APPLICATION OF MODIS DATA TO ASSESS THE LATEST FOREST COVER CHANGES OF SRI LANKA

Assessing forest cover of Sri Lanka is becoming important to lower the pressure on forest lands as well as man-elephant conflicts. Furthermore, the land access to north-east Sri Lanka after the end of 30 years long civil war has increased the need of regularly updated land cover information for proper planning. This study produced an assessment of the forest cover of Sri Lanka using two satellite data based maps within 23 years of time span. For the old forest cover map, the study used one of the first island-wide digital land cover classification produced by the main author in 1988. The old land cover classification was produced at 80m spatial resolution, using Landsat MSS data. A previously published another study by the author has investigated the application feasibility of MODIS and Landsat MSS imagery for a selected sub-section of Sri Lanka to identify the forest cover changes. Through the light of these two studies, the assessment was conducted to investigate the application possibility of MODIS 250m over a small island like Sri Lanka. The relation between the definition of forest in the study and spatial resolution of the used satellite data sets were considered since the 2012 map was based on MODIS data. The forest cover map of 1988 was interpolated into 250m spatial resolution to integrate with the GIS data base. The results demonstrated the advantages as well as disadvantages of MODIS data in a study at this scale. The successful monitoring of forest is largely depending on the possibility to update the field conditions at regular basis. Freely available MODIS data provides a very valuable set of information of relatively large green patches on the ground at relatively real-time basis. Based on the changes of forest cover from 1988 to 2012, the study recommends the use of MODIS data as a resalable method to forest assessment and to identify hotspots to be re-investigated. It’s noteworthy to mention the possibility of uncounted small isolated pockets of forest, or sub-pixel size forest patches when MODIS 250mx250m data used in small regions. * Kithsiri Perera. perera@usq.edu.au


Introduction
Sri Lanka is an island with 65,610 sq km land area and blessed with a rich biodiversity (27% of Sri Lanka's plants are endemic) due to its tropical climate (Mongabay.com, 2006).The central hill region of the island is the source of over 103 major rivers flowing to all directions in radial shape.Sri Lanka is the house for over 20 million people and a rich array of flora and fauna.According to historical estimations, when the British Empire took the control in 1843, about 90% of the Island was covered by forest (Alagan, 2009).This lush forest cover of Sri Lanka has also experienced a rapid decrease in last 100 years parallel to most of other regions of the world.At the early stage of the decrease, forest has been lost due to the spread of plantation agriculture introduced by the British administration.In the twentieth century deforestation was caused by expansion of informal settlements due to population increase, national development projects and planned settlement programs of the government and land encroachments (Rathnayake et. al, 2002;Alagan, 2009;FRA, 2001).Figure 1 presents few typical causes for deforestation, monitored by very high resolution satellite data available through Google Earth.
These various historical and concurrent reasons are causing a significant pressure on the limited forest cover of Sri Lanka.
Assessing this limited forest cover is critically important to arrest the pressure on forest lands as well as to lower manelephant conflicts.The new land access to north-east Sri Lanka after the end of the 30 years long civil war will further increase the need of regularly updated land cover information for proper planning.Other decisive reasons for present study were the availability of previously made land cover map from Landsat MSS data with extensive field data (Perera andTateishi, 1991, Perera et al., 1992) and local research contacts.A consideration has given to resent fluctuations in annual rainfall pattern, when analysing the assessment.

Sri Lanka and Its Forest Cover
When conducting an assessment of forest cover, it's vital to understand the geography, climate, history and socioeconomic background of the considered land.Sri Lanka is characterized with a considerable topographic diversity from high mountain peaks to flat low lands compared to its size (FRA, 2001).The central region of Sri Lanka rises over 2,500m (Mt Pidurutalagala 2,524 m) and a sizable area in the country exceeds 1500 m above sea level, where a thick forest cover blanketed before large scale plantations begun in 18 th century.
The first significant impact on the forest cover of the island has caused by the large scale plantations (coffee then tea, rubber, coconut, and cinnamon) started by British colonial rulers since mid 1800s.A vast flatland along the coastal belt Climatology of Sri Lanka works favourably to have a forest cover at any given location of the country except Jaffna peninsula where soil is unfavourable and few coastal dry lands.The island gets rainfall from two major monsoons (south-west and north-east) apart from inter-monsoon rains.
The south-west monsoon (May to September) poured over 55% of the annual rainfall (GFDRR, 2011).Figure 3 shows the average annual rainfall distribution of the island where total rainfall often exceeds 3000mm.
The pattern of river network, mean annual rainfall, elevation, and the remaining forest in relatively dry regions of the country indicates clear evidences for the historical flush greenery in all regions.However, due to the population increase and economic activities, forest cover has restricted to three key regions of the country, i.e., Mahaweli River basin and northern plains, Central Hills, and Yala sanctuary in southeast.When mean annual rainfall data layer overplayed with rivers and elevations, the geo-climatic background of these remaining forest areas becomes clearer, as presented in Figure 4.The most of the forest cover is located in relatively dry and flat terrain where less human activities are located.
The forest cover of Sri Lanka faced a drastic reduction in last 100 years.According to historical reports 70% of the land under forest in 1900.The area under forest cover in recent years has given with different percentage values in different studies depending on the data source of the map, definition of the forest, and mapping methodology (Alagan, 2009;Perera & Tateishi, 1996;Rathnayake et al, 2002;Suzuki, 2007).Nearly all estimations have placed the percentage of forest cover of Sri Lanka as 25% -30% of the total land area.A descriptive forest classification presented by the forest conservation department, Sri Lanka using government data of 1992 has given seven categories based on climate and elevation (WWCT, 2011).However, forest cover monitoring with satellite images are largely based on spectral information recorded by the satellite sensor and can be used to monitor vegetation cover regardless to elevation and climate conditions.This study investigated the applicability of MODIS 250m satellite images to classify total forest area and compare with a finer spatial resolution based old forest cover data set produced in 1988.

Applicability of Satellite Imagery
The relationship between climate and elevation of Sri Lanka was discussed in this report in order to provide  (Perera et al., 1992;Eurisy report, 2011;Lehmann et al. 2012).Through the advancements of the technology, high resolution imaging sensors have been launched and operated and the data acquired by these satellites are useful for producing detailed land cover maps.
Though these advantages are promising, it is extremely difficult to get cloud-free imageries acquired by these satellites for Sri Lanka due to tropical cloud coverage over the mountains and long recurrent periods of the satellites.
The high cost of finer resolution satellite images is another very significant hurdle faces by developing nations in application studies.Apart from the economic hardships, Sri Lanka suffered 30 years long civil war which just ended in 2009.Spectral data acquired by MODIS system can be utilized as a reasonable solution under these circumstances.
Table 1 presents some of the technical information of four major earth observation systems.
After MODIS (MODerate-resolution Imaging Spectrometer) ( MODIS, About MODIS, 2012) images became available many scientists have successfully produced land cover maps or combined with other GIS (Geographic Information Sysyems) data to make use the advantages of MODIS system (Friedl et al., 2002;Hall et al., 2002: Price, 2003;Zhan et al., 2002).Geometrically corrected MODIS data products such as MODIS NDVI (MODIS WEB, 2006) and true colour image data (NASA, MODIS Rapid Response Systems, 2011; Gumley et al., 2003) are available through NASA for the global community at no cost.Taking advantage of these pre-processed NASA's MODIS products and its inherited technical advantages (recurrent and swath), this study used a near-cloud free true colour MODIS mosaic of Sri Lanka to conduct the forest cover assessment.The ground for present study was supported by a recently published paper by the author on experimenting on application of MODIS and Landsat MSS data to observe land cover changes in a selected sub-region of Sri Lanka (Perera & Tsuchiya, 2009).

APPLICABILITY OF SATELLITE IMAGERY
The relationship between climate and elevation of Sri Lanka was discussed in this report in order to provide information on non-existing positive relation of present forest cover with high rainfall and elevation.The assessment discussed in this study is focused on quantitative change of the forest cover which is one of the most important factors to be considered in any forest conservation plan.However, field survey based forest investigations are costly for mapping and regular updating.
Here, application of satellite images for forest cover investigations provides an ideal cost effective alternation.Though these advantages are promising, it is extremely difficult to get cloud-free imageries acquired by these satellites for Sri Lanka due to tropical cloud coverage over the mountains and long recurrent periods of the satellites.The high cost of finer Table 1.A comparison of basic components of four prominent earth observation satellites.
After MODIS (MODerate-resolution Imaging Spectrometer)( MODIS, About MODIS, 2012) images became available many scientists have successfully produced land cover maps or combined with other GIS (Geographic Information Sysyems) data to make use the advantages of MODIS system (Friedl et al., 2002;Hall et al., 2002: Price, 2003;Zhan et al., 2002)

Forest cover of 1988
The land cover map of 1988 was produced as one of first digital spatial data base produced for entire Sri Lanka in 1980s, using Landsat MSS data (Perera et al., 1991(Perera et al., , 1992)).The land cover map was extensively supported by ground truth surveys.Only a brief explanation is included in this study on the data processing methodology of 1988 map.
After correction geometric distortions of all MSS images using ground control points, the land cover classification of each image scene was performed separately applying the maximum likelihood classifier.About 50% of Sri Lanka was covered by Landsat MSS images of 2 nd August, 1988 (scene number 140-55, 140-56, and 141-54)   Based on the research finding of Liang et al.(2002), this study assumed characteristic of Landsat ETM+ is applicable to Landsat MSS data, and the pixel size adjustment of MSS image data were performed through aggregation of 3×3 pixels converting into 240 m resolution image.The resulted forest map was registered on 2012 map using map to map geometric transfer and presents in Figure 5.

Forest cover of 2012
The near perennial cloud cover in tropical sky is one of the controlling factors for studies based on active remote sensing data in tropical regions.This obstacle was discussed in detailed in a study that used MODIS true colour images to classify land cover of Mekong River basin (Perera et al., 2010).However, unlike the complicated data processing for producing number of land cover categories, isolating forest only cover is an achievable task with fewer difficulties.To meet the ISPRS standard length requirements of the report, some of the preliminary stage data mining and processing steps were not included in this report.After a careful observation of MODIS images from Jan 2011 to Jan 2012, three images (2011067,2012001,2012003) were selected to produce the true colour mosaic to extract of forest cover.The image dates were selected from same rainfall season to avoid any fluctuations in greenery.Image processing was conducted using ENVI 4.6 package.To delineate the clear coastal boundary and isolate inland water features (water mask), band 7(R), 2(G), 1(B), mosaic also produced using MODIS images of same dates.In band 7-2-1 combination, water is visible in dark blue and no sub colours are prominently appearing within water.
Figure 6 presents true colour mosaic superimposed with the water mask produced using band 7, 2, and 1 combination.
After interpolating and registering 1988 image with 2012 data set, same water mask was superimposed on 1988 data set too.As MODIS documentation explains, its true colour band combination is strongly representing vegetation information on the ground while keeping the soil and vegetation differences intact.Table 3 shows key uses of each band in true colour combination of MODIS.To integrate with Google Earth, MODIS KML files of same image dates were downloaded.The set of training sites was selected using the true colour mosaic and integrated with MODIS KML files to locate exact location in high resolution Google Earth images.
When selecting training samples, it was assumed that the forest cover is having fewer fluctuations in green biomass if the images collected in same rainy season.Pixel values within each training site were used as the initial criteria to discriminate pre identified forest cover, with the support of knowledge about the region and existing map information, apart from the very high resolution images in Google Earth.
Each training sample contained well over 1000 pixels to give enough representation for each pre-identified forest and non forest areas.Image classification was conducted by the supervised maximum likelihood classifier and final forest map of 2012 was produced through combining initial forest classes (Figure 7).The same process used for training sample generation was administrated for accuracy assessment of 2012 forest cover map, which gave an over 85% mapping accuracy (based on randomly selected 50 sample points within classified forest polygons).

Analysis of Results and Conclusions
The assessment of forest cover of Sri Lanka was achieved using 1988 and 2012 forest cover maps produced using Landsat MSS and MODIS images respectively.Area under forest has changed significantly within last 23 years, by likelihood classifier and final forest map of 2012 was produced through combining initial forest classes (figure 7).The same process used for training sample generation was administrated for accuracy assessment of 2012 forest cover map, which gave an over 85% mapping accuracy (based on randomly selected 50 sample points within classified forest polygons).The spatial pattern of forest cover changes between 1988 and 2012 shows most of the major forest regions mentioned in Figure 4 are intact apart from some large scale negative changes visible in southeast region and small scale positive Figure 8 shows forest cover change detection image of southeast Sri Lanka, where some major changes have highlighted.As possible reasons for the negative changes, recent large scale agriculture settlements can be pointed out.
Village expansion and large scale settlements have consumed 51,000 hectares of new lands between 1983 and 1992 throughout Sri Lanka (Ratnayake et al, 2002).Various positive changes in forest cover may have caused by favourable weather conditions that created higher spectral values in 2012 MODIS images.The study showed the applicability of MODIS images to ascertain single land cover type (forest) changes with an acceptable level of information.Application of MODIS to update remote sensing data based old land cover products has a cost effective benefit, especially for developing nations, where acquiring high resolution satellite data is costly.Furthermore, MODIS based change detection and assessment is strong enough to identify hotspots to be re-investigated.It's worth to mention the possible unaccountability of isolated forest pockets smaller than MODIS 250mx250m spatial resolution.

Figure 1 .
Figure 1.Major causes of deforestation in Sri Lanka; Atraditional cut and burn farms, B -large scale plantations, C -new agriculture settlements, D -filling wet lands for housing.See approximate image locations in Figure 2.

Figure 2 .
Figure 2. The central highlands and river network of the island (A, B, C & D present approximate locations of Figure 1 images).

Figure 3 .
Figure 3. Mean Annual rainfall of Sri Lanka.

Figure 6 .
Figure 6.Cloud free MODIS image mosaic of Sri Lanka.Figure7.Forest cover of 1988.

Figure 8 .
Figure 8. Forest cover changes detected in southeast Sri Lanka.

Table 2
presents spectral bandwidth information of two satellite systems used in the study.The old forest cover map was extracted from 1988 land cover map produced by the author using Landsat MSS images.Though spectral Figure 4. Regions where major forest regions locate in Sri Lanka, overlays with of mean annual rainfall, central hills, and river network.

Table 2 .
-free imageries acquired by these satellites for Sri Lanka due to tropical cloud coverage over the mountains and long recurrent periods of the satellites.The high cost of finer resolution satellite images is another very significant hurdle faces by developing nations in application studies.Apart from the economic hardships, Sri Lanka suffered 30 years long civil war which just ended in 2009.Spectral data acquired by MODIS system can be utilized as a reasonable solution under these circumstances.Table1presents some of the technical information of four major earth observation systems.Similarities in spectral bandwidth between Landsat MSS and MODIS systems.Table2presents spectral bandwidth information of two satellite systems used in the study.The old forest cover map was extracted from 1988 land cover map produced by the author using Landsat MSS images.Though spectral characters are not perfectly matching between MSS and MODIS, table 2 shows some significant similarities in bands used in old and new map.In this study, application of two different satellite data sets to assess single spatial component bears two important features.1.Assessing forest cover of entire Sri Lanka at 23 years of time span using satellite image products.2.Testing the applicability of two different satellite data sources at entirely different spatial resolutions for mapping task.The relationship between climate and elevation of Sri Lanka was discussed in this report in order to provide information on non-existing positive relation of present forest cover with high rainfall and elevation.The assessment discussed in this study is focused on quantitative change of the forest cover which is one of the most important factors to be considered in any forest conservation plan.However, field survey based forest investigations are costly for mapping and regular updating.Here, application of satellite images for forest cover investigations provides an ideal cost effective alternation.Since the launch of Landsat MSS(Multi-Spectral Scanner System) (Perera et al., 1992;009) et al., 2002: PrGLehmann et al. 2012), 2002)ll et al., 2002: Price, 2003;Zhan et al., 2002).Geometrically corrected MODIS data products such as MODIS NDVI(MODIS WEB, 2006)and true colour image data(NASA, MODIS Rapid Response Systems, 2011;Gumley et al., 2003)are available through NASA for the global community at no cost.Taking advantage of these pre-processed NASA's MODIS products and its inherited technical advantages (recurrent and swath), this study used a near-cloud free true colour MODIS mosaic of Sri Lanka to conduct the forest cover assessment.The ground for present study was supported by a recently published paper by the author on experimenting on application of MODIS and Landsat MSS data to observe land cover changes in a selected sub-region of Sri Lanka(Perera & Tsuchiya, 2009).earthobservationsatellite in July 1972 (NASA, 2012), earth observation including forest cover monitoring from space ran into a technological revolution.In recent years there are many well developed operational satellite services to support decision-making and sustainable forest management, on a global, national and sub-national scale(Perera et al., 1992;  Eurisy report, 2011;Lehmann et al. 2012).Through the advancements of the technology, high resolution imaging sensors have been launched and operated and the data acquired by these satellites are useful for producing detailed land cover maps.

Table 3 .
The usages of MODIS true colour bands 2001), though the area is higher than according to a study which calculated the closed canopy forest area as 25.7%(Ratnayake  et al, 2011).Total forest area (19,330 sq km) published by mongaday.com(2006) is much closer to the 2012 map figure.highlighted.As possible reasons for the negative changes, recent large scale agriculture settlements can be pointed out.Village expansion and large scale settlements have consumed 51,000 hectares of new lands between 1983 and 1992 throughout Sri Lanka (Ratnayake et al, 2002).Various positive changes in forest cover may have caused by favourable weather conditions that created higher spectral values in 2012 MODIS images.The study showed the applicability of MODIS images to ascertain single land cover type (forest) changes with an acceptable level of information.Application of MODIS to update remote sensing data based old land cover products has a cost effective benefit, especially for developing nations, where

Table 3 .
The usages of MODIS true colour bands.Figure 6.Cloud free MODIS image mosaic of Sri Lanka.

Table 3
. The usages of MODIS true colour bands5.ANALYSIS OF RESULTS AND CONCLUSIONSThe assessment of forest cover of Sri Lanka was achieved using 1988 and 2012 forest cover maps produced using Landsat MSS and MODIS images respectively.Area under forest has changed significantly within last 23 years, by 137,700 hectares (table 4).The total area under forest in 2012 map is lower than the calculated forest area figure (30.9%) published in 2000(FRA,  2001), though the area is higher than according to a study which calculated the closed canopy forest area as 25.7%(Ratnayake  et al, 2011).Total forest area (19,330 sq km) published by mongaday.com(2006) is much closer to the 2012 map figure.

Table 4 .
Assessment of forest area changes.

Table 4 .
Assessment of forest area changes.137,700hectares(Table4).The total area under forest in 2012 map is lower than the calculated forest area figure (30.9%) published in 2000(FRA, 2001), though the area is higher than according to a study which calculated the closed canopy forest area as 25.7%(Ratnayake et al, 2011).Total forest area (19,330 sq km) published by mongaday.com(2006) is much closer to the 2012 map figure.