A Comparison of ALOS PALSAR-2 Calibration Data by Using External DEM

Transcription

1 CEOS SAR Calibration and Validation Workshop 2016 A Comparison of ALOS PALSAR-2 Calibration Data by Using External DEM Tokyo Denki University, Japan, 7 th -9 th September 2016 *Choen KIM College of Forest Science, Kookmin Univ. Min-Gee HONG Department of Applied Information Technology, Kookmin Univ.

2 Contents I. INTRODUCTION II. STUDY AREA AND USED DATA III. METHODOLOGY IV. RESULTS V. DISCUSSION AND CONCLUSION

3 Introduction Objectives Since the backscatter of the PALSAR-2 is influenced by the terrain topography and it can be corrected through the use of external DEM. To evaluate radiometric correction influenced by the topographic effects on the calibration of PALSAR-2 data, PALSAR-2 level 2.1 data orthorectified by the 30-m SRTM DEM are compared to both the orthorectified PALSAR-2 level 1.1 data based on 5-m DEM and that 30-m DEM, by using the topographic normalized backscattering coefficient γ 0..

4 Topographic displacement of SAR sensor <Raney, 1998>

5 Study area & Used data Study area Kwangneung Experiment Forest Composite of HH+HV/HV/HH overlaid on digital forest map

6 Overview of KEF Number of Stand Stand size(min~max) Mean stand size Stand volume(min-max) Mean Stand volume 5.4~64.04 ha ha 112~467m 3 /ha m 3 /ha Stand volume (m 3 /ha) The KEF test site consists of 58 stands, which cover approximately 1,000ha and have a steep slope topography, North in latitude and East in longitude. The stand size varied between ha with an average of ha.

7 Used data <Characteristics of the used PALSAR-2> Sensor Observation mode Polarization Bandwidth Incidence angle [ ] Range resolution [m] PALSAR-2 Stripmap (High sensitive) DP (HH+HV) 42 MHz <Definition of processing levels> Level Definition Format The range and single-look azimuth compressed data is 1.1 represented by complex I and Q channels to preserve the magnitude and phase information. CEOS / GeoTIFF 2.1 Level 2.1 data is orthorectified from Level 1.1 data by using a DEM. CEOS / GeoTIFF The ALOS PALSAR-2 data taken over the Kwangneung Experiment Forest (KEF, Korea) was acquired on 7th of April 2015.

8 <Digital topographic map of a scale of 1:25,000 > <Triangulated irregular network > <Digital elevation model> External DEM acquisitions were made in two different digital topographic maps of a scale of 1 : 5,000 and 1 : 25,000, respectively. Generate a 5-m (30-m) DEM based on digital topographic map of a scale of 1:5,000 (1:25,000).

9 Methodology -Generate the Gamma-nought image based on DEM data- foreslope Ascending 30.2 farslope The gamma-nought images were classified by the directions, namely foreslope and far-slope. In the fore-slope(here, west-facing slope) corresponding to the tilted surface toward the sensor In case of the far-slope (here, eastfacing slope) corresponding to the tilted surface opposite to the sensor

10 Calibration of PALSAR-2 image using external DEM <30m DEM based on digital topographic map of a scale of 1:25,000 > <5m DEM based on digital topographic map of a scale of 1:5,000 > Once the DEM and the PALSAR-2 images have been registered, elevation and incidence angles can be calculated for each image pixel and the image data can be properly corrected. This work aim at comparison the geometric calibration of the ALOS PALSAR-2 according to DEM resolution.

11 Relationships between gamma-nought and stand volume in KEF We proposed that the topographic normalized backscattering coefficient γ 0 was applied to the relationships between SAR backscatter and field inventories-based stand stock volume.

12 Methodology -Calculation the Spearman s rank correlation- For comparison, three calibration images for radiometric correction were tested through performing the Spearman s coefficient of rank correction. The common form of the Spearman s rank correlation coefficients is described as where D i is the difference between the ranks of two variables in i th ordered pair and n is the number of paired ranks.

13 Results Adjusted ALOS PALSAR-2 HH images Adjusted PALSAR level HH image using 5m DEM Adjusted PALSAR level HH image using 30m DEM PALSAR level HH image using SRTM DEM <A Comparison of ALOS PALSAR-2 calibration images by using external DEM>

14 A comparison of gamma-nought ranges for PALSAR-2 data Using SRTM DEM Using 30m DEM Using 5m DEM [db] The dynamic range at gamma-nought based on 5-m DEM was less than those at gamma-nought based on 30-m DEM and 30-m SRTM DEM. The average of gamma-nought was similar, being approximately -7.4dB.

15 γ 0 values (db) Differences in the stand stock volume among PALSAR-2 γ 0 values based on 5-m/30-m/30-m SRTM DEM ALOS PALSAR γ 0 based on 5m DEM ALOS PALSAR γ 0 based on 30m DEM ALOS PALSAR γ 0 based on SRTM DEM Stand stock volume (m 3 ha -1 ) The smaller the sum of squared residuals will be the larger R 2

16 Compute the coefficient of determination γ 0 values = ln(stand stock volume) R 2 = ALOS PALSAR γ 0 based on 5m DEM γ 0 values = ln(stand stock volume) R 2 = ALOS PALSAR γ 0 based on 30m DEM γ 0 values = ln(stand stock volume) R 2 = ALOS PALSAR γ 0 based on SRTM DEM The coefficient of determination (R 2 ) for PALSAR-2 level 2.1 image using 30-m SRTM DEM was 0.45 smaller as compared to the PALSAR-2 level 1.1 image using 5-m DEM. Therefore, the explanatory γ 0 ability (R 2 =0.806=80.6%) of PALSAR-2 level 2.1 image using 30-m SRTM DEM was less than that (R 2 =0.851=85.1%) of PALSAR-2 level 1.1 image using 5-m DEM.

17 Calculation of Spearman s rank correlation (using 5-m DEM) n (KEF Stand number) Rank of stand volume Rank of γ 0 value D i D 2 i r s The test statistic r s = is greater than the critical value of for α=0.01

18 Calculation of Spearman s rank correlation (using 30-m DEM) n (KEF Stand number) Rank of stand volume Rank of γ 0 value D i D 2 i r s The test statistic r s = is greater than the critical value of for α=0.01

19 Calculation of Spearman s rank correlation (using 30-m SRTM DEM) n (KEF Stand number) Rank of stand volume Rank of γ 0 value D i D 2 i r s The test statistic r s = is also greater than the critical value of for α=0.01

20 Discussion and Conclusion We present the orthorectified precise data using 5-m and 30-m DEM based on digital topographic map instead of 30-m spaced SRTM DEM, which were tested on a topographic effects. In the case of SAR calibration using external DEM at the rugged forests, there are errors due to topographic effects thus, gammanought computation is necessary.

21 For the Spearman s rank-correlation test, there is sufficient evidence to conclude that the stand volume and γ 0 image from the three kind data are correlated. Compared to PALSAR-2 γ 0 ranges through the use of three kind DEMs, the dynamic range for 5-m DEM has smaller than both that for 30-m DEM and for 30-m SRTM DEM. In each of the three DEMs, the determination coefficient R 2 of 5-m DEM indicates more accurate relation between stand volume and γ 0 value than those of 30-m DEM and of 30-m SRTM DEM. As well, all three PALSAR-2 orthorectified calibration data (5-m DEM, 30-m DEM, and 30-m SRTM DEM) are appropriate (i.e. fit) for making γ 0 image.

22 for your attention

23 Spearman s rank correlation test The Spearman's rank correlation is the nonparametric version of the Pearson product-moment correlation. Spearman's rank correlation coefficient measures the strength and direction of association between two ranked variables. Spearman's rank correlation measures the strength and direction of monotonic association between two variables. A monotonic relationship is not strictly an assumption of Spearman's rank correlation. Spearman's rank correlation on a non-monotonic relationship to determine if there is a monotonic component to the association.