 2.5.

MODIS EVI data were extracted from the MODIS VI product (MOD13Q1) using the MODIS Reprojection Tool (MRT) and the selected output format was GeoTIFF and coordinate system was geographic coordinate systems (GCS) on datum World Geodetic System of 1984 (WGS-84). The main island of Sumatra is covered by four MODIS tiles: h27v08, h27v09, h28v08, and h28v09 (total image is 1932 scenes).

b. Land Cover Map

Time series land cover maps provided by the Ministry of Forestry (MoF); 2000, 2003, 2006, and 2009 were used as a reference to select pixels of land use/cover change (Figure 2). These land cover maps were generated based on interpretation of Landsat 5 TM/7 ETM+, and performed by visual classification. This classification was performed by digitizing polygons based on image-interpretation with minimum mapping unit (MMU) defined by means of the smallest polygon identified.

The MoF is concerned primarily with natural vegetation categories. Using the thematic baseline forestry and designation of forestry-aquatic area maps for 30 provinces at the scale of 1:250,000 [16], delineated polygons for those classes could be labeled with two codes, representing categories un-disturbed by human activities (primary type) and those disturbed (secondary type). Finally, the satellite data was classified into 23 land cover types (Table 1). The 23 categories seem sufficient to represent land cover types in Indonesia. This is near the maximum number of categories that could be discriminated effectively; identification of other categories would likely be difficult to perform.

In addition, although the land cover map provided by the MoF was based on the Landsat ETM+ imagery, for which one pixel of image represents 30m x 30m of the earth surface, the product was generalized based on the smallest polygons that could be identified by visual interpretation. As a result, the MoF classification resolution was effectively coarser than a spectral classification using the Landsat data.

2.3. Image Processing

a. Training sample sets

Previous studies have derived land cover maps from NDVI at continental and regional scale based on phenological differences among cover types as reflected in temporal profiles of NDVI [17, 18]. Similar assumptions were used in this study; those locations which display similar temporal patterns are assumed having similar land use types. The land use types are, in turn inferred to have relatively homogeneous characteristics such as fractions of vegetation covers and amount of biomass. The use of the long time series (2001-2012) of EVI was based on the

Fig.2. Land cover map produced periodically by Ministry of Forestry

hypothesis that the land use can be distinguished based on land cover dynamics because consistent land use has a typical, distinct and repeated temporal pattern of vegetation index inter-annually [19].

Table 1. The 23 categories of land cover type by the Ministry of Forestry

No Land cover type

Description

1 Primary dry land forest

2 Secondary dry land forest

3 Primary swamp forest

4 Secondary swamp forest

5 Primary mangrove forest

6 Secondary mangrove forest

7 Bush/slash

8 Swamp

9 Swamp bush

10 Savannah

11 Industrial forest plantation

12 Plantation

13 Dry land agriculture

14 Mixed dry land agriculture

15 Rice land

16 Fishpond

17 Bare land

18 Transmigration

19 Mining

20 Airport area

21 Built-up area / Housing

22 Water body

23 Cloud

Forest in the lowlands, hills and mountains, where there was no logging activity.

Forest in the lowlands, hills and mountains, where logging had occurred.

Forest in swampy areas, including peat swamps without any signs of logging activity.

Forest in swampy areas, where logging had occurred.

Mangrove forests and palm around the coast that have not been logged.

Mangrove forests and palm that have been logged, as revealed by the pattern of grooves.

Natural plants with multiple stems and canopy cover is greater than agriculture but was is not a forested area

Vegetation or swamp area that was not forested swamp

Bush / shrub converted from forest in swampy areas, including bush in swamp areas that will be converted into pond, forest plantation, or plantation forests

Natural vegetation non-forest, grassland with few trees

Large-scale of agriculture in which a crop plantation is planted

Plantation area covered by plantation crops. Plantation areas that are bare land or shrubs are classified according to those land covers (bare land or bush).

Agricultural activities on dry lands, i.e., mixed garden and uplands, include agricultural crops, fruit trees.

Agricultural activities on dry land, which is mixed with bushes or shrubs. Agricultural area in wetlands, planted with rice and characterized by a ridge pattern Fish farming area along the shore.

Area without vegetation (rock outcrops and mountain peaks, volcanic craters, sand)

Settlement area including agricultural land, when its not classified as agricultural land.

Area without vegetation, open area

Airport area which could be identified by its runway

Settlement areas, rural areas, ports, airports, etc. industries.

Water features, including sea, rivers, lakes, reservoirs, coral reefs and sea-grass

An area covered by clouds.

b. Change detection on vegetation dynamics

The change of vegetation dynamics is recognized using a distance of annual vegetation index values from two different time series data, which computed for all pixels and included all consecutive study years during periods from 2001 to 2012 by a function shown in the equation below [20]:

Nk 2 Ni 2 dk,l = ^-I Mfc _ V-new I + w-I № ~ Vnew I

™new ™new (2)

where k and I are two successive time series data, and dkj is the distance between index values of the two successive patterns of k and I data, Nh Ni is the number of observations in k and I data, Nnew is the number of observation of the two pattern of k and i data (Nnew = N + Ni) and ¡dh is the mean of vegetation index values in k and i data, ^new is the mean of index values of the two pattern of k and i data.

Figure 3 illustrates how a distance of index value in two successive time series data is used to detect the change of forest cover (based on vegetation pattern change).

2005 2006

-1-1—--1-1-1---1-1-1-•-1-1-1-•-1-n-T-!-r

0 50 100 150 200 250

Layer (DOY 2001*2012)

Fig. 3. Detecting change pattern using distance of two successive years in a forestland (e.g. the changes during 2005-2006)

These significant change patterns and their trajectories were identified based on their temporal profile, a corresponding series MoF's land cover maps or detailed information which was obtained using Google Earth.

3. Results and Discussion

3.1. Change Detection of Temporal Vegetation Dynamics

Characterizing the long-term vegetation dynamics in this research provides information about the change process in the forestlands, including how and when the change is occurred accurately, and the size of these affected areas. Some significant forest cover changes in the vulnerable areas (e.g. tropical peat soil area) as well as deforestation and forest degradation are identified systematically.

Regarding the vegetation pattern changes that can be detected by the distance comparison of EVI, there are many significant patterns indicated the changes in forestland. Figure 4 shows the patterns for each period from 2001 to 2012. Moreover, based on land cover type classified by the MoF, several change trajectories were identified, for example: forest mixed bushes were converted to barren lands through burning as well as logging activities, which was planted and converted to upland; an upland which was converted into infrastructure/settlement, the area which was first converted to barren land (land clearing/preparation); change pattern of agricultural development due to crop planting in barren land; and re-vegetation processes occurred in an area affected by management activity, where forest (plantation forest) changed to open land because of the harvesting and the later tree planting process.

In many cases, the pattern changes could indicate a continuation change occurred after the forest cover loss, where then the change leads to two possible change pathways, namely: developed to be agriculture lands (plantations) or new vegetation start to grow, either secondary forest or bushes. Although these pathways are not consistently equal with the change of uses in some locations, but the change of these land surface will be interrelated to a specific ecosystem.

The distribution of the changes of Sumatra's forestlands that have been detected from this study is shown in Figure 5. The detail of landscape changes in Riau is given in Figure 6.

Extending the themes of change pathways, trajectories of forest cover change in the production and protected (conservation) forest shall be obtained. The abrupt and gradual change occurred in the production forest indicates that land management caused a change in vegetation pattern dynamics in this forestland. Meanwhile, in the protected forest lands, the change patterns are driven mainly by forest conversion activities and natural phenomena, such as forest fire.

3.2. Accuracy Assessment

Furthermore, the performance of this change detection method was evaluated by 10,000 reference points which revealed an overall accuracy of 80.10%. Comparison of the accuracy in change detection among those forest types

are varied. Forest plantation area had the greatest overall accuracy (94.02%), followed by swamp forests (87.57%), secondary dryland forest (86.62%), mangrove (82.55%), and primary dryland forest/natural forest (81.97%). In the natural forest and mangrove cases, 44.61% and 38.67% of the errors, respectively, are due to omission errors, meaning that the change area in those classes was assigned incorrectly. The result also indicates that the change in forest plantations and swamp forest could be detected more accurate relatively to other forest classes.

Fig.4. Significant change pattern per period indicate the forest cover change

Moreover, 94.02% of industrial forest plantations was assigned to a spesific changed area; even when this area is not actually changed. At the same time, change in non-vegetated land use types including built-up (settlement, mining and open-area), fishpond and water-body were not examined, because the reference data for those classes were unacceptable as change categories. Additional analysis was conducted in order to get a better understanding of the dynamics change in forestland from over long-term periods and its detection accuracies. Although the overall accuracies for long-term periods were consistently in the range of 80-90% from 2001 to 2012, the omission error of the change areas in each period revealed a difference.

Fig.5. Vegetation pattern change in Sumatra Island detected by the change of temporal pattern per period

Fig.6. Detail change pattern in Riau detected by the change of temporal pattern per period

Regarding to the results of accuracy assessment, human activities such as clear cutting or logging, where was caused an area changed smaller than the minimal limit of the detectable area by the approach, have a low probability of being detected. Our analysis indicates that the change in MODIS pixels could be detected by the approach if approximately more than 40% of the sub-pixel had changed (2.5 ha) [10]. This result indicates that there is a disparity between the image pixel size and the average patch size of the change area.

Furthermore, this study helps to understand the terrestrial environmental changes such as carbon storage change and sequestration by terrestrial plants. However, this research work is carried out under the framework of regional analysis, so that better understanding of such large are as Sumatra can be achieved.

4. Conclusion

This study shows some change trajectories in Sumatra's forestlands. The main types of change are recognized, as follows: (1) agricultural development, including some trajectories such as non-agricultural lands converted into intensive agricultural lands, (2) change in forest lands, either by forest fire or human factors which is converted into oil palm plantation as well as an open area, and (3) temporary change in plantation area, either oil palm plantation or industrial forest plantation. Each change category has specific change mechanisms or processes, which then used as new category of spatial model for each land's type.

Based on the validation results, the mixed pixel issue is quite problematic, as a means to identify all of change events in the area. The abrupt changes caused by the human activities, such as logging or clear cutting, have a low probability of being detected, since the change area typically occurs in small areas or patches.

Regional shifts in temporal vegetation dynamics, including the actual changes of land use and temporary changes of land cover, have numerous consequences relevant to the environment as well as changes in carbon and nitrogen storage, land degradation and loss of biodiversity. Determining the change of temporal vegetation dynamics is the first step in understanding their implications, for example, long-term crop production, and environmental, agricultural and economic sustainability. An understanding of temporal vegetation dynamics to explain the mechanisms and pathways of land use change is important because of its relationship to ecosystem characteristics and socio-economic attributes of the land and it will be discussed separately in further research works.

Acknowledgements

The authors would like to thank the Center for Environmental Research, Bogor Agricultural University (PPLH-IPB) for giving us opportunity to get the research funding. This research was partly funded by Osaka Gas Foundation of International Cultural Exchange (OGFICE) Japan for fiscal year 2013-2014.

References

1. Lambin EF., Modeling and monitoring land-cover change processes in tropical regions. Progress in Physical Geography 1997; 21: 375-393.

2. Verburg PH, Veldcamp A, Willemen L, Overmars KO, Castella JC. Landscape level analysis of the spatial and temporal complexity of land-use change. Geophysical Monograph-American Geophysical Union 2004; 153: 217-230.

3. Geist HJ, Lambin EF. Proximate causes and underlying driving forces of tropical deforestation. BioScience 2002; 52: 143-150.

4. Turner II BL, Lambin EF, Reenberg A. The emergence of land change science for global environmental change and sustainability. Proceedings of the National Academy of Sciences 2007; 104: 20666-20671.

5. Skole DL and Tucker CJ. Tropical deforestation and habitat fragmentation in the Amazon: Satellite data from 1978 to 1988. Science 1993; 260: 1905—1910.

6. Turner II BL, Meyer WB, Skole DL. Global land-use/land-cover change: Towards an integrated program of study. Ambio1994; 23: 91-95.

7. Verburg PH, Veldkamp TA, Bouma J. Land Use Change under Conditions of High Population Pressure: The Case of Java. Global Environmental Change 1999; 9: 303-312.

8. Evans TP, Moran EF. Spatial integration of social and biophysical factors related to landcover change. Reprinted from: Population and Environment: Methods of Analysis. A Supplement to Vol. 28. Anthropological Center for Training and Research on Global Environmental Change, Indiana University; 2002.

9. Prasetyo LB, Kartodihardjo H, Adiwibowo S, Okarda B, Setiawan Y. Spatial model approach on deforestation of Java Island, Indonesia. Journal of Integrated Field Science 2009; 6: 37-44.

10. Setiawan Y, Yoshino K. Detecting land use change from seasonal vegetation dynamics in regional scale with MODIS EVI 250 m time-series imagery. Journal of Land Use Science 2014; 9: 304-330.

11. Land Resources Department/Bina Program. The Land Resources Of Indonesia: A National Overview From Regional Physical Planning Programme for Transmigration (Reppprot). Land Resources Department, Natural Resources Institute, Overseas Development Administration, London, United Kingdom; and Direktorat Bina Program, Direktorat Jenderal Penyiapan Pemukiman, Departemen Transmigrasi, Jakarta, Indonesia; 1990.

12. USGS LP DAAC. MOD13Q1. USGS/Earth Resources Observation and Science (EROS) Center, Sioux Falls, South Dakota; 2009.

13. Huete A., Didan K., Miura T, Rodriguez EP, Gao X., Ferreira LG. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 2002; 83: 195-213.

14. Gao X., Huete AR, Ni W, Miura T. Optical-biophysical relationships of vegetation spectra without background contamination, Remote Sensing of Environment 2000; 74: 609- 620.

15. MODLAND. Vegetation Indices 16-Day L3 Global 250m. Accessed on 29 May, 2014 at https://lpdaac.usgs.gov/lpdaac/products/ vegetation_indices; 2002

16. Ministry of Forestry. Recalculation of Land Cover in Indonesia 2008. Directorate General of Forestry Planning, Ministry of Forestry, Republic of Indonesia; 2008.

17. Townshend JRG, Justice CO, Kalb V. Characterization and classification of South American land cover types using satellite data. International Journal of Remote Sensing 1987; 8: 1189-1207.

18. Knight JF, Lunetta RS, Ediriwickrema J, Khorram S. Regional scale land cover characterization using MODIS-NDVI 250 m multi-temporal imagery: A phenology- based approach. GIScience & Remote Sensing 2006; 43: 1-23.

19. Setiawan Y, Yoshino K., Philpot WD. Characterizing temporal vegetation dynamics of land use in regional scale of Java Island, Indonesia. Journal of Land Use Science 2013; 8: 1-30.

20. Bouman, CA. Digital Image Processing. Lecture Material. School of Electrical and Computer Engineering, (Purdue University); 2009.