Vegetation classification on Prathong Island Phang Nga Thailand

Vegetation classification on Prathong Island, Phang Nga, Thailand Naiyana Srichai & Chanida Suwanprasit Faculty of Technology and Environment, Prince of Songkla University, Phuket Campus APAN 33 rd Meeting 13 -17 February 2012

2 Introduction • vegetation type study date back to the Nineteenth Century : ecologists, plant geographers, vegetation scientists • three major determinants of vegetationcompetition, stress and disturbance (Grime, 1974)

3 Objectives • To classify vegetation on Prathong Island, Phang Nga province, southern Thailand

4 Study area: Prathong Island, Phang Nga THAILAND 7 th biggest island 1. 5 km off the coast Size : width 9. 7 km length 15. 4 km Area : 92 sq. km Unseen Thailand 2002 deer, hornbill, adjutant stork, green turtle, dugong

5 Source: Dept. of Marine and Coastal Resources, 2005 Wild orchids Local plants Local vegetables 79 spp. 96 spp. 65 spp.

Source: Dept. of Marine and Coastal Resources, 2005 11 Mammals spp. 86 Reptiles spp. 137 Birds > 20 Freshwater animals 6

Source: Dept. of Marine and Coastal Resources, 2005 Koh Ra Pak Jok 19 households 109 people 87 households 134 people Koh Ra Tong Dab village 49 households 272 people Tha Paeyow 123 households 409 people Koh Prathong 7

Source: Dept. of Marine and Coastal Resources, 2005 8 Koh Ra and Prathong Size 71, 000 Rais or 92 sq. km Mangrove Beach forest Swamp forest 32% (green) 7% (orange) 13% (pink) Tropical forest 13% (Koh Ra, purple) Grassland 8% (yellow) Beach Seagrass 26 km (orange) 4, 550 Rais (blue) Coral 43 Rais (lighter green)

9 Swamp forest Grassland Mangrove forest Beach forest

10 Tsunami 26 Dec. 2004 Area affected : 18. 55 % (6. 25% agricultural, 92. 88% others)

11 Vegetation change after Tsunami Fragile land Salt tolerant tree invasion Casuarina equisetifolia

12 Data set: THEOS Multispectral Achieved on 19 Jan 2009 Spatial Resolution 15 m Spectral Band Wavelength ( m) Band 0 (Blue) 0. 45 -0. 52 Band 1 (Green) 0. 53 -0. 60 Band 2 (Red) 0. 62 -0. 69 Band 3 (NIR) 0. 77 -0. 90

13 THEOS Spectral bands Band 0 (Blue) Band 1 (Green) Band 2 (Red) Band 3 (NIR)

14 Process Outline THEOS image 2009 Pre-image processing Image Classification Classes Maximum Likelihood (MLC) Support Vector Machines (SVMs) Vegetation Mapping • • • Grassland Beach forest Mangrove forest Wetland (swamp forest) Water Other

15 Support Vector Machines • SVMs : a supervised classifier, which requires training samples but SVMs are not relatively sensitive to training sample size (works with limited quantity and quality). • The SVM-based approach used a recursive procedure to generate prior probability estimates for known and unknown classes by adapting the Bayesian minimum-error decision rule (Mountrakis, et. at. 2011; Fauvel 2008).

16 Support vector machines (SVMs) : numerous applications in remote sensing. 108 relevant papers, published in 20072010. (G. Mountrakis, Jungho Im, C. Ogole, 2011)

17 Unsupervised Classification: • K-Mean • 10 Classes

18 ROI Separability Classes Grassland Beach Forest Mangrove Forest Swamp Forest Sand Water 1. 982 Mangrove Forest 2. 000 - 1. 766 1. 881 2. 000 - 1. 996 2. 000 - 1. 648 1. 997 - 2. 000 Grassland Beach Forest - Swamp forest Sand Water 1. 610 1. 959 2. 000 -

19 Classification Results MLC SVMs Grassland Swamp Forest Beach Forest Mangrove Forest Sand Water Other

20 RGB(0, 1, 2) MLC SVMs

21 Class Confusion Matrix Class MLC Prod. Acc. (%) SVMs User Acc. (%) Prod. Acc. (%) User Acc. (%) Grassland 98. 68 100. 00 96. 71 100. 00 Beach Forest 97. 26 97. 06 100. 00 97. 14 Mangrove Forest 97. 20 100. 00 99. 15 99. 39 Swamp 46. 55 56. 84 61. 21 83. 53 Water 97. 58 70. 35 97. 58 82. 31 Sand 98. 21 100. 00 99. 40 98. 82 Over all Accuracy 94. 29 % (Kappa Co. = 0. 921) 96. 72 % (Kappa Co. = 0. 954)

22 Conclusions • SVM classifier compared to the more conventional maximum likelihood approach gave slightly better accuracy using THEOS image for class : swamp forest of Prathong Island.

23 Acknowledgement • Geo-Informatics and Space Technology Development Agency (Public Organization) • Uni. Net • Prince of Songkla University, Phuket campus

24 References: • Department of Marine and Coastal Resources. 2005. Strategies for sustainable development of Koh Ra and Koh Prathong with people participation. Unpublished report. • Fauvel, M. , Benediktsson, J. A. , Chanussot, J. , Sveinsson, J. R. . 2008. Spectral and Spatial Classification of Hyperspectral Data Using SVMs and Morphological Profiles. Geoscience and Remote Sensing, 46 (11), 3804 - 3814 • Giorgos Mountrakis, Jungho Im, Caesar Ogole. 2011. Support vector machines in remote sensing: A review. ISPRS Journal of Photogrammetry and Remote Sensing, 66, 247– 259. • Grime, J. P. 1974. Vegetation classification by reference to strategies. Nature, 250 (5461), 26 -31.

25 Kob Khun Ka : Thank You