PALSAR Full-Polarimetric Observation for Peatland

Size: px
Start display at page:

Download "PALSAR Full-Polarimetric Observation for Peatland"

Transcription

1 PALSAR Full-Polarimetric Observation for Peatland M. Watanabe 1*, K. Kushida 2, C. Yonezawa 3, M. Sato 1 and M. Fukuda 4 1 Center for North East Asian Studies, Tohoku University, Kawauchi 41, Aoba-ku, Sendai, Miyagi , Japan 2 Center for Far Eastern Studies, University of Toyama, Gofuku 3190, Toyama-shi, Toyama , Japan 3 The Graduate School of Agricultural Science, Tohoku University, 1-1 Amamiya-machi, Tsutsumidori, Aoba-ku, Sendai, Miyagi , Japan 4 Fukuyama City University, Minatomachi 2, Fukuyama-shi, Hiroshima , Japan Abstract PALSAR full-polarimetric observation was carried out on May 27, 2010, for peatlands in Palangka Raya, Indonesia, and the field experiment was carried out at the end of September PALSAR analysis shows that σ 0 is larger than HH σ0, and this could not be explained by a one-layer radar reflection scattering model from bare soil. The field experiment shows that there is no effective vertical moisture variation under the surface to contribute to the σ 0 value observed with PALSAR in the peatland. We conclude that a small amount of vegetation, i.e., 2.4 tons/ha, covering the ground should be taken into account to explain the radar reflection from the peatlands, even if the SAR frequency is high, e.g., L-band. Key words: L-band, SAR, Soil moisture, Peatland fire 1. Introduction Tropical peat (including swamps and forests) found on islands in the Indonesian and Malaysian Archipelagos, in the Amazon lowlands, and in Central Africa comprise some 42 million ha, and they are estimated to store approximately 148 Gt of carbon. Because peatlands are found to be one of the earth s major carbon sinks and stores, estimating the amount of carbon, and management of the peatlands, have attracted much more interest recently. When investigating peatlands, the synthetic aperture radar (SAR) is found to be advantageous because the data is collected on cloudy or rainy days, which occur frequently in tropical areas. Owing to the Mega Rice Project (MRP), Indonesia is threatened by large-scale deforestation, canal drainage, and forest fires, which cause enormous carbon emissions (Goldammer, 1999). The MRP in Central Kalimantan 2011 AARS, All rights reserved. * Corresponding author involved the conversion of over 1 million ha of peat swamp forest to rice fields and opened up the ecosystem to free logging and the use of fire for deforestation. The irrigation and drainage systems associated with the MRP also aggravated the impact of fire. Several studies have been carried out to monitor the peatlands. Tetuko, et al. (2003) developed a model for the relationship between the backscattering coefficient and the thickness of a burnt coal seam. The model was applied to estimate the thickness of a burnt coal seam in a central Borneo fire event that occurred in 1997, using Japanese Earth Resources Satellite (JERS-1) SAR data. D. Hoekman and M. Vissers (2007) use the time series of historical JERS- 1 SAR data, and they reveal the extent and nature of recent disturbances, such as those caused by excess drainage and severe El Niño Southern Oscillation in Indonesia. Wetland class characterizations were carried out using polarimetric SAR. Pope et al. (1994, 1997) have shown that the phase

2 PALSAR Full-Polarimetric Observation for Peatland difference between the HH and polarizations is the most useful parameter for flooded wetland classification and detection of seasonal flooding wetland changes. Kasischke et al. (2011) show that significant linear correlations were found between the log of above ground biomass (ranging between 0.02 and 22.2 tons/ha) and both σ 0 (L-HH) and σ 0 (L- HV) for the phased array type L-band synthetic aperture radar (PALSAR) data taken from fire-disturbed black spruce forests in the Alaskan interior. In this paper, radar scattering mechanisms for peatlands are examined in order to develop a moisture estimation algorithm. It is known that fire occurrence and volumetric water content of a soil are closely related, and the possibility of a fire occurring becomes high when the volumetric water content of a soil is low. Estimating the moisture map within a global area is one of the prospective applications for satellite data. L-band SAR, such as PALSAR, launched in 2006, uses a moderate wavelength to estimate the volumetric water content of a soil within a global area. Unlike previous SAR, PALSAR not only has a single/dual polarization mode with a 10 m resolution, but also has a full polarimetric mode with a 30 m resolution. The former mode is useful when high resolution monitoring of an area is required, and the later mode is useful to determine the mechanism of radar scattering from the earth s surface. PALSAR full-polarimetric mode data are used to understand the radar reflection from peatland. 2. Radar Reflection from Soil Radar reflection from soil is described by three parameters: dielectric constant (ε), root mean square (RMS) height of the ground surface (s), and correlation length (l), where s and l are usually normalized by the wave number, k = (2π/λ), and represented by ks and kl. Soil moisture (Mv) is calculated from the well-known empirical equation developed by Topps (1980). By using these parameters, a theoretical model, the IEM model, was developed by Fung (1994). For cases where (ks) (kl) < 1.2 (ε), the backscattering coefficient can be calculated using the analytical formula: k z = kcosθ and k x = ksinθ where θ is a local incident angle. p and q represent polarization (h or v). W (n) is the Fourier transformation of the n th power of the surface correlation function, and the exponential, Gaussian, and 1.5-power forms of this function are well known. I n qp is a function of k z, s, ε, and Fresnel s reflection coefficient. Based on this model, Shi et al. (1997) provided an estimation of soil moisture and surface roughness parameters from the full polarimetric SAR data in the L-band. (1) (2) Oh (2004) proposed a semi-empirical model in which some parameters were tuned using GB-polarimetric scatterometers and AIRSAR data taken over various soil conditions. Three parameters are defined in this model: Both models indicate that σ 0 is always more than σ0 HH, if the reflection comes from a single layer rough surface, without any vegetation, and the validation range of the volumetric water content of a soil is at most less than 40%, because the sensitivity of σ 0 to the moisture is low beyond this. The IEM model and the Oh model describe the full polarimetric radar reflection taken with PALSAR from a ground surface (Watanabe et al., 2010). 3. Field Experiment and PALSAR Observation The field experiment was carried out from September 22 to 26, The test site, Region 1, is located 30 km from Palangka Raya at longitude E and latitude S (Figure. 1). There is a small degree of vegetation on the peat soil [Figure. 2(b)]. Surface roughness parameters, ks and kl, were measured using a needle profilometer with a length of 1.5 m and a needle interval of 2 cm. Soil moisture was measured using time domain reflectometry (TDR). We dug several 15 cm holes in Region 1 and measured the volumetric water content of the soil at depths of 5 and 15 cm to check whether the volumetric water content is dependent on soil depth. The volumetric water content ranged from 33.4% to 51.2%. Since σ 0 shows little sensitivity in this Table1. Parameters derived from a field experiment Volmetric water content (@ 0 cm) Volmetric water content (@ 5 cm) Volmetric water content (@ 15 cm) Dielectric constant ( ) Surface roughness (ks) Correlation length (kl) 38.8% 33.4% 51.2% Surface biomass 2.4 tons/ha

3 Asian Journal of Geoinformatics, Vol.11,No.3 (2011) moisture range, we conclude that the volumetric water content observed in the vertical direction make a small contribution to the σ 0 obtained using PALSAR in the peatland. Surface dry biomass was estimated from the grasses in Region 1. We selected eleven 2 2 m sub-plots from which all the vegetation was cut and collected. Relative density was assumed to be 0.4, and the averaged biomass in Region 1 was estimated from the eleven subplots. Table 1 shows the summary of the field parameters collected in the field. PALSAR full-polarimetric observation was carried out on May 27, The flight direction was ascending, and the off-nadir angle was , corresponding to an incident angle of Level 1.1 and 1.5 data processed by the Japan Aerospace Exploration Agency (JAXA) was used for the analysis. The full polarimetry data were analyzed using the Polarimetric SAR Data Processing and Educational Tool (PolSARpro) (Pottier,2009), and level-1.1 PALSAR data were converted to a coherence matrix. Two of the unsupervised classification techniques, the four component decomposition model (Yamaguchi,2001) and the H/α classification scheme (Cloud,1997), were applied to the data. The four component decomposition image and site photos are presented in Figures. 1 and 2. Five peatland areas (Regions 1, 2, 3, 5, and 6) were selected in the scene. Few trees were observed in Regions 1, 2, and 3, and the field experiment was carried out in Region 1. One forest area (Region 4) was selected near Region 1. One bare soil area (Region 7) was selected in the Palangka Raya airport, where strong surface scattering reflection was observed in the four component decomposition image shown in Figure. 3. The parameters derived from PALSAR data are summarized in Table 2. Figure 1. Four component decomposition image (red: double bounce scattering, green: volume scattering, and blue: surface scattering), position of test site, and peatland, where radar parameters are picked up Table 2. Parameters derived from PALSAR data Reg. 1 Reg. 2 Reg. 3 Reg. 4 Reg. 5 Reg. 6 Reg. 7 Reg. 8* Peatland Fores t Peatland Bare soil Peatlan d 0 HH (db) VH (db) (db) Entropy ( ) PDouble (%) PVolume (%) PSurface (%) PHelix (%) * Pixels, where is less than 20, are used to calculate the value.

4 PALSAR Full-Polarimetric Observation for Peatland M. Watanabe et al./asian Journal of Geoinformatics Vol. xx No. xx (2010) Figure Fig. 2. (a)2avnir-2 imageimage of Regions 1-4, and 8.and The 8. field experiment wasetcarried out Region 1. (e)xx(f) Four (a) AVNIR-2 of Regions 1-4, The fieldm.experiment was carried out in Watanabe al./asian Journal ofingeoinformatics Vol.(c) xx No. (2010) Region 1. (c) (e) (f) image. Four component image. (red: double bounce scattering, component decomposition (red: doubledecomposition bounce scattering, green: volume scattering, blue: surface scattering). 0 the figures represent green: volume scattering, blue: surface scattering). Yellow polygons Yellow polygons in the figures represent the areas, where parameters, such as σin are calculated. (b) (d) Site photos 0 the areas, where parameters, such as 4. Unsupervised Classification Image are calculated. (b) (d) Site photos. Figures. 1 and 2(c),(e), (f) are images of the four component decomposition model. Red, green, and blue colors are used to represent double bounce, volume, and surface scattering components respectively. An image obtained using an Advanced Visible and Near Infrared Radiometer type 2 (AVNIR-2) is shown in Figure. 2(a). Forest areas, which result in volume scattering, are represented by green, and urban areas, which result in double bounce scattering, are colored red (Figure. 1). Peatland produces surface scattering, represented by blue, or weak backscattering, represented by black. Complex color patterns are observed in the peatland. Several double bounce scattering component areas are observed in the peatland, as shown in Regions 5 and 6 of Figures. 2(e) and 2(f). A possible cause of the double bounce scattering component observed in the peatland is the presence of patchy forestation. This may cause a double bounce reflection from the ground surface and the tree trunks. This kind of information is never extracted from single- or dualpolarization data, which is the merit of full polarimetry mode data. Quantitative discussion is presented in the next section. 3. Four component image and Fig. 3Figure Four component decomposition image and decomposition position of site 7 (left). Yellow polygons in position ofthesite (left). Yellow in the Photofigure around the figure represents areas,7 where parameters, suchpolygons as 0 are calculated. 0 represents the site 7 (right). the areas, where parameters, such as σ are calculated. Photo around the site 7 (right)

5 Asian Journal of Geoinformatics, Vol.11,No.3 (2011) 5. Radar Scattering from Peatland σ 0 is calculated from the IEM model for co-polarization, and the Oh model for cross-polarization. The surface roughness parameters of ks, kl, and ε, taken in the field experiment, were inserted into the models. The calculated σ 0 is compared with the value of σ 0 derived from the PALSAR data, and the results are summarized in Table 3. The σ 0 for co-polarization shows a small db difference between the PALSAR data and the IEM model. However, the σ 0 for cross-polarization shows an 7 db difference between them. Another important characteristic observed in the PALSAR data is that the σ 0 HH is always larger than the σ 0 for the peatland, while the bare soil area of Region 7 shows almost the same value. As mentioned in Section 2, and as the model-predicted σ 0 in Table 3, σ 0 should always be more than σ0 for the case HH of radar backscattering from single layer surface scattering. The PALSAR data indicates that radar backscattering from the peatland could not be represented by a single layer model. Two possible scenarios might explain this situation. One is that peatland consists of two underground layers with different dielectric constants (or volumetric water content of a soil). The other is that vegetation on the surface affects the radar reflection from the ground. No effective vertical moisture variation was observed for the field experiment in Region 1. Therefore, we conclude that small amounts of vegetation, where the biomass is 2.4 t/ha, affects the radar reflection, so that the observed σ 0 HH is larger than the σ0. The larger discrepancy of the σ 0 between PALSAR data VH and the model is also explained by a small amount of vegetation. Two possible scenarios might explain this situation. Scenario 1) Layer 1 magnifies the surface roughness of layer 2. Even if there is little vegetation, the surface parameters magnify the roughness (ks) and correlation length (kl) of the ground surface by a factor equal to the square root of the dielectric constant. For example, if the dielectric constant of the vegetation layer is assumed to be 3, the surface roughness and the correlation length are 1.7 times larger than the original. It magnifies the radar backscattering from db to db for σ 0, -5.5 db to -4.7 db for VH σ0, and -3.5 db HH to -2.6 db for σ 0 (Table 3). The presence of a vegetation layer is more significant when considering cross polarization data. Scenario 2) Volume scattering by the vegetation layer. As pointed out by Kasischke et al. (2011), significant linear correlations were found between the log of the surface biomass (ranging from 0.02 to 22.2 tons/ha) and σ 0 (L-HV) for the PALSAR data. They suggest the following experimental formula: Table 3. σ 0 derived from PALSAR data at Region 1, and calculated from IEM (co-polarization) and Oh models (like-polarization) 0 HH (db) 0 VH (db) 0 (db) BM a BM is the log of surface biomass, and a and b are parameters which are decided by the PALSAR data. By using this formula with a = 0.214, and b = 4.11, which show the highest correlation in their data, σ 0 is estimated to be db. HV One or both of the scenarios may explain the discrepancy between the σ 0 derived from PALSAR observation and that derived from the one layer surface scattering model. Larger α values observed in the peatland also support this conclusion. The α value is expected to be near 0 0 for simple surface scattering, and values of to are reported in bare soil areas using PALSAR full-polarimetric observation (Watanabe,2010). Bare soil areas of Region 7 also show a low α value of On the other hand, the observed α values for the peatland show relatively large values ranging from to This also supports the fact that the single surface scattering model could not describe the radar reflection from peatland, and a small amount of vegetation, such as 2.4 tons/ha, may affect the backscattering, even if the radar frequency is L-band. Next, we selected the pixels where α is less than 20 0 in Region 1 and calculated the σ 0. The results are presented in Table 2 (Region 8). As we expected, the differences between the σ 0 HH and the σ0 are reduced from ~3dB to ~1dB. This means that single surface scattering is more dominant in these pixels, while the σ 0 HH is still larger than the σ Conclusion PALSAR Model 0 HV ( =1) Model ( =3) b (6) The PALSAR full-polarimetric mode data was analyzed, and the derived σ 0, HH σ0, and σ0 were compared with IEM and VH Oh models. These models are two representative models for describing radar backscattering from rough surfaces, and surface parameters measured in the field are inserted to derive the σ 0. The σ 0 for co-polarization show a few db difference between PALSAR data and the IEM model. On

6 PALSAR Full-Polarimetric Observation for Peatland the other hand, σ 0 for cross-polarization show 7 db difference between them. The observed σ 0 with PALSAR shows that σ 0 is larger than HH σ0. Because of the following three reasons, we conclude that this discrepancy between the PALSAR data and the model is due to the existence of small amounts of vegetation cover on the peatland, and vegetation of 2.4 tons/ha cannot be neglected, even if the frequency band is L-band. 1. Vertical structure of the moisture observed in the peatland gives a small contribution to the σ 0 observed with PALSAR. 2. The discrepancy between the PALSAR data and the models is a few db for co-polarization, and 7 db for crosspolarization. These may be explained by the presence of a vegetation layer, which magnifies especially in the case of cross-polarization, and/or causes volume scattering. 3. A larger α value ranging from to indicates that the scattering from the peatland cannot be explained by a single layer surface scattering model. Acknowledgments The field measurements in Indonesia were funded by the Japan International Cooperation Agency and the Japan Science and Technology Agency (JICA and JST). This work was supported by the Japan Society for the Promotion of Science (JSPS) Grant-in-Aid for Scientific Research (S) , (B) , and (C) References Cloud S. R., and Pottier E. (1997), An entropy based classification scheme for land applications of polarimetric SAR, IEEE Transactions on Geosciences and Remote Sensing, 35(1), Fung A. K. (1994), Microwave scattering and emission models and their applications, Artech House Goldammer J. G. (1999), Forest on Fire, Science, 284, Hoekman D., and Vissers M. (2007), ALOS PALSAR radar observation of tropical peat swamp forest as a monitoring tool for environmental protection and restoration, IGARSS 2007, /IGARSS , Kasischke E. S., Tanase M. A., Bourgeau-Chavez L. L., and Borr M. (2011), Soil water content limitations on monitoring boreal forest regrowth using spaceborne L-band SAR data, Remote Sensing of Environment, 115, Oh Y. (2004), Quantitative retrieval of soil moisture content and surface roughness from multipolarized radar observations of bare soil surfaces, IEEE Trans. Geosci. Remote Sensing, 42(3), Pope K. O., Rey-Benayas J. M., and Paris J. F. (1994), Radar remote sensing of forest and wetland ecosystems in the Central American tropics, Remote Sens. Environ., 48(2), Pope K. O., Rejmankova E., Paris J. F., and Woodruff R. (1997), Detecting seasonal flooding cycles in marches of the Yucatan Peninsula with SIRC polarimetric radar imagery, Remote Sens. Environ., 59(2), Pottier E., Ferro-Famil L., Allain S., Cloude S., Hajnsek I., Papathanassiou K., Moreira A., Williams M., Minchella A., Lavalle M., Desnos Y. (2009), Overview of the PolSARpro V4.0 software. the open source toolbox for polarimetric and interferometric polarimetric SAR data processing, Proceedings of IEEE International Geoscience and Remote Sensing Symposium, 4, Shi J., Wang J., Hsu A. Y., O Neill P. E., and Engman E. T. (1997), Estimation of Bare Surface Soil Moisture and Surface Roughness Parameter Using L-band SAR Image Data, IEEE Trans. Geosci. Remote Sensing, 35(5), Tetuko S. S. J., Tateishi R., and Takeuchi N. (2003), A physical method to analyse scattered waves from burnt coal seam and its application to estimate thickness of fire scars in central Borneo using L-Band SAR data, Int. J. Remote Sensing, 24(15), Topp G. C., Davis J. L., and Annan A. P. (1980), Electromagnetic determination of soil water content, Water resource Res., 16, Yamaguchi Y., Moriyama T., Ishido M., and Yamada H. (2001), Four-component scattering model for polarimetric SAR image decomposition, 43(8), Watanabe M., Kadosaki G., Kim Y., Ishikawa M., Kushida K., Sawada Y., Tadono T., Fukuda, M., and Sato. M. (2010), Observation of active layer in permafrost using the PALSAR/full polarimetry mode, IEEE Trans. Geosci. Remote Sensing, Submitting.