Home Articles Land Cover Mapping Using ALOS PALSAR Data Over Penang Island, Malaysia

Land Cover Mapping Using ALOS PALSAR Data Over Penang Island, Malaysia

C. K. Sim
PhD Scholar
Universiti Sains Malaysia
Malaysia.
[email protected],

K. Abdullah
Lecturer
Universiti Sains Malaysia
Malaysia.
[email protected]

M. Z. MatJafri
Universiti Sains Malaysia
Malaysia.
[email protected]

H. S. Lim
Universiti Sains Malaysia
Malaysia.
[email protected]

Abstract
Remote sensing technology offers a wide variety of digital imagery that makes it extremely interesting to develop monitoring systems capable of regularly updating land-cover maps. The objective of this study is to access the capability of Advanced Land Observation Satellite (ALOS) Phase Array L-type Synthetic Aperture Radar (PALSAR) data on land cover mapping over Penang Island, Malaysia. This paper presents the basic information of the project, the status of the research and preliminary result including data acquisitions, data processing and data analysis. ASF MapReady programs from Alaska satellite Facility Geographical Institute at the University of Alaska Fairbanks was used for the preprocessing of ALOS-PALSAR data. Standard supervised classification techniques such as the maximum likelihood, minimum distance-tomean, and parallelepiped were applied using the same training areas derived from high resolution optical satellite imagery.

Filtering and enhancement methods had to be applied in order to reduce speckle noise and to contrast the images. Composite color images were produced for visual interpretation and field surveys. After investigation of the ground truth, representative areas of each land cover type were identified and allocated to the images. The ALOS-PALSAR data of training areas were choose and selected based on the high resolution optical satellite imagery and was classified using supervised classification methods. The land cover information was extracted from the digital data (HH and HV Polarization) bands using PCI Geomatica 10.1 software package. An accuracy assessment was also carried out in this study. High overall accuracy 82.5% and Kappa coefficient 0.70 was achieved by the Maximum Likelihood classifier (HH+HV Polarization) in this study. Finally maximum likelihood classifier (HH+HV Polarization) was used to classify the land features into a land cover map. This study indicated that the land cover of Penang Island, Malaysia can be mapped accurately using ALOS-PALSAR data.

1. INTRODUCTION
Quantitative assessment of land cover is important for a country to make proper planning, and in global scale the database will be helpful to understand the trends of Earth surface alteration and its linkage to the climate change. Synthetic Aperture Radar (SAR) both independently and jointly with optical sensors are suitable to prepare land cover maps [1]. Optical remote sensing data such as LANDSAT TM and SPOT have been successfully applied in certain parts of Malaysia, particularly in land use/ land cover mapping, land use change detection etc. However, the use of only optical imagery is unreliable under conditions of continued cloud cover that can persist over large areas of the earth’s surface [2].

In recent years, researchers have been investigating the use of longer wavelength radar imagery to obtain additional land cover information. Radar imagery provides reliable data under cloud and haze conditions at any time of the year. These are the major reasons that the SAR data are very popular to use in land cover mapping. Commercial and experimental SAR data are available from European Resource Satellite 1/2 (ERS-1/2), Envisat ASAR, SIR A, B or C, Radarsat-1/2, Japanese Earth Resource Satellite-1 (JERS-1), Advanced Land Observation Satellite (ALOS) Phase Array L-type Synthetic Aperture Radar (PALSAR) etc. The potential of SAR in land cover analysis has been clearly demonstrated [3, 4].

The objective of this study was to identify the land cover/use feature over Penang Island, Malaysia. This research investigated the multi polarized data of ALOS-PALSAR data for land cover/use mapping. Supervise classification technique was applied to the digital satellite images. The monitoring task can be accomplished by supervised classification techniques, which have proven to be effective categorization tools [5]. Post-classification of accuracy assessment was carried out in this study.

2. MATERIALS AND METHODS

2.1 Descriptions of Study Area
The study area is the Penang Island, Malaysia, located within latitudes 5o 12’ N to 5o 30’ N and longitudes 100o 09’ E to 100o 26’ E. The map of the region is shown in Figure 1. The satellite image was acquired on 1 November 2007.

2.2 Data Sets
Geocoded ALOS-PALSAR L-band polarimetric data with 12.5m spatial resolution and 21.5 degree incident angle recorded on 1 November 2007 was used in the analysis of land cover classification in Penang Island, Malaysia. Figure 2 shows the raw satellite image. The data has two different modes: HH, and HV polarization (Table 1).

Table 1. The characteristics of data-sets

A Landsat TM satellite image of 128/56 (path/row) on 8 February 2007 was use for interpretation and validation purposes.

Digital elevation model (DEM) of Shuttle Radar Topographic Mission (SRTM, https://srtm.usgs.gov) 90 m resolution elevation data was used to geometric correction ALOS PALSAR data.

2.3 Preprocessing
The ASF MapReady program from Alaska satellite Facility Geographical Institute at the University of Alaska Fairbanks was used for a radiometric correction, geometric correction and terrain correction to the ALOS-PALSAR data. After converting DN to sigma-nought, we can obtain radar backscattering coefficients. Then we calculated a geometric terrain correction (orthorectifies) using SRTM data to remove artifacts commonly seen in SAR data such as layover and shadow. We applied a medium filter with a 5×5 window size to reduce speckle noise of ALOS-PALSAR images.

2.4 Classification
Standard supervised classification techniques such as the maximum likelihood, minimum distance-to-mean, and parallelepiped were used for land cover classification. Every single acquisition mode was classified using the same training areas derived from optical satellite data. The satellite image was classified into 3 classes namely forest, water and urban.

2.4 Validation
A total of 200 samples were chosen randomly over the whole study area by using a high resolution Landsat 5 satellite data. Accuracy assessments determined the correctness of the classified map. Two method of accuracy assessments were the overall classification accuracy and kappa coefficients.
RESULTS AND DICUSSION
The classified map produce by each classifier was checked with the confusion or error matrix and kappa statistic. The results obtained are show in Table 2. The Maximum Likelihood classifier (HH+HV Polarization) produced the highest accuracy with overall classification accuracy of 82.5% and Kappa coefficient of 0.70. A classified image using Maximum Likelihood classifier (HH+HV Polarization) is shown in Figure 3. An increase of the overall classification accuracy and kappa coefficient was obtained with dual-mode data (HH+HV) in comparison with the single band input data (HH, HV).

Table 2. The overall classification accuracy and kappa coefficient

CONCLUSION
The result of this preliminary study show that analyzing the accuracies of the single band input data in comparison to the multi-mode data. We clearly detected the use of multimode data indicate an increase accuracy in land cover identification. As the result of this study the Maximum Likelihood classifier (HH+HV Polarization) produced the highest degree of accuracy.

ACKNOWLEDGEMENT
This research is conducted under the agreement of JAXA Research Announcement titled ‘2nd ALOS Research Announcement for the Advanced Land Observation Satellite between the Japan Aerospace Exploration Agency and the Research – The use of ALOS data in studying environmental changes in Malaysia’ (JAXA – 404). The author would like to express special thanks to Alaska satellite Facility Geographical Institute at the University of Alaska Fairbanks for providing the ASF MapReady programs free software use in this study. Thanks are extended to USM for support and encouragement.

REFERENCES
[1] M. Mahmudur Rahman , Josaphat Tetuko Sri Sumantyo. ALOS PALSAR data for tropical forest interpretation and mapping. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008, Microwave Remote Sensing Laboratory, Center for Environmental Remote Sensing, Chiba University. P. 185-190.

[2] Roebig, J.H., E. Hardy, R. Bryant, and B, Guetti. 1984. SPOT potential for land use/land cover classification in using image enhancement and computer processing. Proc. SPOT Symposium, Scottsdale, Arizona, U.S.A., p. 251-258.

[3] Dobson, M.C., L.E. Pierce, and F.T. Ulaby. 1996. Knowledge-based land cover classification using ERS-1/JERS-1 SAR composites. IEEE Transactions on Geoscience and Remote Sensing. 34:83-99.

[4] Henderson, F.M. and A.J. Lewis (eds.). 1998. Manual of Remote Sensing, Vol. 2: Principles and Applications of Imaging Radar. Wiley: New York, U.S.A.

[5] Bruzzone, L., Cossu, R. and Vernazza, G. (2002).Combining parametric and non-parametric algorithms for a partially unsupervised classification of multitemporal remote-sensing images. Information Fusion 3, 289 –297.