XIV WORLD FORESTRY CONGRESS, Durban, South Africa, 7-11 September 2015

Potential use of synthetic aperture radar in detecting forest degradation in the protected areas of the : a case study of Roven Tumaneng1, Angelica Kristina Monzon1, Joanne Rae Pales1, and Jose Don de Alban1

1Fauna & Flora International - Philippines, [email protected], Tagaytay City, Cavite, Philippines

Abstract

With the forest of the Philippines under intense pressure, one of the innovative ways of detecting and monitoring deforestation and forest degradation aside from using optical imagery is using radar images such as Synthetic Aperture Radar (SAR) data. This study aims to recognise the potential use of SAR technology in the Philippines for determining the extent of deforestation and forest degradation particularly in protected areas. One of the identified protected areas that are being degraded at an alarming rate is Mt. Guiting-guiting Natural Park (MGGNP) in Sibuyan Island, . Dual-polarised L-band ALOS/PALSAR mosaic data at 25-meter spatial resolution, taken at two acquisition years (2007 and 2009), were utilised in this study to provide information on forest cover and habitat change in Sibuyan Island between year 2007 and 2009. In-situ data was used for forest/non-forest classification and 15 habitat transects were surveyed to identify the different habitat types using Mallari et al.’s (2011) broad habitat types (4 classes) and used to train and validate it. Results showed that there is moderate to severe degradation of old growth forests (at the rate 24% between 2007-2009) as seen in land cover change detection models while the annual deforestation rate in the island was computed at 0.36%. It indicates that the forest cover almost remained unchanged from that span of 2 years but the rate of degradation is significantly increasing evinced by the decrease in extent of old growth forest whilst gaining early secondary growth forest. Some of the major identified causes of forest degradation in the island is due to unregulated harvesting of forest resources such as fuel wood, charcoal and poles/post, unregulated opportunistic logging and unsustainable harvesting of non-timber forest products. Overall classification accuracies achieved using Mallari et al.’s (2011) broad habitat types at 70.29% and 69.70% for 2007 and 2009 PALSAR data, respectively.

Keywords: ALOS/PALSAR, forest degradation, habitat change, protected area management, change detection, forest monitoring

Introduction, scope and main objectives

Philippines have a numerous protected areas, estimated to cover about 25, 995 km2, or 87% of the country’s area (Ong et al., 2002). Although each protected area in existence is very important, much more needs to be done particularly on monitoring its forest area. On the other hand, information on forest classification, extension, degradation, biomass, and temporal evolution of forest areas is therefore tremendously requested, especially in regions with scarce accessibility (Petinatto et al. 2014).

Remote sensing has become a useful and important tool for evaluating the implementation of environmental policies and decision-making (Mayer and Lopez, 2011). Together with standardized ground plots and regular in situ measurements, remote sensing is a powerful monitoring device as well especially on land cover/forest habitat change detection and identifying change hotspots. Earth observation data have been used in the past (c. since 1970s) in the Philippines for national and sub-national forest cover mapping, mainly by applying visual interpretation techniques on optical (multispectral and hyperspectral) satellite data (e.g., Landsat, SPOT, etc.) (Swedish Space Corporation, 1988; Kummer, 1992). However, persistent cloud cover and inclement weather conditions in the Philippines limits the utility of optical/multispectral data that can be used for either wall-to-wall or even site- level land cover mapping and consistent forest change monitoring.

The Japan Aerospace Exploration Agency’s (JAXA) Advance Land Observing Satellite (ALOS) Phased Array L band Synthetic Aperture Radar (PALSAR) provided an opportunity to utilise synthetic aperture radar (SAR) as a suitable alternative data source to overcome these limitations of optical/multispectral data. SAR is a side-looking active remote sensing system, which transmits microwaves and receives measurements from the signal backscattered from a surface, to produce an image after appropriate processing (Wheeler et al. 2014). The main benefit of SAR its ability of its sensor, day- and night-time imaging capability and independence from weather conditions and solar illumination. SAR imaging has been available in remote sensing since 1951 (Almeido-Filho et al. 2009), and its applications cover a wide range in mapping and monitoring of natural resources. Moreover, its capability has been effective for periodic forest monitoring and change detection (Rosenqvist, et al. 2010; Rahman and Sumantyo, 2010). SAR data are available internationally from a large number of satellites with different frequency bands, polarizations, and variable imaging geometries (Walker et al. 2010). Longer wavelength L-band SAR, such as ALOS/PALSAR, have been demonstrated to be suitable for monitoring forest and wetland areas largely because of its sensitivity to vegetation (Almeido-Filho et al. 2009; Walker et al. 2010; Rosenqvist et al. 2010)

This study focuses on the potential of using SAR to map and delineate forest cover in Sibuyan Island. This paper also provides the ability of SAR to distinguish vegetation-based habitat categories and the use of remote sensing for assessing the conservation status of habitat types.

Study Area and Data

A. Study Area

Sibuyan Island is the second largest island of Romblon Province with an approximate land area of 45,600 has. The Islands of , , Masbate and Tablas surround the island. Sloping lands dominate the topography of the area. Approximately, 77% of the land is considered to have a slope of 18% and above. The highest peak in the island is located at Mt. Guiting-guiting with an elevation of 2,058 meters above sea level. Sibuyan Island is under Type III climate based on Corona’s classification, which indicates that there is no very pronounced maximum rain period in the area, and has a short dry season usually lasting from 1-3 months (March-May or December-February) (DENR- , 2014). There are 3 municipalities and 36 barangays with an annual growth rate of 0.96%. Currently, the population of Sibuyan is 57,248 and a population density of 118.2/km2 (NSO, 2010). The indigenous dwellers in Sibuyan Island are called Sibuyan Mangyan Tagabukid (SMT). Two Certificate Ancestral Domain Titles (CADTs) were issued to SMT covering a total of 8,408.84 has located in two non-contiguous areas within and San Fernando (Tongson & McShane, 2006). The island is recognized as one of the key biodiversity area. It was believed that many groups of plants and animals demonstrate medium to high levels of endemism because of diversification brought about by its complex geological history and long periods of isolation from other islands (Brown and Diesmos, 2009). Aside from that, this island still has high remaining forest cover. Due to its ecological importance, Sibuyan Island’s most prominent peak, Mt. Guiting-guiting has been declared Natural Park by Pres. Fidel V. Ramos under Proclamation No. 746 on February 20, 1996. The terrestrial reserve with an area of 15,260.48 hectares was mandated to be protected and conserved in a sustainable basis (DENR- MIMAROPA, 2014). In spite of its protected status, Mt. Guiting-guiting Natural Park (MGGNP) and many other natural areas on the island remain under serious threats.

B. ALOS/PALSAR Data

Eight dual-polarisation (HH and HV) ALOS/PALSAR image mosaics at 25-meter spatial resolution taken at two image acquisition years (2007 and 2009) were used for the forest cover and habitat assessment in Sibuyan Island, Romblon. Each tile was identified with the reference address of latitude-longitude of N13 E122 in each of the two acquisition years. Each 1x1-degree image mosaic tile (4500 x 4500 pixels) was pre-processed using Japan Aerospace Exploration Agency’s (JAXA) mosaicking algorithm, which includes calibration, orthorectification, slope correction, and intensity normalisation with neighbouring data strips (Rahman and Sumantyo, 2010).

C. Field Data

Figure 1: Collected ground truth points over the 2010 land cover map of Sibuyan Island, Romblon. (Data source: National Mapping and Resource Information Authority, 2010) Field data were collected during the habitat and biodiversity surveys, land cover assement and socio-economic survey conducted in three municipalities of Sibuyan Island namely, Cajidiocan, San Fernando and Magdiwang in May to June 2014. Habitat surveys were conducted using transect route. Transects comprised of a 2-kilometer line with transect points marked at every 25 meters. A total of 15 transects were surveyed covering a full range of habitat and disturbance gradients across all three sites of Sibuyan Island. Habitat types was identified for every transect points using the broad habitat classification of Mallari et al. (2011) which consists of four habitat categories: old growth forest, advanced second growth forest, early second growth forest, and cultivated areas (Table 1). These habitat categories will be used in generating supervised classification of the ALOS/PALSAR mosaic images of 2007 and 2009. GPS coordinates for each point were also recorded in field datasheets. A total of 1,712 ground truth points were gathered, comprised of 770 ASG samples, 62 OG samples, 336 ESG samples and 544 CVT/other-areas samples were used in image classification process (Figure 1).

Table 1: Description of the broad habitat classification adapted from Mallari et al. (2011).

Habitat types Description Areas with active or recently abandoned farmland; included grasslands, Cultivated areas (CVT) brushlands, agricultural plots, and small orchards with fruit trees ≤ 4 meters tall. Early Second Growth Areas of newly regenerating forest (< 20 years old) dominated by saplings forests (ESG) and other small- to medium-sized trees Advanced Secondary Forests that are c. 20–40 years old, which have a less dense understorey and Growth forests (ASG) are dominated by medium to large trees (30-50 cm) Primary forest or forests >40 years old which are dominated by large to very large trees and have a less complex understorey compared to ESG and ASG. Old Growth forests (OG) Extent of the area is not limited to the lower and upper altitudinal limit of the mixed-dipterocarp and/or mossy forest.

Methods

ALOS/PALSAR Data Processing

(1) Forest and Habitat cover mapping. Prior to classification and interpretation, a series of pre-processing steps involving geometric and radiometric corrections developed by Shimada et al. (2010) were applied to the pre-processed mosaic tiles of 2007 and 2009 images. For geometric pre-processing, the mosaic tile was geocoded, and re-projected to Universal Transverse Mercator Zone 51 North coordinate system World Geodetic System 1984 datum (UTM WGS84) coordinate reference system with less than 0.5 total root mean square error (RMSE) to ensure the co-registration across other tiles and other spatial data. Radiometric pre-processing includes speckle filtering in the reprojected images to compensate the signal noise observed. It was implemented using Lee speckle filter with a medium-sized kernel (5 x 5) to reduce signal noise from the radar data and used to preserve edge information between transition areas of different habitat types. Once the speckles were reduced, the digital number (DN) were converted from amplitude format to normalized radar cross-section, or sigma- naught (σ0; units in decibels or dB) using the equation:

σ0 = 10 log10 [DN2] + CF; (Eq.1)

Where CF is the calibration factor with a value of -83 dB [31]. Ratio images were computed (HH/HV) and stacked together with the HH and HV image bands. Image masks were used to exclude water bodies, layover, and shadow regions where radar data may be erroneous due to the terrain and sensor limitations caused by acquisition geometry, that is specific for each acquisition time & season. The reprojected image masks for each year were merged together, for consistent land area computation between the two image dates of the study area. All point coordinate data collected from field observations was consolidated and organised according to the habitat classification schemes adopted in Sibuyan Islands. The support vector machine (SVM) classifier was used to generate the supervised habitat types classification for the two image dates, since the capability of its algorithm can generalise well, even with very limited training samples, and often produce higher classification accuracy than traditional methods (Rosenqvist, et al. 2007). The 1, 712 collected ground truth data were used in classifying the images into the different habitat types. The consolidated field database was divided into two sets as training and validation data for image classification: 80% for training data and 20% for validation data for each classification type. Classification accuracy was assessed using an error matrix, which provided the basis on which to both describe classification accuracy and characterize errors of each category present in the classification (Stehman, 1997; Elith et al. 2006). The broad habitat classification was further re-clustered to produce the forest cover extent (combining old growth forest, and advanced and early secondary growth forest); and non-forest cover extent (including cultivated and other non-forest areas).

(2) Forest and Habitat change detection methods/analysis. Change detection was conducted between the two image dates by producing a change matrix to show the changes between forest cover/habitat types at an initial state (2007) and a final state (2009) of the study area. Forest/non-forest change detection will locate the gain/loss of forest cover in the island while habitat change detection will identify changes in the quality of forest habitat. In determining the habitat change, the broad habitat classification was congregate into the following: good forest (combining old and advanced secondary growth forest); degraded forest (early secondary growth forest); and non-forest (including cultivated and other non-forest areas). Areas that demonstrated high disturbance levels (conversion from forest to non-forest or good forest to degraded forest) were identified as hotspots.

Results

A. Forest and Habitat Cover Mapping

(1) Forest cover. ALOS/PALSAR images of 2007 and 2009 were processed for forest/non- forest cover mapping and subsequently analysed for forest change detection. The results in Table 2 show that 80.55% and 81.13% of the total land area of the island is still covered by forest in 2009 and 2007, respectively. However, bulk of all these forests are found in uplands or in high elevation areas. Around 80% of the total land area of the municipalities is still forested, with the largest forest block found in the municipality of San Fernando with total land area of 15,379.83 hectares.

Table 2: Area summary of the forest and non-forest cover classess of each municipality in Sibuya Island from 2009 and 2007 ALOS/PALSAR mosaic data. All percentages presented were computed with respect to the total area of the island.

2009 Forest/Non-forest 2007 Forest/Non-forest Land Area Net Change Municipalit cover cover (Area in in Forest y Forest Non-forest Forest Non-forest ha) (Area in ha) (Area in ha) (Area in ha) (Area in ha) (Area in ha) 12,182.32 2,978.62 12,226.06 2,934.87 (43.74) Cajidiocan 15,160.94 (28.44%) (6.95%) (28.54%) (6.85%) 0.10% 6,945.47 2,119.47 7,162.69 1,902.26 (217.22) Magdiwang 9,064.93 (16.21%) (4.95%) (16.72%) (4.44%) 0.51% San 15,379.83 3,236.43 15,370.24 3,246.02 9.59 18,616.26 Fernando (35.9%) (7.55%) (35.88%) (7.58%) (0.02%) Sibuyan 42,842.1 34,507.62 8,334.52 34,758.99 8,083.15 (251.37) Island 4 (80.55%) (19.45%) (81.13%) (18.87%) 0.59%

(2) Habitat cover. The results of the habitat cover mapping as illustrated in Table 3 show that less than 50% total land area of the island is covered with old growth forest in 2009. Advanced second growth forest was found to cover 7.74% of the total land area of the island in 2009, while it was 19.09% in 2007. Early second growth forest gains its cover from 13.14% in 2007 to 28.85% in 2009. Table 3: Area summary of broad habitat cover of each municipality in Sibuyan Island from 2009 and 2007 ALOS/PALSAR mosaic data. All percentages presented were computed with respect to the total area of the island

Habitat Types Land Advanced Secondary Early Secondary Cultivated areas and Old Growth Net Net Net Net Municipal Area Growth Growth others (Area in ha) Change Change Change Change ity (Area in (Area in ha) (Area in ha) (Area in ha) ha) (Area in (Area in (Area in (Area in 2009 2007 ha) 2009 2007 ha) 2009 2007 ha) 2009 2007 ha) 6,676.25 7,480.17 1,160.84 2,870.34 4,345.23 1,875.55  2,469 2,978.62 2,934.87 Cajidiocan 15,160.94  803.93  1,709.49  43.75 (15.58) (17.46) (2.71) (6.70) (10.14) (4.38) .68 (6.95) (6.85) 3,566.01 4,252.92 619.38 1,705.17 2,760.07 1,204.60  1,555 2,119.47 1,902.26 Magdiwang 9,064.93  686.91  1,085.78  217.21 (8.32) (9.93) (1.45) (3.98) (6.44) (2.81) .48 (4.95) (4.44) San 8,585.63 9,220.26 1,537.57 3,602.03 5,256.63 2,547.95  2,708 3,236.43 3,246.02 18,616.26  634.63  2,064.46  9.59 Fernando (20.04) (21.52) (3.59) (8.41) (12.27) (5.95) .68 (7.55) (7.58) Sibuyan 42,842.1 18,827.89 20,953.36  2,125.4 3,317.80 8,177.54  4,859.7 12,361.93 5,628.10  6,73 8,334.52 8,083.15  251.3 Island 4 (43.95) (48.91) 6 (7.74) (19.09) 4 (28.85) (13.14) 3.83 (19.45) (18.87) 7

Figure 2: Pixel-based change detection of Sibuyan Island using 2007 & 2009 ALOS-PALSAR Mosaic Images.Change detection analysis refers to the tracking of conversion in the pixel classification generated from the processing of radar backscatter using a categorical reference derived in time 1 to a change or unchanged category in time 2. B. Forest and Habitat Cover Changes

The results of the change detection for forest/non-forest cover show that the island lost a total of 251.37 hectares of forest cover from 2007 to 2009 (Figure 3). This is equivalent to 0.59% of the island’s total land area and the highest forest cover loss is found in the municipality of Magdiwang with 217.22 hectares. From the results of the habitat cover change detection as illustrated in Table 4, Sibuyan Island lost a total of 2,125.46 hectares of old growth forest and 4,859.74 hectares of advanced second growth forest (Figure 4). In comparison, the total area of early second growth forest increased its area coverage to a total of 6,733.83 hectares and the cultivated/other lands increased by 251.37 hectares from 2007 to 2009 as well.

Figure 3: 2007-2009 Change detection of forest cover showing areas without change, forest areas that became non-forests and non-forest areas that became forests. Figure 4: 2007-2009 Change Detection of Habitat quality showing changes from OG to ASG, vegetation regrowth, and no changes in habitat quality.

C. Accuracy Assessments

Overall classification accuracies at 90.70% and 89.33% were achieved using the SVM classifier for the 2007 and 2009 PALSAR data, respectively, which satisfies the universally accepted minimum 85% overall accuracy (Foody, 2002). Error matrices show that forest classes derived from 2007 and 2009 PALSAR data are in agreement with the ground-truth validation data at 99.54% and 94.95%, respectively; while non-forest classes have lower agreement compared with ground-truth validation data at 67.07% and 74.39%, respectively (Table 4a and Table 4b).

T a b l e 4: Error matrices for: (a) 2007 forest/non-forest cover classification, and (b) 2009 forest/non-forest cover classification from the ALOS/PALSAR 25-meter mosaic data.

(a)

2007 Forest/Non-forest Ground Truth (%) cover map Class (%) Forest Non-forest Total Forest 99.54 32.93 81.40 Non-Forest 0.46 67.07 18.60 Overall Accuracy 100.00 100.00 100.00 (b)

2009 Forest/Non-forest Ground Truth (%) cover map Class (%) Forest Non-forest Total Forest 94.95 25.61 76.00 Non-Forest 5.05 74.39 24.00 Overall Accuracy 100.00 100.00 100.00

Overall classification accuracies achieved using the Mallari’s et al. (2011) broad habitat classification were 70.29% and 60.70% for 2007 and 2009 PALSAR data, respectively. Error matrices show that classification of more than two habitat cover types yield lower accuracy compared to the accuracies computed for only two classes of forest/non-forest. The highest classification accuracy for 2007 is computed from the old growth forest with 83.33% while the lowest is from advanced second growth forest with 23.33% (Table 5a). The highest classification accuracy for 2009 is computed from the advanced second growth forest with 81.36%, while the lowest is from early second growth forest with 42.86% (Table 5b).

Table 5: Error matrices for (a) 2007 broad habitat cover classification, and (b) 2009 broad habitat cover classification result from the ALOS/PALSAR 25-meter mosaic data.

(a) Error Matrix for 2007 Broad habitat types Map Ground Truth (%) Class (%) OG ASG ESG CVT Total OG 83.33 46.67 16.67 3.03 19.70 ASG 0.00 23.33 10.00 1.52 8.33 ESG 16.67 10.00 63.33 3.03 18.94 CVT 0.00 20.00 10.00 92.42 53.03 TOTAL 100.00 100.00 100.00 100.00 100.00

(b)

Error Matrix for 2009 Broad habitat types Map

Ground Truth (%) Class (%) OG ASG ESG CVT Total OG 71.43 8.47 34.29 9.46 16.57 ASG 28.57 81.36 22.86 0.00 33.14 ESG 0.00 8.47 42.86 16.22 18.29 CVT 0.00 1.69 0.00 74.32 32.00 TOTAL 100.00 100.00 100.00 100.00 100.00

Discussion

Effectiveness of ALOS/PALSAR in mapping forest/habitat cover types in Sibuyan

It was demonstrated that mapping forest and non-forest areas using the 25-meter ALOS/PALSAR mosaic could yield classification maps with satisfactory overall accuracies of greater than 85%. The results of the habitat cover mapping and change detection yield lower classification accuracies especially between the forest habitat types. One of the possible reasons that may affect this result was the assessment of the forest habitat quality, which may not only depend on the forest stand and physical structure of the assessed area, but also the biological parameters (i.e. tree/plant species) regardless of the size of the trees observed.

Potential of SAR in detecting deforestation and forest degradation

Eighty-one percent (34,508 ha) of Sibuyan Island is covered by forest while around 50% of it is old growth forest. The forest cover of the island has decreased with an annual rate of 3.45% and an annual regrowth rate of 3.08%, with respect to the 2007 forest cover. Hence, the net rate of change of the forest cover of the island was computed at 0.36% (evinced by the land cover change from forest to non-forest). It indicates that the forest cover almost remained unchanged and intact from that span of 2 years (Figure 3). While only 251.37 ha net loss of forest cover from 2007 & 2009, results from broad habitat cover classification (Table 2) indicate that the island has experienced moderate to severe degradation of old growth forests. A staggering 6,734 ha (54.5%) increase of degraded forests was only observed between 2007 and 2009 (Table 3). The results also suggest that there is 24% rate of degradation in the island between 2007 and 2009 evinced by the decrease in extent of old growth and advanced second growth while there is an observed expansion of early secondary growth forests (at 9.82% per annum). The change maps exhibited clear spatial variation of forest degradation (Figure 4). The results from the broad habitat cover classification confirm that there was minimal net change from any of the forest habitat classes to the non-forest habitat (cultivated/other land classes) over the period of the assessed images. However, results from the change detection of the forest habitat cover classes indicate that while the general classification of the island’s landscape, remain to be identified as forested, the quality of the forest relating to its habitat had experienced drastic changes between the period of the image analysis.

Causes of Forest Degradation in Sibuyan Island

The severe forest degradation in the island are due to unregulated harvesting of forest resources such as fuel wood, charcoal and poles/post, unregulated opportunistic logging and unsustainable harvesting of non-timber forest products especially gathering of wild honey. Rampant collecting of honeys inside the forest results to the establishement of many trails inside the forest. Since these honey collectors have the opportunity to go further in the forest, they also tend to collect other non-timber forest products for their subsistence needs. Although their area of mobility has been constricted due to increasing population pressure and access restriction imposed by protected area and regulated by forestry law. This has made them to go further and access other forest areas for their subsistence needs. However, subsitent forest gatherer or people that corresponds to dependency on forest resources as main livelihood usually comprised of younger generations and newly settled migrants was found to be possible main driver for forest degradation.

Conclusions

This study demonstrated the capability of dual-polarised ALOS/PALSAR data mosaic data for detecting and monitoring land and forest cover change and generating spatially explicit activity data information between different land cover types and especially forest degradation. The output also of this study has been viewed as input to the enhancement of the management plan. But also improve the governance of the remaining forest of the protected areas in the Philippines. Moreover, this information is intended to address the drivers of habitat conversion and reduce biodiversity loss. These can potentially be applied and scaled-up to provide inputs for national-level monitoring of forest cover in the Philippines.

Acknowledgements

The work on the ALOS/PALSAR mosaic imageries were undertaken within the framework of the Japan Aerospace Exploration Agency (JAXA) Kyoto & Carbon Initiative, and through the joint collaboration of Fauna & Flora International Philippines, the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH, and the University of the Philippines – Department of Geodetic Engineering (UP-DGE). ALOS/PALSAR data have been provided by the JAXA Earth Observation Research Center (EORC). This study was implemented through the Mainstreaming Indigenous People’s Participation in Environmental Governance (MIPPEG) Project of Foundation for the Philippine Environment, Inc. (FPE) which funded by European Union – Fundacion Desarollo Sustenido (EU- FUNDESO). The authors wish to thank our colleagues at Fauna & Flora International Philippines, our project partners, and especially the local and indigenous Sibuyan Mangyan Tagabukid communities who have been integral in the conduct of the resource socio-economic assessments in Sibuyan Island. The authors are also grateful to Drs. Shimada and Rosenqvist for their continued leadership at JAXA and the K&C Initiative.

References

ALMEIDA‐FILHO, R. SHIMABUKURO, Y. E. ROSENQVIST, A. & SÁNCHEZ, G. A. (2009) Using dual‐polarized ALOS PALSAR data for detecting new fronts of deforestation in the Brazilian Amazônia,” International Journal of Remote Sensing, 30 (14): 3735–3743.

BROWN, R.M. & DIESMOS, A.C. 2009. Philippines, Biology; p. 723-732 In R. Gillespie D. Clague (eds.). Encyclopedia of Islands. Berkeley: University of California Press.

DENR-MIMAROPA. 2014. Mt. Guiting-guiting Natural Park. Available at: http://mimaropa.denr.gov.ph/index.php/mt- guiting-guiting-natural-park [accessed 12.06.14]

ELITH, J., GRAHAM, C. H., ANDERSON, R. P., DUDIK, M., FERRIER, S., GUISAN, A., HIJMANS, R. J., HUETTMANN, F., LEATHWICK, J. R., LEHMANN, A. & OTHERS. 2006. Novel methods improve prediction of species’ distributions from occurrence data, Ecography, 29(2): 129–151.

FOODY, G.M. 2002. Status of land cover classification accuracy assessment. Remote Sensing of Environment, 80, 185–201. KASISCHKE, E. S., MELACK, J. M. & DOBSON, M. C. 1997. The use of imaging radars for ecological applications: a review. Remote Sensing Environment, 59(2): 141–156.

KUMMER, D.M. 1992. Remote sensing and tropical deforestation: a cautionary note from the Philippines. Photogrammetric Engineering and Remote Sensing, 58, 1469–1471.

MALLARI, N. A., COLLAR, N. J. LEE, D. C., MCGOWAN, P. J., WILKINSON, R. & MARSDEN, S. J. 2011. Population densities of understorey birds across a habitat gradient in , Philippines: implications for conservation, Oryx, 45(2): 234–242.

MAYER, A. L., & LOPEZ, R. D. 2011. Use of Remote Sensing to Support Forest and Wetlands Policies in the USA. Remote Sensing, 2(3): 1211-1233.

ONG, P.S., AFUANG, L.E., & ROSELL-AMBAL, R.G (eds.) 2002. Philippine Biodiveristy Conservation Priorities: A Second iteration of the National Biodiveristy Strategy and Acton Plan. Department of Environment and Natural Resources- Protected Areas and Management Wildlife Bureau, Conservation International Philippines, Biodiversity Conservation Program University of the Philippines Center for Integrative and Development Sutidies and Foundation for the Philippine Environment, Quezon City, Philippines. RAHMAN, M. M. & SUMANTYO, J. T. S. 2010. Mapping tropical forest cover and deforestation using synthetic aperture radar (SAR) images. Applied Geomatics. 2(3): 113–121.

PETTINATO, S., PALOSCIA, S., & SANTI, E. 2014. The Potential of SAR Images in Identifying Forest Characteristics. In ForestSAT2014 Open Conference System. Available at: http://ocs.agr.unifi.it/index.php/forestsat2014/ForestSAT2014/paper/ viewPaper/87 [accessed 14.06.14]

ROSENQVIST, A., SHIMADA, M., LUCAS, R. M., CHAPMAN, B. D., PAILLOU, P., HESS, L.L. & LOWRY, J. 2010. The Kyoto & Carbon Initiative - a brief summary, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 3(4): 551–553.

ROSENQVIST, A., SHIMADA, M., ITO, N. & WATANABE, M. 2007. ALOS PALSAR: a pathfinder mission for global- scale monitoring of the environment. IEEE Transactions on Geoscience and Remote Sensing, 45, 3307–3316. SHIMADA, M. AND OHTAKI. T. 2010. Generating large-scale high-quality SAR mosaic data sets: Application to PALSAR data for global monitoring, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 3(4): 637-656

STEHMAN, S. V. 1997. Selecting and interpreting measures of thematic classification accuracy, Remote Sensing of Environment, 62(1): 77 – 89.

SWEDISH SPACE CORPORATION. 1988. Mapping of the Natural Conditions of the Philippines. Swedish Space Corporation, Solna, Sweden, Final Report FSF101.

TONGSON, E. & MCSHANE, T. 2006. Securing Indigenous Rights and Biodiversity Conservation through Partnerships in Sibuyan Island, Romblon, Philippines. Policy Matters. pp. 286-296

WALKER, W. S., STICKLER, C. M., KELLNDORFER, J. M., KIRSCH, K. M. & NEPSTAD, D.C. 2010. Large-area classification and mapping of forest and land cover in the Brazilian Amazon: a comparative analysis of ALOS/PALSAR and Landsat data sources, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 3(4): 594–604.

WHEELER, J., TANSEY, K., BALZTER, H. AND NICOLAS-PEREA, V. 2014. Deforestation map of the Congo basin from 1996 to the present. GMES Initial Operations – Network for Earth Observation Research Training - European Centre o f E x c e l l e n c e i n E a r t h O b s e r v a t i o n R e s e a r c h T r a i n i n g . A v a i l a b l e a t : http://www.gionet.eu/wp- content/uploads/2014/06/Deforestation-map-of-the-Congo-basin-from-1996-to-the-present.pdf [accessed 14.06.14]