Detecting an area affected by forest fires using ALOS PALSAR Keiko Ishii (1), Masanobu Shimada (2), Osamu Isoguchi (2), Kazuo Isono (1) (1)Remote Sensing Technology Center of Japan (2)Japan Aerospace Exploration Agency 3 rd ALOS Joint PI Symposium, Nov. 2009, Kona, Hawaii
Contents Objectives Study Areas and PALSAR Datasets Analysis Methods Results Summary 2
Objectives To find a specific polarization in L-band SAR best suited for detecting fire scars To assess the availability of the fine beam dual polarization mode (FBD) or polarimetric mode (PLR) of ALOS PALSAR for forest fires. 4
Study Areas and PALSAR Datasets (1) Location Evia Island, Greece California, USA (Buckweed Fire) Fire Event 2007.8.24 2007.10.25 PALSAR Data (before the fire) PALSAR Data (after the fire) Off-nadir Angle, Ascend./ Descend. 2007.7.17 (FBD) 2006.9.8(PLR) 2007.9.1 (FBD) 2007.10.27 (PLR) 34.3A 21.5A RSP Path No. 620 212 6
Study Areas and PALSAR Datasets (2) Location (1) Victoria, Australia (East Region) (2) Victoria, Australia (Central Region) Fire Event 2009.2.7 2009.2.7 PALSAR Data (before the fire) PALSAR Data (after the fire) Off-nadir Angle, Ascend./ Descend. 2008.5.31 (FBD) 2008. 10. 21 (FBD) 2009.3.3 (FBD) 2009.3.8 (FBD) 34.3A 34.3A RSP Path No. 379 382 2 1 7
Analysis Methods Generate color composite image before and after fires To check visually whether some changes are detected or not Calculate differences of backscatter coefficients between before and after fires Input data: geo-coded images (not ortho-rectified) Calculation of the backscatter coefficients (σ 0 ) from the amplitude images (DN) before and after fires σ 0 = 10*log 10 <DN 2 > + CF (db), CF = -83.0 Change detection from the difference of backscatter coefficients Δ = σ 0 a σ0 b (db) σ 0 a : σ0 after a fire, σ 0 b : σ0 before a fire, Δ : difference of backscatters Compare PALSAR images with the image of AVNIR-2 or MODIS To confirm if the changes of PALSAR images are burnt areas or not 9
Evia Island, Greece (2007) (1) Areas in the yellow circles in the image of HV polarization are more reddish than in HH polarization 11
Evia Island, Greece (2007) (2) Dark areas in the yellow circles correspond to the red areas in RGB images The areas of HV polarization are darker than that of HH polarization Volume scattering decreased after the fires 12
Evia Island, Greece (2007) (3) In the AVNIR-2 image, vegetation was lost in the same areas as the changed areas in the PALSAR image Burnt Areas 13
Evia Island, Greece (2007) (4) Comparison of the backscatter coefficients before and after fires -5-10 13 3 2 1 4 5 12 Backscatter (db) -15 Burnt areas -20 Mean = -1.02 SD = 0.52 Un-burnt areas Mean = 0.21 SD = 0.16 20070717 HH Pol. 20070901 HH Pol. 14 6 7-25 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Area No. 11 No.1-7: Burnt areas No.8-14: Un-burnt areas 8 10 9 Differences -5 of HV polarization are larger than those of HH polarization in the changed areas. Back Scatter (db) -10-15 -20 Mean = -2.31 SD = 0.59 Mean = 0.12 SD = 0.17 20070717 HV Pol. 20070901 HV Pol. -25 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Area No. 14
Buckweed, California (2007) (1) : The multi-temporal AVNIR-2 images show that change in this area is not the burned scar caused by the fire on Oct. 25, 2007. HV polarization was more effective to detect fire scars than HH and VV polarization 16
Buckweed, California (2007) (2) Dark areas in the yellow circles correspond to the red areas in RGB images Volume scattering decreased after the fires 17
Buckweed, California (2007) (3) AVNIR-2 Images before and after the fire No changes were detected (Before the fire) (After the fire) This area may be far slope from the antenna 18
Buckweed, California (2007) (4) Comparison of the backscatter coefficients before and after fires -5-5 -10-10 Back Scatter (db) -15-20 Mean = -0.75 SD = 0.75 20060908 HH Pol. 20071027 HH Pol. -15 Mean = 0.17 SD = 0.23 Mean = -0.75-20 SD = 0.74 Back Scatter (db) Mean = 0.19 SD = 0.24 20060908 VV Pol. 20071027 VV Pol. -25 1 2 3 4 5 6 7 8 9 Area No. 6 7 1 2 3 4 5 8 No.1-5: Burnt areas No.6-9: Un-burnt areas Back Scatter (db) -25 1 2 3 4 5 6 7 8 9-5 -10-15 Area No. In the burnt areas, differences of HV polarization are larger than those of HH and VV polarizations. Mean = -2.96 SD = 1.27 Mean = 0.02 SD = 0.28 20060908 HV Pol. 20071027 HV Pol. -20 9-25 1 2 3 4 5 6 7 8 9 Area No. 19
East Region of Victoria, Australia (2009) (1) Forest region changed similarly in the images of both HH and HV polarizations. 21
East Region of Victoria, Australia (2009) (2) 22
East Region of Victoria, Australia (2009) (3) Region A is distinguished a little in the PALSAR difference image. Region B is not distinguished clearly in both PALSAR images. 23
East Region of Victoria, Australia (2009) (4) Comparison of the backscatter coefficients before and after fires -5-10 5 1 Back Scatter (db) -15 Mean = -1.62-20 SD = 0.62-25 Mean = -2.59 SD = 0.54 20080531 HH Pol. 20090303 HH Pol. 3-30 1 2 3 4 5 6 7 Area No. 4 7 6 2 Back Scatter (db) -5 Mean = -1.84-10 SD = 0.85 Mean = -1.53 SD = 0.64-15 -20 20080531 HV Pol. 20090303 HV Pol. -25 No.1-2: Burnt areas No.3-7: Un-burnt areas -30 1 2 3 4 5 6 7 Area No. In the burnt areas, differences of both HH and HV polarizations are similar. 24
East Region of Victoria, Australia (2009) (5) Multi-temporal Variation of Backscatter Coefficients on Burnt / Un-burnt Areas using four data between Aug. 29, 2007 and Mar. 3, 2009 HH Pol. HV Pol. -5-10 -7-12 -9-14 Backscatter (db) -11-13 -15-17 1 2 3 4 5 Backscatter (db) -16-18 -20-22 1 2 3 4 5-19 -24-21 -26-23 20070829 20080531 20080831 20090303 Observation Date -28 20070829 20080531 20080831 20090303 Observation Date No.1-2: Burnt areas No.3-5: Un-burnt areas Backscatter coefficients of Aug. 29, 2007 and May 31, 2008 show a slight difference Backscatter coefficients of Aug. 31, 2008 are higher than former two observation dates Backscatter coefficients of March 3, 2009 are lower than others Seasonal variation? They have to be assessed further. 25
Central Region of Victoria, Australia (2009) (1) A B C In the image of HV polarization, region A, B and C changed in the forest regions, but in the image of HH polarization, most of the forest regions (and farmlands) changed. 27
Central Region of Victoria, Australia (2009) (2) 28
Central Region of Victoria, Australia (2009) (3) The regions A and B in PALSAR image are similar to the burnt areas in the MODIS image. The region C can not be seen in MODIS image because of cloud cover, however can be seen in PALSAR image. 29
Central Region of Victoria, Australia (2009) (4) Comparison of the backscatter coefficients before and after fires -5-10 Backscatter (db) -15-20 -25 Mean = -0.66 SD = 0.90 Mean = -0.25 SD = 0.70 20081021 HH Pol. 20090308 HH Pol. -30 In the 1 burnt 2 3areas, 4 5differences 6 7 8 of 9HV 10 Area No. polarization are larger than those of HH polarization. -5-10 Backscatter (db) -15-20 -25 Mean = -1.93 SD = 0.64 Mean = -0.53 SD = 0.88 20081021 HV Pol. 20090308 HV Pol. No.1-5: Burnt areas No.6-10: Un-burnt areas -30 1 2 3 4 5 6 7 8 9 10 Area No. 30
Summary From the results of the analyses in the Greek, Californian and Australian forest fires, fire scars are detected clearly by using HV polarization. HV polarization is more sensitive to detect fire scars than HH and VV polarizations. In the east region of Victoria, a fire scar is detected in HV polarization, but other un-burnt mountain areas also changed in both HH and HV polarizations. It may be caused by seasonal dependency of backscatter variations, but it has to be investigated further. In some cases, it is difficult to detect fire scars or to identify cause of changes, dependent on the topography and land use. Ex. far-slope areas of mountainous region and farmland. Future Work: The data of other forest fires have to be analyzed to increase the case studies of PALSAR data, to clarify issues and to confirm the further availability of cross-polarization data of PALSAR. 31