A tasseled cap transformation for CBERS-02B CCD data

A tasseled cap transformation for CBERS-02B CCD data 1 2 first-author 1 2 corresp ‡Corresponding Author ‡Corresponding Author †E-mail:hjf@zju.edu.cn †E-mail:hjf@zju.edu.cn hjf@zju.edu.cn 3 Institute of Agricultural Remote Sensing and Information Application, Zhejiang University, Hangzhou 310029, China Institute of Agricultural Remote Sensing and Information Application, Zhejiang University, Hangzhou 310029, China Key Laboratory of Agricultural Remote Sensing and Information System of Zhejiang Province, Hangzhou 310029, China Key Laboratory of Agricultural Remote Sensing and Information System of Zhejiang Province, Hangzhou 310029, China Yuhang Branch Bureau of Planning, Hangzhou Urban Planning Bureau, Hangzhou 311100, China Yuhang Branch Bureau of Planning, Hangzhou Urban Planning Bureau, Hangzhou 311100, China The tasseled cap transformation of remote sensing data has been widely used in agriculture, forest, ecology, and landscape. In this paper, tasseled cap transformation coefficients appropriate for data from a new sensor (China & Brazil Earth Resource Satellite (CBERS-02B)) are presented. The first three components after transformation captured 98% of the four-band variance, and represent the physical characteristics of brightness (coefficients: 0.509, 0.431, 0.330, and 0.668), greenness (coefficients: −0.494, −0.318, −0.324, and 0.741), and blueness (coefficients: 0.581, −0.070, −0.811, and 0.003), respectively. We hope these results will enhance the application of CBERS-02B charge-coupled device (CCD) data in the areas of agriculture, forest, ecology, and landscape. Tasseled cap transformation CBERS-02B CCD Reflectance factor Introduction 1 The tasseled cap transformation has provided a mechanism for data volume reduction and enhanced data interpretability by emphasizing the structures in the multispectral imagery (Crist, 1985 1985 ). The transformation is sensor-dependent, which means that the transformation coefficients for one sensor cannot be applied successfully to another sensor. The tasseled cap transformation was first applied to Landsat multispectral scanner (MSS) data (Kauth and Thomas, 1976 1976 ), but has been extended to other sensors, e.g., thematic mapper (TM) (Crist and Cicone, 1984 1984 ; Crist, 1985 1985 ), enhanced thematic mapper plus (ETM+) (Huang et al., 2002 2002 ), IKONOS (Horne, 2003 2003 ), and moderate-resolution imaging spectroradiometer (MODIS) (Zhang et al., 2002 2002 ; Lobser and Cohen, 2007 2007 ). The tasseled cap transformation turns original, highly covariant data into three uncorrelated indices called brightness, greenness, and wetness (the third tasseled cap transformation components for MSS was called yellow stuff), which can reflect the state of vegetation and soil. In recent years, these three indices have been widely studied and successfully used in studies of agriculture, forest, ecology, and landscape. For instance, the wetness component was used to characterize the dynamics of irrigated crops (Serra and Pons, 2008 2008 ); tasseled cap indices were used to estimate the age of forest (Wulder et al., 2004 2004 ); tasseled cap indices can improve the accuracy of mapping and land cover classification (Dymond et al., 2002 2002 ); a linear combination of tasseled cap indices called the disturbance index can produce significantly more accurate disturbance maps in Russia (Healey et al., 2005 2005 ); and, wetness and brightness indices were proven to reflect qualitatively both fuel moisture and its distribution, the most important determinants of fire propagation in savanna areas (Mbow et al., 2004 2004 ). China & Brazil Earth Resource Satellite (CBERS)-02B was launched in September 2007 in China. The sensor charge-coupled device (CCD) carried by CBERS-02B obtains five bands (one panchromatic band and four multispectral bands) with 19.5 m spatial resolution. In this paper, we aim to present a tasseled cap transformation coefficient appropriate for CBERS-02B CCD data. Materials and methods 1 Preparation of CBERS-02B CCD data 2 The relative spectral response curves of CBERS-02B CCD data are shown in Fig. 1 1 . The CBERS-02B CCD data can be downloaded from the website http://www.cresda.com http://www.cresda.com for free. The level 2 products employed in our study were radiometrically and geometrically corrected and provided by the China Centre for Resources Satellite Data and Application. We selected 10 scenes of CBERS-02B CCD data that captured a variety of land cover types and growing stages. The data preprocessing consisted of two major steps. First, we converted the raw digital number (DN) values to at-sensor radiance. Calculation of at-sensor spectral radiance is necessary in order to normalize the data into a physically meaningful radiometric scale (Chander et al., 2009 2009 ) that also compensates for the temporal variation of irradiance that can adversely affect the derived tasseled cap values (Huang et al., 2002 2002 ). Second, atmospheric correction was performed using a 6S model. The 6S input parameters include geographic conditions (acquisition date, sun zenith, sun azimuth, viewing zenith, and viewing azimuth), atmospheric conditions, aerosol model, ground visibility, target altitude, sensor altitude, spectral band conditions, ground reflectance characteristics, and atmospheric correction mode. The 10-scene data information and main 6S parameters are listed in Table 1 1 . Geographic conditions were from data information file; ground visibility was from meteorological data; target altitude was from digital elevation model (DEM); and, spectral band conditions were defined by the user. Study shows that the atmospheric model in the region of latitude about 40° N in China is close to US62, so here we chose US62 for Huairou in spring and autumn The scene locates in the midlatitude area in summer or winter 2 Path 2 Row 2 Geographic location 2 Acquisition data 3 6S input parameter Atmospheric model Aerosol model Ground reflectance 367 65 Suzhou Nov. 29, 2008 Midlatitude winter * * Continental Non-homogenous 367 65 Suzhou June 29, 2008 Midlatitude summer * * Continental Non-homogenous 370 64 Chaohu Nov. 20, 2008 Midlatitude winter * * Continental Non-homogenous 370 64 Chaohu Jan. 11, 2009 Midlatitude winter * * Continental Non-homogenous 373 55 Huairou Apr. 17, 2008 US62 # # Continental Non-homogenous 373 55 Huairou Sept. 22, 2009 US62 # # Continental Non-homogenous 1 55 Beijing Aug. 22, 2008 Midlatitude summer * * Urban Non-homogenous 1 55 Beijing Dec. 30, 2008 Midlatitude winter * * Urban Non-homogenous 373 66 Wuhan May 13, 2008 Midlatitude summer * * Urban Non-homogenous 373 66 Wuhan Nov. 11, 2008 Midlatitude winter * * Urban Non-homogenous The following equation is used to covert DN value to at-sensor spectral radiance: L =DN/ a , where L is spectral radiance at the sensor’s aperture [W/(m 2 ·sr·μm)]; DN is digital number value of level 2 product; and, a is the calibration coefficient from the data information file. The top-of-atmosphere (TOA) is computed as follows: ρ TOA =π LD 2 /( E cos θ ), where ρ TOA is TOA reflectance (unitless); π is mathematical constant ≈3.14159 (unitless); L is spectral radiance at the sensor’s aperture [W/(m 2 ·sr·μm)]; D is Earth-Sun distance (astronomical unit); E is mean exoatmospheric solar irradiance [W/(m 2 ∙μm)], shown in Table 2 2 ; and, θ is solar zenith angle (°). From http://www.cresda.com http://www.cresda.com Band E (W/(m 2 ·μm)) 1 1 919 2 1 812 3 1 523 4 1 042 5 1 636 In the 6S model, the relationship between TOA reflectance and target reflectance can be expressed as: $${\rho _{TOA}}({\theta _S},{\theta _V},{\phi _S} - {\phi _V}) = {\rho _a}({\theta _S},{\theta _V},{\phi _S} - {\phi _V}) + \frac{\rho }{{1 - \rho S}}T({\theta _S})T({\theta _V})$$ , where θ S , θ V , ϕ S , and ϕ V are solar zenith angle, view zenith angle, solar azimuthal angle, and view azimuthal angle, respectively; ρ TOA ( θ S , θ V , ϕ S − ϕ V ) is TOA reflectance; ρ a ( θ S , θ V , ϕ S − ϕ V ) is path radiance reflectance; ρ is the reflectance of target; S is the spherical albedo of the atmosphere; T ( θ S ) is the total downward transmittance; and, T ( θ V ) is the total upward transmittance. Method of tasseled cap transformation 2 The basic idea of tasseled cap transformation is compressing spectral data into a few bands associated with physical scene characteristics. Principal component analysis is frequently used to evaluate the data dimensionality. The first two principal components will define a plane very similar to the tasseled cap transformation, and can be rotated to provide features essentially equivalent to the tasseled cap features (Crist and Cicone, 1984 1984 ). As a first step, principal component transformation coefficients of the 10 scenes containing Bands 1–4 were calculated in ENVI4.3, and a general set of vectors was calculated to determine the initial axes. The approach was to calculate least-squares fitting of the 10 principal component transformation coefficients by minimizing the residual sum of squares ( χ ): $$\chi = \sum\limits_{A = 1}^{10} {\sum\limits_{i = 1}^4 {\sum\limits_{j = 1}^4 {{{(PC_{ij}^A - {R_{ij}})}^2}} } }$$ . The matrix R is a unitary matrix. The matrix PC A is principal component transformation coefficients for Image A . The second step was to rotate the initial axes to the tasseled cap features. At early stage, this was an exploratory, visual process that Crist and Cicone ( 1984 1984 ) described rotation method in this way: “these components were then rotated, two or three at a time, in a linear fashion which preserved the orthogonality of the six components. By the process of applying various rotations, the data relationships in the TM data space were discovered and defined.” It is obvious that when determining the tasseled cap transformation for a new sensor by this method, the rotating is deemed most appropriate by the investigator. Lobser and Cohen ( 2007 2007 ) put forward that the TM tasseled cap transformation is so widely used and well established that it can be referenced as rotation standards which are developed through the relationship between tasseled cap features and their corresponding biophysical parameters. Thus, the process of rotation we adopted was visually interpreted by reference to the TM tasseled cap. First, based on prior knowledge regarding important scene characteristics or data structures of tasseled cap transformation (Table 3 3 ) and considering the fourth component was mainly noise, we made assumptions that brightness and greenness can be obtained and tried to find the characteristics for the third component. Second, tasseled cap indices would discriminate different land cover types, and an attempt had been made to represent a wide variety of land cover types in the TM tasseled cap. Hundreds of vegetation, water, soil, and manmade material samples were selected from each scene to guide the rotation process. MSS: multispectral scanner; TM: thematic mapper; ETM+: enhanced thematic mapper plus; MODIS: moderate-resolution imaging spectroradiometer; CBERS: China & Brazil Earth Resource Satellite; NIR: near-infrared; SWIR: short wave-infrared; TC: tasseled cap transformation component 2 Sensor 5 Band width (μm) 3 Tasseled cap tranformation result Blue Green Red NIR SWIR TC 1 TC 2 TC 3 MSS 0.50–0.60 0.60–0.70 0.70–0.80, 0.80–1.10 Brightness Greenness Yellowness TM/ETM+ 0.45–0.52 0.52–0.60 0.63–0.69 0.76–0.90 1.55–1.75, 2.08–2.35 Brightness Greenness Wetness IKONOS 0.44–0.51 0.51–0.59 0.63–0.70 0.78–0.85 Brightness Greenness Undefined MODIS 0.46–0.48 0.54–0.56 0.62–0.67 0.84–0.87, 1.23–1.25 1.63–1.65, 2.10–2.15 Brightness Greenness Wetness CBERS-02B 0.45–0.52 0.52–0.59 0.63–0.69 0.77–0.89 To make this method easily understood, here we assumed a three-dimensional (3D) space to illustrate the process of transformation (Fig. 2 2 ). The original data with three bands ρ 1 , ρ 2 , and ρ 3 , after principal component transformation, are turned into PC 1 , PC 2 , and PC 3 (the first, second, and third principal components, respectively). PC 3 is mainly noise, so rotations were performed in the plane of PC 1 -PC 2 with the orthogonality of PC 3 axis. Under the guide of samples, angle θ rotation is applied to shift PC 1 and PC 2 axes to TC 1 and TC 2 axes (the first and second tasseled cap transformation components). ρ 1 , ρ 2 , and ρ 3 are the original data axes; PC 1 and PC 2 are the initial axes after transformation; TC 1 and TC 2 are the final axes after a series of rotations ( θ ) of PC 1 and PC 2 In the course of PC axes to TC axes, rotations are applied in a 3D space (the fourth component was mainly noise) which preserves the orthogonality of three axes. A series of small angle rotations were made based on the following equations: $$({R'_1},{R'_2},{R'_3},1) = ({R_1},{R_2},{R_3},1)\left[ {\begin{array}{*{20}{c}}{{\text{ }}\cos \gamma {\text{ }}\sin \gamma {\text{ }}0{\text{ }}0} \\{ - \sin \gamma {\text{ }}\cos \gamma {\text{ }}0{\text{ }}0} \\{{\text{ }}0{\text{ }}0{\text{ }}1{\text{ }}0} \\{{\text{ }}0{\text{ }}0{\text{ }}0{\text{ }}1} \\\end{array} } \right],$$ , $$({R'_1},{R'_2},{R'_3},1) = ({R_1},{R_2},{R_3},1)\left[ {\begin{array}{*{20}{c}}{1{\text{ }}0{\text{ }}0{\text{ }}0} \\{0{\text{ }}\cos \alpha {\text{ }}\sin \alpha {\text{ }}0} \\{0{\text{ }} - \sin \alpha {\text{ }}\cos \alpha {\text{ }}0} \\{0{\text{ }}0{\text{ }}0{\text{ }}1} \\\end{array} } \right],$$ , $$({R'_1},{R'_2},{R'_3},1) = ({R_1},{R_2},{R_3},1)\left[ {\begin{array}{*{20}{c}}{\cos \beta {\text{ }}0{\text{ }}\sin \beta {\text{ }}0} \\{{\text{ }}0{\text{ }}1{\text{ }}0{\text{ }}0} \\{ - \sin \beta {\text{ }}0{\text{ }}\cos \beta {\text{ }}0} \\{{\text{ }}0{\text{ }}0{\text{ }}0{\text{ }}1} \\\end{array} } \right],$$ , where $${R'_1}$$ , $${R'_2}$$ , and $${R'_3}$$ are the axis coefficients after angle rotation of the original axis coefficients R 1 , R 2 , and R 3 ; and α , β , and γ are rotation angles in three planes, respectively. Results and discussion 1 Analysis of CCD data 2 The CBERS-02B CCD relative spectral response curves show that there are overlaps between the multispectral bands (Fig. 1 1 ), which means that they have correlations. The correlation coefficients among the four multispectral bands of the 10 scenes of CCD data were calculated, and the means of these correlations are shown in Table 4 4 . Note that there are high correlations among the three visible region bands ( ρ 1 , ρ 2 , and ρ 3 ), while ρ 4 in the near-infrared (NIR) region is substantially less correlated with the other bands. These moderate to high inter-band correlations demonstrate that there is a lot of redundancy in the original CCD data. The tasseled cap transformation components are uncorrelated and contain enhanced biophysical information. 2 Reflectance 4 Correlation coefficient ρ 1 ρ 2 ρ 3 ρ 4 ρ 1 1.000 0.980 0.981 0.396 ρ 2 0.980 1.000 0.978 0.477 ρ 3 0.981 0.978 1.000 0.417 ρ 4 0.396 0.477 0.417 1.000 Tasseled cap transformation components 2 Table 5 5 gives the tasseled cap transformation coefficients for CBERS-02B CCD data based on reflectance factor. The first three components can extract 98% of the total variance of the original data and represent the physical characteristics of brightness (TC 1 ), greenness (TC 2 ) and blueness (TC 3 ), respectively. 2 Component 4 Coefficient ρ 1 ρ 2 ρ 3 ρ 4 TC 1 0.509 0.431 0.330 0.668 TC 2 −0.494 −0.318 −0.324 0.741 TC 3 0.581 −0.070 −0.811 0.003 Fourth −0.449 0.845 −0.285 −0.051 Brightness (TC 1 ) is the weighted sum of four bands, but is dominated by the contribution from the near-infrared (NIR) band. Greenness (TC 2 ) is the contrast between the three visible bands and the NIR band, which captures the characteristic green vegetation reflectance pattern: high absorption and low reflectance in the visible bands contrasted with the high reflectance in the NIR band. Blueness (TC 3 ), this component contrasts strong blue reflectance and strong red absorptance. The very weak contributions from the green and NIR bands more or less cancel one another. The blue plastic roof (Fig. 3 3 ), due to its high reflectance in the blue band and high absorption in the other bands, shows brightest in TC 3 . Discussion of the tasseled cap transformation 2 A 400×400 extract from the CBERS-02B CCD image of the Suzhou area, acquired in November 2008, is shown as an example to explain the tasseled cap transformation results (Fig. 4 4 ). Both TC 1 (Fig. 4a 4a ) and the panchromatic image (Fig. 4d 4d ) display similar physical characteristics—object brightness, but TC 1 enhances the contrast of the data and eliminates noise compared to the panchromatic image. The CBERS-02B panchromatic band captures very little radiance from the blue and NIR portions of the spectrum. TC 1 , on the other hand, is dominated by NIR radiance with a strong contribution from the blue band. In TC 1 , buildings tend to be bright, dense vegetation is intermediate, and water is dark. In TC 2 (Fig. 4b 4b ), vegetation tends to be bright, while buildings and water are both dark. In TC 3 (Fig. 4c 4c ), the blue plastic roofs (Fig. 3 3 ), mainly in the industrial area, appear the brightest in TC 3 . (a) TC 1 ; (b) TC 2 ; (c) TC 3 ; (d) Panchromatic image The 2D scatter diagrams of Figs. 4a–4c 4a–4c are illustrated in Fig. 5a 5a . Several key land cover types are plotted on these three planes in Fig. 5b 5b . Conclusions 1 An appropriate suggested set of tasseled cap transformation coefficients for the CBERS-02B CCD data is presented. The transformation is based on reflectance factor. The first three components after transformation capture 98% of the four-band variance, and represent the physical characteristics of brightness, greenness, and blueness, respectively. We hope these results will enhance the application of CBERS-02B CCD data in the areas of agriculture, forest, ecology, and landscape. References G Chander , BL Markham , DL Helder . Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors . Remote Sens Environ , 2009 . 113 ( 5 ): 893 - 903 . DOI: 10.1016/j.rse.2009.01.007 http://doi.org/10.1016/j.rse.2009.01.007 . Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors EP Crist . A TM tasseled cap equivalent transformation for reflectance factor data . Remote Sens Environ , 1985 . 17 ( 3 ): 301 - 306 . DOI: 10.1016/0034-4257(85)90102-6 http://doi.org/10.1016/0034-4257(85)90102-6 . A TM tasseled cap equivalent transformation for reflectance factor data EP Crist , RC Cicone . A physically-based transformation of Thematic Mapper data—the TM tasseled cap . IEEE Trans Geosci Remote Sens , 1984 . 22 ( 3 ): 256 - 263 . DOI: 10.1109/TGRS.1984.350619 http://doi.org/10.1109/TGRS.1984.350619 . A physically-based transformation of Thematic Mapper data—the TM tasseled cap CC Dymond , DJ Mladenoff , VC Radeloff . Phenological differences in tasseled cap indices improve deciduous forest classification . Remote Sens Environ , 2002 . 80 ( 3 ): 460 - 472 . DOI: 10.1016/S0034-4257(01)00324-8 http://doi.org/10.1016/S0034-4257(01)00324-8 . Phenological differences in tasseled cap indices improve deciduous forest classification SP Healey , WB Cohen , ZQ Yang , ON Krankina . Comparison of tasseled cap-based Landsat data structures for use in forest disturbance detection . Remote Sens Environ , 2005 . 97 ( 3 ): 301 - 310 . DOI: 10.1016/j.rse.2005.05.009 http://doi.org/10.1016/j.rse.2005.05.009 . Comparison of tasseled cap-based Landsat data structures for use in forest disturbance detection JH Horne . A Tasseled Cap Transformation for IKONOS Images . 2003 . ASPRS Annual Conference Proceedings . Alaska, USA . A Tasseled Cap Transformation for IKONOS Images C Huang , B Wyli , L Yang , C Homer , G Zylstra . Derivation of a tasseled cap transformation based on Landsat 7 at-satellite reflectance . Int J Remote Sens , 2002 . 23 ( 8 ): 1741 - 1748 . DOI: 10.1080/01431160110106113 http://doi.org/10.1080/01431160110106113 . Derivation of a tasseled cap transformation based on Landsat 7 at-satellite reflectance RJ Kauth , GS Thomas . The Tasseled Cap—A Graphic Description of the Spectral-Temporal Development of Agricultural Crops as Seen by Landsat . 1976 . Proceedings of the Symposium on Machine Processing of Remotely Sensed Data . West Lafayette, Indiana 41 - 51 . . The Tasseled Cap—A Graphic Description of the Spectral-Temporal Development of Agricultural Crops as Seen by Landsat SE Lobser , WB Cohen . MODIS tasselled cap: land cover characteristics expressed through transformed MODIS data . Int J Remote Sens , 2007 . 28 ( 22 ): 5079 - 5101 . DOI: 10.1080/01431160701253303 http://doi.org/10.1080/01431160701253303 . MODIS tasselled cap: land cover characteristics expressed through transformed MODIS data C Mbow , K Goıta , GB Benie . Spectral indices and fire behaviour simulation for fire risk assessment in savanna ecosystems . Remote Sens Environ , 2004 . 91 ( 1 ): 1 - 13 . DOI: 10.1016/j.rse.2003.10.019 http://doi.org/10.1016/j.rse.2003.10.019 . Spectral indices and fire behaviour simulation for fire risk assessment in savanna ecosystems P Serra , X Pons . Monitoring farmers’ decisions on Mediterranean irrigated crops using satellite image time series . Int J Remote Sens , 2008 . 29 ( 8 ): 2293 - 2316 . DOI: 10.1080/01431160701408444 http://doi.org/10.1080/01431160701408444 . Monitoring farmers’ decisions on Mediterranean irrigated crops using satellite image time series MA Wulder , RS Skakun , WA Kurz , JC White . Estimating time since forest harvest using segmented Landsat ETM+ imagery . Remote Sens Environ , 2004 . 93 ( 1-2 ): 179 - 187 . DOI: 10.1016/j.rse.2004.07.009 http://doi.org/10.1016/j.rse.2004.07.009 . Estimating time since forest harvest using segmented Landsat ETM+ imagery XY Zhang , CB Schaaf , MA Friedl , 等 . MODIS Tasseled Cap Transformation and Its Utility . 2002 . Proceedings of the International Geosciences and Remote Sensing Symposium . Toronto, Canada 1063 - 1065 . . MODIS Tasseled Cap Transformation and Its Utility The authors wish to express a gratitude to the China Centre for Resources Satellite Data and Application for CBERS-02B level 2 data. 10.1631/jzus.B1100088 Project (No. 40871158/D0106) supported by the National Natural Science Foundation of China 2011-03-21 2011-08-18 2011-07-26 2022-01-05 Copyright © Zhejiang University and Springer-Verlag Berlin Heidelberg 2011 2011 Journal of Zhejiang University-SCIENCE B (Biomedicine and Biotechnology) 09 12