Forum Geografi, 31(1), 2017; DOI: 10.23917/forgeo.v31i1.2518

 

Temporal Decorrelation Effect in Carbon Stocks Estimation Using Polarimetric Interferometry Synthetic Aperture Radar (PolInSAR) (Case Study: Southeast Sulawesi Tropical Forest)

 

Laode Muh. Golok Jaya*,1,2, Ketut Wikantika2,3, Katmoko Ari Sambodo4, Armi Susandi5

 

1 Faculty of Engineering, Universitas Halu Oleo, Jl. HEA. Mokodompit No. 8, Kampus Hijau UHO, Anduonohu Kendari Southeast Sulawesi (Indonesia)

2 Center for Remote Sensing Institute of Technology Bandung (ITB) Jl. Ganesha No. 10, Bandung, West Java (Indonesia),

3 ForMIND Institute (Forum Peneliti Muda Indonesia)

4 Indonesia National Institute of Aeronautics and Space (LAPAN) Jl. Lapan No. 70, Pekayon Pasar Rebo, Jakarta (Indonesia)

5 Meteorology Department, Faculty of Earth Science and Technology, ITB (Indonesia), Jl. Ganesha No. 10, Bandung West Java (Indonesia)

 

*) Corresponding Author (e-mail: laode251@yahoo.com)

 

Received: 30 March 2017 / Accepted: 13 April 2017 / Published: 01 July 2017

Abstract

This paper was aimed to analyse the effect of temporal decorrelation in carbon stocks estimation. Estimation of carbon stocks plays important roles particularly to understand the global carbon cycle in the atmosphere regarding with climate change mitigation effort. PolInSAR technique combines the advantages of Polarimetric Synthetic Aperture Radar (PolSAR) and Interferometry Synthetic Aperture Radar (InSAR) technique, which is evidenced to have significant contribution in radar mapping technology in the last few years. In carbon stocks estimation, PolInSAR provides information about vertical vegetation structure to estimate carbon stocks in the forest layers. Two coherence Synthetic Aperture Radar (SAR) images of ALOS PALSAR full-polarimetric with 46 days temporal baseline were used in this research. The study was carried out in Southeast Sulawesi tropical forest. The research method was by comparing three interferometric phase coherence images affected by temporal decorrelation and their impacts on Random Volume over Ground (RvoG) model. This research showed that 46 days temporal baseline has a significant impact to estimate tree heights of the forest cover where the accuracy decrease from R2=0.7525 (standard deviation of tree heights is 2.75 meters) to R2=0.4435 (standard deviation 4.68 meters) and R2=0.3772 (standard deviation 3.15 meters) respectively. However, coherence optimisation can provide the best coherence image to produce a good accuracy of carbon stocks.  

 

Keywords: Temporal Decorrelation, Carbon Stocks, PolInSAR, ALOS PALSAR.

 

Abstrak

Penelitian ini bertujuan untuk meneliti pengaruh dekorelasi temporal dalam pendugaan cadangan karbon. Pendugaan cadangan karbon sangat penting untuk memahami daur karbon secara global in atmosfer terkait upaya mitigasi perubahan iklim. Teknik PolInSAR menggabungkan dua buah teknik pemetaan dengan citra Synthetic Aperture Radar (SAR) yaitu Polarimetric Synthetic Aperture Radar (PolSAR) dan Interferometry Synthetic Aperture Radar (InSAR) yang terbukti berkontribusi dalam teknologi pemetaan sejak beberapa tahun terakhir. Teknik PolInSAR dapat memberikan informasi struktur tinggi vegetasi yang digunakan untuk menduga besarnya cadangan karbon pada tutupan hutan. Dua buah citra ALOS PALSAR yang koheren dan full-polarisasi dengan perbedaan waktu perekaman 46 hari digunakan dalam penelitian ini. Wilayah studi penelitian ini adalah hutan tropis Sulawesi Tenggara. Metode penelitian ini adalah membandingkan tiga buah citra fase interferometrik yang dipengaruhi oleh dekorelasi temporal dan dampaknya dalam pembentukan model RVoG. Hasil penelitian ini menunjukkan bahwa adanya perbedaan perekaman citra selama 46 hari memiliki dampak yang signifikan dalam estimasi cadangan karbon dimana akurasi tinggi vegetasi berkurang dari R2=0.7525 (simpangan baku 2.75 meter) menjadi R2=0.4435 (simpangan baku 4.68 meter) dan R2=0.3772 (simpangan baku 3.15 meter). Namun demikian, dengan proses optimisasi koherensi dapat menghasilkan citra dengan nilai koherensi yang terbaik untuk menghasilkan nilai cadangan karbon dengan akurasi yang baik.

 

Kata Kunci: Dekorelasi Temporal, Cadangan Karbon, PolInSAR, ALOS PALSAR

Introduction

Estimation of carbon stocks plays an important role in the context of climate change mitigation. Forest layers absorb carbon emission from the atmosphere and store it as carbon pool (IPCC, 2007). Indonesia is one of the countries having the largest tropical forest that contains carbon stocks and gives impact to climate change. The mapping of forest carbon stocks in Indonesia will be very useful to support mitigation of climate change under the scheme of Reducing Emission from Deforestation and Forest Degradation (REDD). Some efforts have been conducted to estimate carbon stocks of Indonesian tropical forest including by using optical remote sensing (Kustiyo et al, 2015). However, optical remote sensing system does not generate optimum results such as the unclear and low quality of satellite image due to cloud cover and other meteorological conditions. Therefore, the radar remote sensing system becomes another option and has been widely applied for that purpose.   

In the model development for carbon stocks mapping, radar polarimetric (PolSAR) has been popular in the last few years. Radar polarimetric uses polarisation component to develop the relationship between forest carbon stocks and backscatter coefficient from different frequencies (Mitchell et al, 2012; Villard & Le Toan, 2014) and different sensors (Kurvonen et al, 1999). However, this method has a problem of saturation (Moreira et al, 2002; Le Toan et al, 2011). Meanwhile, radar interferometric (InSAR) has some advantages including it is not influenced by saturation. Some researchers such as Neeff et al, (2005) and Kugler et al, (2006) have discussed the contribution of the InSAR method to assess biophysical parameters in forest areas. Thus, the integration of radar polarimetry and interferometry (PolInSAR) can overcome saturation problem in the development of physical object model. This method was initially published by Cloude and Papathanassiou (1998). The PolInSAR is based on the formation of two coherence images from two difference positions and times of acquisition. It means that the PolInSAR is particularly vulnerable to temporal decorrelation (Mette, 2006; Richards, 2009). The objective of this paper was to analyse the effects of temporal decorrelation in carbon stocks estimation using ALOS PALSAR full-polarimetric image.

Research Method

Location

 

This research was conducted in Southeast Sulawesi, Indonesia. Specifically, the study are is a part of Wolasi Tropical Protection Forest (4°06’10.22”S to 4°12’15.74”S and 122°29’2.15”E to 122°30’33.4”E) and the altitude at 300-700 meter above mean sea level. The average temperature is 25 –34° and annual precipitation is 1,469 mm. The topography is mountainous with a gradient up to 10-20%. Furthermore, 20 sample plots were established in the study area with  20x20m2 in size. The map of the study area is displayed in Figure 1.

 

Figure 1. Map of study area based on SRTM data of Southeast Sulawesi, Indonesia. White rectangle indicates position of ALOS PALSAR data coverage used in this study.

 

The dominant vegetations in this area are  Eha (Castanopsis buruana), Batu-Batu (Ptemandra spp.), Dange (Dillenia sp.) and Ruruhi (Syzygium spp.). Some of trees aged between 5 (five) to older than 40 years old with a range of 5 to 40 cm in diameter. The tree heights are between 3-35 m and the density is 200-300 trees per hectare. The forest in the study are is displayed in Figure 2.

Figure 2. (a) Situation of forest in the study area, (b) one of dominant tree species: Eha (Castanopsis buruana).

 

SAR Data

Two levels of L1.1 full polarimetric ALOS PALSAR data were used in this research. The polarisations were HH, HV, VH and VV, respectively. Temporal baseline between these two images was 46 days. Data characteristics are described in Table 1. The images were divided into master and slave.

 

Table 1. ALOS PALSAR data characteristics

Description

Master Image

Slave Image

Center Frequency

1,25 GHz

1,25 GHz

Wavelength

23 cm

23 cm

Polarization

HH, HV, VH, VV

HH, HV, VH, VV

Range Resolution

9.4 m

9.4 m

Azimuth Resolution

3.8 m

3.6 m

Date of acquisition

2 May 2010

17 March 2010

Incidence Angle

23°

21.5°

Pass

Ascending

Ascending

Mode

Single Look Complex (SLC)

SLC

 

SAR Data Processing

SAR data processing is a fundamental key to obtain good quality of interferogram. Its fundamental steps consist of multilooking, polarimetric calibration, coregistration, spectral filtering and interferogram generation. It also proceeds two coherence ALOS PALSAR images from two different periods and two slightly different look angles, further referred as master and slave as presented in Figure 3 in Red Green Blue (RGB) composite.

 

Figure 3. Pairs of ALOS PALSAR images in RGB composite (HH, HV, HH/HV): (a) image assigned as master and (b) slave.

 

PolInSAR technique is an integration of the advantages of PolSAR and InSAR techniques. The coherence between two identical images in full-polarizations environment is the key of PolInSAR technique (Cloude & Papathanossiou, 1998). ALOS PALSAR image obtained on May 2nd, 2010 was assigned as master image and those on March 17th, 2010 was assigned as slave. Each image was performed as scattering matrix contains all polarisations as displayed in equation (1).

 

 

The scattering matrix, namely the S2 matrix, contains backscatter properties of the earth objects (Richards, 2009). The scattering matrix characterises the scattered wave for any polarisation of incidence wave at every image pixel. The elements of the scattering matrix are very useful in the analysis of pixels, for example, on the classification process using decomposition (Richards, 2009). The matrix component can be transformed into a vector with four elements arranged in the form of the scattering matrix-vector, called Pauli basis target vector (Cloude & Papathanossiou, 1998; Richards, 2009) as shown in Equation (2). “T” indicates transpose matrix.

 

 

For coregistration of two full polarimetric SAR coherent images, equation (2),  was transformed into 6x6 [T6] Hermitian matrix as shown in Equation (3) (Richards, 2009).

 

 

Where < > represented multi-looking operator, * is Hermitian transformation, k1 and k2 are 3D Pauli scattering vectors.  and  are standard Hermitian coherency matrix containing full-polarimetric information from each full-polarimetric image. The matrix  states new complex matrix in 3x3 dimension, which contains not only polarimetric information, but also interferometric phase between polarimetric channels in the coherent pair images (Richards, 2009). The T6 was then proceeded to estimate the coherence of interferogram and to produce the RVoG model.

Results and Discussion

Estimation of carbon stocks in Indonesia has been carried out by several researchers by using several ways, including backscatter radar to identify the characteristics of forest stand in tropical forest (Wuryanta, 2016), discrimination of mangrove ecosystem objects on the visible spectrum using spectroradiometer HR-1024 (Arfan, 2015) and calculation based on land cover changes in the Leuser Ecosystem Area (LEA) in Aceh (Hermon, 2015). However, the efforts produced uncertainty results. In this paper, we attempted to minimise the uncertainty by using PolInSAR method and to investigate the impediment aspect, namely the temporal decorrelation.

In Indonesia, the usage of PolInSAR through INDREX-II airborne campaign was initiated in 2004 (Hajnsek et al, 2005). The campaign was conducted to develop database of aboveground biomass in Kalimantan forests by using L- and P-bands in polarimetric and interferometric components using Experimental Synthetic Aperture Radar (E-SAR) of the German Aerospace Center (DLR). Radar technology is allegedly suitable to map carbon stocks in Indonesian forest environment due to the obstacles of cloud cover, smoke and haze, especially in Sumatera, Kalimantan, Sulawesi and Papua.

Temporal Decorrelation is the main problem in the PolInSAR (Mette, 2006; Richards, 2009; Meng et al, 2010). Decorrelation itself is divided into two ways. First, it refers to the critical baseline. A larger baseline means a greater phase shift leads to a more sensitive interferometer. However, there is a limit at 2π phase shift for certain pixels. If the change exceeds 2π , then it is hard to recover the inter-pixel variation in elevation (Richard, 2009). He also suggested that the > 1000m baseline  of spaceborne will inhibit to understand the flat earth variation. In relation with airborne SAR image,  Lee and Pottier (2009) explained their preference to use 10 m baseline pair in their study in order to avoid phase unwrapping problems in forest height estimation as well as 20 m baseline to gain better sensitivity in height estimation, which could be used for height estimation of lower vegetation. Second, decorrelation comes from any mechanism that leads to statistical differences between the signals received by two channels (Richards, 2009). Repeat-pass interferometry will change ground scattered in two acquisitions then the interferometric phase difference will be affected. Shortly, temporal decorrelation will lead to uncertainty in interferogram formation. However, it cannot be avoided in repeat-pass mode of SAR system, for instance ALOS PALSAR. To reduce the impact of temporal decorrelation, coherence optimisation was conducted.

Figure 4, 5, and 6 showed the coherence level between pair of images produced by coherence optimisation. Coherence value will be zero (0) for minimum called incoherence (black colour) and one (1) for maximum called fully coherence (white colour). Indicated by the histogram of coherence, Figure 4 is identified as the best coherence since the histogram value is near to one. The optimum coherence value was obtained by ground topography but dropped down in the forested area because of the residual volume component which cannot be removed (Cloude & Papathanassiou, 2003). Furthermore, the figures 4, 5, and 6 indicated the effect of temporal decorrelation on the coherence interferometric phase of a PolInSAR method to establish Random Volume over Ground (RVoG) model. RVoG is biophysical model of forest which is separated into two layers; top canopy layer and underlying topography (Cloude & Papathanassiou, 2003).  Figure 4 showed the most coherent image in compared to Figure 5 and 6, in which Figure 5 is more coherent than 6. The effect of temporal decorrelation can be observed in forest height inversion from RVoG model as showed in Figure 7 (a), (b) and (c).

 

 

Figure 4. Optimised interferometric coherence #1 (OPT_1) and its histogram.

 

  

 

Figure 5. Optimised interferometric coherence #2 (OPT_2) and its histogram.

 

  

Figure 6. Optimised interferometric coherence #3 (OPT_3) and its histogram.

 

Figure 7. Forest height from RVoG model. (a) from optimized interferometric coherence #1 (OPT_1), (b) optimized interferometric coherence #2 (OPT_2), and (c) optimized interferometric coherence #3 (OPT_3).

 

Temporal decorrelation affected not only the inaccuracy of forest heights but also the decline of the top phase height of forest canopy layer. In this research, the maximum canopy layer height decreased by 22%, from 32 m to 25 m as an effect of temporal decorrelation. Furthermore, the accuracy of forest height obtained from field survey In compared to RVoG heights also decreased.

Figure 8 (a), (b) and (c) showed the effect of temporal decorrelation on the accuracy of canopy layer heights. The accuracy decreased from R2=0.7525 (standard deviation of tree heights is 2.75 meters) to R2=0.4435 (standard deviation 4.68 meters) and R2=0.3772 (standard deviation 3.15 meters), respectively. In the repeat-pass mode, the temporal decorrelation cannot be avoided (Askne & Santoro, 2005; Lee et al., 2009). Hence, spaceborne single pass mode program, e.g., TanDEM-X has been developed to overcome the temporal decorrelation (Kugler et al., 2014). However, from this research, it can be suggested to consider the effect of temporal decorrelation in PolInSAR technique to minimise the uncertainty of carbon estimation in the context of climate change mitigation. Consequently, coherence optimisation must be conducted to get the highest accuracy of tree heights estimation (Lee & Pottier, 2009). Thus the highest accuracy carbon stocks can be gained.  

Figure 8. Temporal decorrelation effect on the accuracy of forest heights (a) from optimized interferometric coherence #1 (OPT_1), (b) optimized interferometric coherence #2 (OPT_2), and (c) optimized interferometric coherence #3 (OPT_3).

Conclusions

In this research, the effect of temporal decorrelation in carbon stocks estimation was discussed. The result indicates temporal decorrelation has an effect on the uncertainty of forest heights and carbon stocks estimation. In fact, it cannot be avoided in a repeat-pass mode such as ALOS PALSAR or other spaceborne SAR (RADARSAT, ENVISAT, etc). Temporal decorrelation reduces the accuracy of forest heights estimation. In this study, 46 days temporal baseline using ALOS PALSAR resulted in about 37.72% to 75.25% accuracy with standard deviation varies from 4.68 to 2.75 meters. The highest value of accuracy and the lowest standard deviation can be obtained through coherence optimisation. The highest coherency of phase interferometric will result in the highest accuracy of carbon stocks estimation (as shown in Figure 7 (a) and 8 (a)). Also, single pass spaceborne mode, e.g.,  TanDEM-X is developed to obtain high coherency.

Acknowledgment

We would like to thank Prof. M Shimada from JAXA Japan who provided ALOS PALSAR Full-Polarimetric. Also thanks to our colleagues at Faculty of Forestry and Faculty of Engineering Universitas Halu Oleo, CRS ITB, and ForMIND Institute (Prof. Ketut Wikantika) for the assistances in completing this work.

References

Arfan, A. (2015) Discrimination of Mangrove Ecosystem Objects on the Visible Spectrum Using Spectroradiometer HR-1024, Forum Geografi, Vol. 29 (1), pp.83 – 88.

Askne, J. and Santoro, M. (2005) Multitemporal repeat-pass SAR interferometry of boreal forest, IEEE Transactions on Geoscience and Remote Sensing, 43: 1219-1228.

Cloude, S.R and Papathanassiou, K. (1998) Polarimetric SAR Interferometry, IEEE Transactions on Geoscience and Remote Sensing, Vol. 36 No. 5, pp. 1551-1565.

Cloude, S.R and Papathanassiou, K. (2003) Three-stage inversion process for Polarimetric SAR Interferometry, IEE Proc-Radar Sonar Navigation, 150: 125-134.

Florian, K., Papathanassiou K., Hajnsek, I. and Dirk, H. (2006) Forest height estimation in tropical rain forest using Pol-Insar techniques, German Aerospace Center (DLR), IEEE 0-7803-9510-7/06.

Hajnsek, I., Kugler, F., Papathanossiou, K., Horn, R., Schieber, R., Moreira, A., Hoekman, D., Davidson, M. (2005) INDREX II – Indonesian airborne radar experiment campaign over tropical forest in L- and P- band: first results, Proceedings of Geoscience and Remote Sensing Symposium 2005 (IGRASS 2005) Page(s): 4335-4338, July 25-29, 2005, Seoul, South Korea.

Hermon, D. (2015) Estimate of Changes in Carbon Stocks Based on Land Cover Changes in the Leuser Ecosystem Area (LEA) Indonesia, Forum Geografi, Vol 29 (2), pp. 187-196.

IPCC (Inter-Governmental Panel on Climate Change) (2007) The Carbon Cycle and Atmospheric Carbon Dioxide Content.

Kugler, F., Papathanassiou K.P., Hajnsek I. (2006) Forest height estimation over tropical forest by means of polarimetric SAR interferometry. Proceedings of: VII Seminário de Atualização em Sensoriamento Remoto e Sistemas de Informações Geográficas Aplicados à Engenharia Florestal, October 2006. Curitiba: 504–512.

Kugler, F., Schulze, D., Hajnsek, I., Pretzsch, H. and Papathanassiou, K.P. (2014) TanDEM-X Pol-InSAR Performance for Forest Height Estimation, IEEE Transactions on Geoscience and Remote Sensing, DOI:10.1109/TGRS. 2013.2296533.

Kurvonen, L., Pulliainen, J., and Hallikainen, M. (1999) Retrieval of biomass in boreal forest from multitemporal ERS-1 and JERS-1 SAR images, IEEE Transactions on Geoscience and Remote Sensing, 37, pp. 198-205.

Kustiyo, O. Roswintiarti, A. Tjahjaningsih, R. Dewanti, S. Furby, J. W. (2015) Annual Forest Monitoring as Part of the Indonesia’s National Carbon Accounting System, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XL-7/W3, 36th International Symposium on Remote Sensing of Environment, Berlin, Germany.

Lee, S.K., Kugler, F., Papathanassiou, K.P., and Moreira, A. (2009) Forest Height Estimation by means of Pol-InSAR Limitations posed by Temporal Decorrelation, K&C Science Report – Phase 1, Microwaves and Radar Institute (DLR-HR), German Aerospace Center (DLR).

Lee, J. and Pottier, E. (2009) Polarimetric Radar Imaging from Basics to Applications, CRC Press, ISBN 978-1-4200-5497-2.

Le Toan, T., Quegan, S., Davidson, M.W.J., Baltzer, H., Paillou, P., Papathanassiou, K., Plummer, S., Rocca, F., Saatchi, S., Shugart, H., Ulander, L. (2011) The BIOMASS Mission: Mapping global forest biomass to better understand the terrestrial carbon cycle, Remote sensing of Environment 115, pp. 2850-2860.

Meng, W. and Sandwell DT. (2010) Decorrelation of L-band and C-band interferometry over vegetated areas in California, IEEE Transactions on Geoscience and Remote Sensing, 48: 2942-2952.

Mette, T., Papathanassiou, K.P., Hajnsek, I., Zimmermann, R. (2003) Forest Biomass Estimation using Polarimetric SAR Interferometry.

Mette, T. (2006) Forest Biomass Estimation from Polarimetric SAR Interferometry, Dissertation, Technischen Universität München.

Mitchell, A.L., Williams, M., Tapley, I., Milne, A.K. (2012) Interoperability of Multi-Frequency SAR Data for Forest Information Extraction in Support of National MRV System, IEEE Trans. on Geoscience and Remote Sensing.

Moreira, A., Papathanassiou, K., Krieger, G. (2002) Polarimetric SAR Interferometry with a Passive Polarimetric Micro-Satellite Concept, Institute of Radio Frequency Technology and Radar Systems, German Aerospace Center (DLR).

Neeff, T., Biging G.S., Dutra L.V., Freitas C.C. and Santos J.R. (2005) Modeling spatial tree pattern in the Tapajós forest using interferometric height, Revista Brasileira de Cartografia 57 (1): 1–6.

Richards, J. A. (2009) Remote Sensing with Imaging Radar, Signals and Communication Technology, ISBN: 978-3-642-02019-3, Springer.

Villard, L. and Le Toan, T. (2014) Relating P-Band SAR Intensity to Biomass for Tropical Dense Forests in Hilly Terrain: γ0 or t0?, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.

Wuryanta, A. (2016) Radar Data for Identifying the Characteristics of Tropical Forest Stands, Forum Geografi, Vol 30 (1), pp. 69-76.

 

© 2017 by the authors. Submitted for possible open access publication under the terms and conditions of the Creative Commons Attribution (CC-BY-NC-ND) license (http://creativecommons.org/licenses/by/4.0/).

 

Article Metrics

Abstract view(s): 788 time(s)

Refbacks

  • There are currently no refbacks.