SCIENCE CHINA Earth Sciences • RESEARCH PAPER •
January 2017 Vol.60 No.1: 143–155 doi: 10.1007/s11430-016-0069-7
A high-accuracy method for simulating the XCO2 global distribution using GOSAT retrieval data ZHAO MingWei1, ZHANG XingYing2*, YUE TianXiang3†, WANG Chun1, JIANG Ling1 & SUN JingLu4 1
Anhui Center for Collaborative Innovation in Geographical Information Integration and Application, Chuzhou University, Chuzhou 239000, China; 2 National Satellite Meteorology Center, China Meteorological Administration, Beijing 100081, China; 3 State Key Laboratory of Resources and Environment Information System, Institute of Geographical Science and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China; 4 Anhui Institute of Economics, Hefei 230051, China Received March 30, 2016; accepted September 26, 2016; published online November 24, 2016
Abstract A high-accuracy surface modeling (HASM) method based on the fundamental theorem of surfaces, is developed to simulate XCO2 surfaces using the GOSAT retrieval XCO2 data. Two tests are designed to investigate the simulation accuracy. The first test divides the existing satellite retrieval XCO2 data into training points and testing points, and simulates the XCO2 surface using the training points while computing the simulation error using the testing points. The absolute mean error (MAE) of the testing points is 1.189 ppmv, and the corresponding values of the comparison methods, Ordinary Kriging, IDW, and Spline are 1.203, 1.301, and 1.355 ppmv, respectively. The second test simulates the XCO2 surface using all the satellite retrieval points and uses the TCCON (Total Carbon Column Observing Network) site observation values as the ture values. For the six typical TCCON sites, the HASM simulation MAE is 1.688 ppmv, and the satellite retrieval MAE at the same sites is 2.147 ppmv. These results indicate that HASM can successfully simulate XCO2 surfaces based on satellite retrieval data. Keywords HASM, GOSAT, TCCON, XCO2 surface Citation:
Zhao M W, Zhang X Y, Yue T X, Wang C, Jiang L, Sun J L. 2017. A high-accuracy method for simulating the XCO2 global distribution using GOSAT retrieval data. Science China Earth Sciences, 60: 143–155, doi: 10.1007/s11430-016-0069-7
Introduction 1. Carbon dioxide (CO2) is the anthropogenic greenhouse gas making the greatest contribution to global climate change (Buchwitz et al., 2006). A better understanding of the sources and sinks of CO2 is important to accurately predict climate change. Moreover, insufficient knowledge regarding CO2 leads to large uncertainties in future climate predictions because observation of CO2 is spatially and temporally limited around the globe (Yoshida et al., 2011). * Corresponding author (email:
[email protected]) † Corresponding author (email:
[email protected]) © Science China Press and Springer-Verlag Berlin Heidelberg 2016
Satellite measurement is one of the most effective approaches to monitor the global distributions of greenhouse gases at high spatiotemporal resolution and is expected to improve the accuracy of the source and sink estimates of these gases (Rayner and O’Brien, 2001). Several satellites have been designed specifically to measure the column-averaged dry air mole fraction of CO2 (XCO2), such as Japan’s Greenhouse Gases Observing Satellite (GOSAT), and NASA’s Orbiting Carbon Observatory-2 (OCO-2). Both missions attempt to identify cloud-free scenes for their retrievals, because clouds can cause large errors in retrieved CO2 concentrations (Baker et al., 2010). GOSAT carries a thermal and near infrared sensor earth.scichina.com link.springer.com
144
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
for carbon observation-Fourier transform spectrometer (TANSO-FTS) to collect information about the concentration of CO2 in the upper-troposphere (Chevallier et al., 2009). Column-averaged CO2 is retrieved from spectra in the shortwave infrared (SWIR) band. GOSAT observes the CO2 column-averaged concentrations under cloud-free conditions, with little disturbance from aerosols. Bias and random error are simulated under the assumption that the bias depends on the aerosol optical thickness (AOT) and surface albedo. The amount of successfully retrieved GOSAT data is corrected under the assumption of a global mean clear-sky probability of 11% (Kadygrov et al., 2009). Clouds and aerosols are major sources of disturbance in greenhouse gas observations from space because they strongly affect the equivalent optical path length (Mao and Kawa, 2004; Houweling et al., 2005). Only 3.2% and 2.7% of the measurement scenes were suitable for retrieval analysis in July 2009 and January 2010, respectively. The monthly mean fraction of measurement scenes suitable for retrieval analysis is approximately 3% (Yoshida et al., 2011), i.e., no valid observations are available for GOSAT retrieval analysis on 97% of the Earth’s surface. The GOSAT research team published monthly mean XCO2 data (data edition v01.50) at a global scale, with missing data on the CO2 surface defined as voids. Our previous work filled these voids based on HASM, and the accuracy test showed that HASM was effective in filling the voids of the global XCO2 surface (Yue et al., 2015). However, the preliminary retrieval data had a system deviation of approximately 8 ppmv (Yoshida et al., 2013); thus, the research team published the new retrieval data edition (v02.11). In contrast to the previous version, the new data set is not for surfaces with voids; instead, it is a data set with validation data points. Therefore, spatial interpolation must be performed based on the data points to simulate the global distribution of XCO2. Since the successful launch of the GOSAT satellite, studies on CO2 concentration and its variation has been conducted under the support of the retrieval products. The global land area distribution of XCO2 was computed based on the GOSAT-XCO2, and the uncertainty was also estimated (Hammerling et al., 2012). Studies have analyzed the correlation between GOSAT-XCO2 and ecological variables (such as GPP, NPP) and built a regression formula for the XCO2 concentrations in different regions of the world, demonstrating that the distribution of XCO2 follows specific rules (Guo et al., 2012). In addition, the seasonal variation of the XCO2 concentration in East Asia and its influencing factors have been studied (Shim et al., 2013; Guo et al., 2013; Liu et al., 2012); and additional studies focused on the temporal and spatial variation of XCO2 in China (Zeng et al., 2013; Xu et al., 2013; Lei et al., 2014). Although many studies on the temporal and spatial variation of XCO2 have been conducted, most used classic inter-
polation methods to obtain the distribution of XCO2 based on discrete XCO2 retrieval, and the accuracy was not sufficient to estimate the carbon sources and sinks. Therefore, research focused on improving the accuracy of the XCO2 distribution is essential. The spatial interpolation precision of HASM has been proved to be superior to the classic methods; therefore HASM is used to simulate the distribution of XCO2 at the global scale, and to analyze its temporal and spatial variation in this paper.
Study 2. data Satellite 2.1 data The GOSAT spacecraft, launched in January 2009, is dedicated to measuring carbon dioxide (CO2) and methane (CH4), using the thermal and near-infrared sensor for carbon observation Fourier transform spectrometer (TANSO-FTS). The TANSO-FTS detects gas absorption in the shortwave infrared (SWIR) and thermal infrared (TIR) region of the spectrum. The SWIR consists mainly of reflected solar radiation, and, therefore, provides sensitivity to variations in the abundance of CO2 throughout the troposphere and into the boundary layer. GOSAT is in a sun-synchronous polar orbit at an altitude of 666 km, with a repeating cycle of 3 d (Kadygrov et al., 2009). The retrieved concentration is obtained by the maximum a posteriori (MAP) method with a priori information from the radiative transfer model (RTM) and the pre-processed measured spectra. The overall retrieval algorithm for the GOSAT CO2 and CH4 product is described in Yoshida et al. (2013). We used the GOSAT XCO2 Level 2 (L2) products from the column abundance retrieved from short wave infra-red (SWIR) radiance spectra (version v02.21) based on the retrieval led by the National Institute for Environmental Studies (NIES). The details of the products, including the retrieval processes and observation results, can be found at http://www.gosat.nies.go.jp/index_e.html. In-situ 2.2 data The in-situ XCO2 data from TCCON (Total Carbon Column Observing Network) (http://tccon.ipac.caltech.edu/) were used to validate the satellite and HASM model results. The TCCON sites use ground-based Fourier transform spectrometers to measure high-resolution spectra (0.02 cm−1) in the near infrared (3,800–15,500 cm), from which XCO2 is retrieved. TCCON XCO2 has been rigorously calibrated against the integrated CO2 profiles measured by WMO-standard instrumentation aboard aircraft. The precision and accuracy of the TCCON XCO2 are both 0.8 ppm (Wunch et al., 2010), and the TCCON XCO2 has been widely used in the accuracy verification and error correction of the satellite
Sci China Earth Sci January (2017) Vol. 60 No. 1
145
Zhao M W, et al.
retrieval CO2 data (Wunch et al., 2011). Eighteen TCCON sites were used in this research (Table 1), of which six sites (Marked Key Site) were selected to perform time series analysis because there is sufficient valid data for 2010–2013 at these sites.
Methodology 3.
second order partial derivative with respect to y, and fxy represents the mixed partial derivative first with respect to x and then with respect to y. The entire derivative can be calculated using finite difference methods (Zhao N et al., 2014). Partial differential equations of the surface theory require E, F, G, L, M, N to satisfy the following Gauss equations (Somasundaram, 2005; Toponogov, 2006; Yue, 2011):
A surface is determined by the first and the second fundamental coefficients based on differential geometry (Henderson, 1998). Suppose a surface can be expressed as z=f(x, y), the first fundamental coefficients E, F, G can be expressed as eq. (1), and the second fundamental coefficients L, M, N can be expressed as eq. (2). E = 1 + f x2 , F = f x . f y , G = 1 + f y2 ,
L= N=
fxx 1+fx + fy 2
2
f yy 1 + f x2 + f y2
,M =
fxy 1 + f x2 + f y2
b± b 2a
2
4ac
(1)
,
which fx represents the first order partial derivative of the function z=f(x, y) with respect to x, fy represents the first order partial derivative with respect to y, fxx represents the second order partial derivative with respect to x, fyy represents the
1 11 x
f yy =
1 22 x
f xy =
1 12 x
1 11
where 1 22
=
1 12
=
, (2)
f xx =
2GFy
f + f +
GGx
2(EG
F 2)
,
1
f +
M E+G
1
2FFx + FEy
2(EG FGy
2(EG F 2) GEy FGx
N E+G
2 12 y
GEx
2
F )
,
,
2 12
=
,
1
f +
2 22 y
f +
=
L E+G
f +
2 11 y
EGx 2(EG
FEy F 2)
2 11
=
2 22
=
,
(3)
, 2EFx
EEy
FEx 2
2(EG F ) EGy 2FFy + FGx 2(EG
F 2)
, ,
. The subscript of E,
F, G represents the first order derivative with respect to x or y. The Christoffel symbols 111, 121 , 221, 112, 122 , 222 only depend on the first fundamental coefficients E, F, G and their derivatives. All the partial derivatives in eq. (3) can be computed based on the central difference; then, the discrete equations of the HASM are as follows:
Table 1 TCCON site details used in this paper Site
Est
Lon
Lat
Alt. (km)
Ascension, Island
May-12
14.333°W
7.917°S
0.01
P.S.
Bialystok, Poland
Mar-09
23.025°E
53.230°N
0.18
Key site
Bremen, Germany
Jul-04
8.850°E
53.100°N
0.027
Key site
Darwin, Austrlia
Aug-05
130.892°E
12.424°S
0.03
Eureka, Canada
Aug-06
86.420°W
80.050°N
0.61
Garmisch, Germany
Jul-07
11.063°E
47.476°N
0.74
Key site
Izana, Tenerife
May-07
16.500°W
28.300°N
2.37
Key site
Caltech, USA
Jul-12
118.127°W
34.136°N
0.23
Karlsruhe, Germany
Sep-09
8.439°E
49.100°N
0.12
Lamont, OK (USA)
Jul-08
97.486°W
36.604°N
0.32
Lauder, New Zealland
Jun-04
169.684°E
45.038°S
0.37
Ny Alesund, Spitsbergen
Apr-02
11.900°E
78.900°N
0.02
Orleans, France
Aug-09
2.113°E
47.970°N
0.13
Park Falls, WI (USA)
May-04
90.273°W
45.945°N
0.44
Reunion Island
Sep-11
55.485°E
20.901°S
0.087
Sodankyla, Finland
Jan-09
26.633°E
67.368°N
0.188
Tsukuba, Japan
Dec-08
140.120°E
36.050°N
0.03
Wollongong, Australia
May-08
150.879°E
34.406°S
0.03
Key site
Key site
146
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
f in++1,1j
2f in, j+ 1 + f i n +1,1j
=
h2
)
f in, j++11
f i, j
)
h2 n n f i, j + 1 i,j
f in++1,1j + 1
+
1
2h
2f in, j+ 1 + f in, j+-11 2 22
( )
n
f i, j + 1
2 n 11 i,j
+(
+(
n
=
f in+ 1, j
1 n 11 i, j
2h L
1 n 22 i, j
1
2h f in++1,1j -1
+
1
f in+ 1, j
n
f i, j
Comparison 4.1 with the classic interpolation methods
n i, j
Ei n, j + Gin, j
( )
Results 4. and discussion
f in 1, j
f in 1, j 2h
N
n i, j
E + Gi n, j n i, j
f in-1,+j1+ 1 + f in +1,1j
,
1
,
(4)
1
2h 2
=
1 n 12 i, j
( )
+
f in+ 1, j
f in 1, j 2h
+(
2 n 12 i,j
)
f i ,nj + 1
f in, j
All months during the studied period are selected to evaluate simulation accuracy. For each month, 90% of the validated data points are selected as the simulation training points and the remaining are used to evaluate the simulation accuracy. Then, the HASM is employed to simulate the global distribution of XCO2 and to extract the simulation values using the test points. Finally, we calculate the MAE and RMSE for each verification point based on its original XCO2 value and the interpolated value following, MAE =
1
2h
1 n
n
oi
si ,
(oi
si )2
n
M i ,nj Ei ,nj + Gin, j
1
RMSE =
.
A min B x n + 1 C
d q p
n
,
(5)
Sx (n + 1) = k , where A, B, C is the left-hand side of eq. (4). By introducing a Lagrange parameter λ, we obtain the linear equation of the modified HASM,
Wx n + 1 = V n,
(6)
where V= W = A A + B B + C C + S S; AT d + BTq + C Tp + 2 S Tk ; x is a vector where each element denotes the simulated value of the corresponding grid point. Thus, the process of simulating the distribution of XCO2 using the modified HASM can be divided into four steps: (1) Building the matrix A, B, C according to the study area and resolution, and building the matrix S and vector V according to the sample information, including the value and position of each sample point; (2) Building the large-scale sparse group of linear equations based on eq. (6), where the default value of λ is 10,000; (3) Solving the large-scale sparse group of linear equations, where the preconditioned conjugate-gradient technique (PCG) is used to solve the linear equations (Golub and van Loan, 2009); (4) Output the solution of the equations using the specific data format. The modified HASM does not require a driving field anymore; the driving field provides the initial value of the study area and is uesd to compute the first and second fundamental coefficients in the former HASM. T
i
n
The constraint equation for the sampling points’ information is added to eq. (4), and the formulation of HASM can be expressed as
T
T
2
T
(7)
i
,
(8)
1
where oi represents the original value of each test point, s i represents the simulation value of the same test point, and n is the total number of test points. The HASM algorithm used in this paper is a type of spatial interpolation method; thus, although studies hava proven that HASM has better simulation accuracy than the classic interpolation methods, such as IDW, and Kriging, in this paper, a comparison between the HASM and other interpolation methods is conducted, to prove that HASM is also superior for simulating the gas concentration distribution. Ordinary Kriging, IDW, and Spline are used as representatives of the classic interpolation methods, to perform the same analysis and to compute the MAE and SMSE. Tables 2 and 3 show the MAE and RMSE of the HASM and classic interpolation methods in 2011, 2012, and 2013. The HASM results are slightly better than the classic interpolation methods. Although the HASM is not far superior to the classic interpolation methods based on the MAE and RMSE, we will explain the advantages of the HASM from other perspectives. Figure 1 shows the correlations between GOSAT-XCO2 and the interpolation results of Kriging and HASM. The formulaat the top left is the fitting function of the interpolation values and the retrieval values. For the three years, all the simulation results of the HASM are superior to the Kriging results. The main advantage of the HASM compared with the classic interpolation methods is that its interpolation results remove the points that have significant differences with the inversion values, which is of great importance in simulating the XCO2 distribution because the classic interpolation methods produce a maximal value and minimal value, and these differences can lead to inaccurate estimations of the carbon sink and source. Therefore, based on the above analyses, HASM is superior in the global distribution simulation of the monthly-mean XCO2. In the following sections,
Sci China Earth Sci January (2017) Vol. 60 No. 1
147
Zhao M W, et al.
Table 2 Accuracy validation of the classic interpolation methods and HASM: MAE (ppmv) Time
Kriging
IDW
Spline
Table 3 Accuracy validation of the classic interpolation methods and HASM: RMSE (ppmv)
HASM
Time
Kriging
IDW
Spline
HASM
1.207
1.202
1.646
1.142
2011-01
1.207
1.259
1.633
1.174
2011-01
2011-02
1.245
1.218
1.727
1.177
2011-02
1.245
1.147
1.701
1.054
1.167
2011-03
1.202
1.145
1.642
1.088
2011-04
1.33
1.27
1.702
1.194
2011-05
1.403
1.503
1.843
1.493
2011-06
1.575
1.615
2.046
1.525
2011-07
1.435
1.341
1.982
1.296
2011-08
1.439
1.496
1.923
1.487
2011-09
1.215
1.439
2.05
1.385
2011-10
1.296
1.274
1.822
1.25
2011-03
1.202
1.207
1.845
2011-04
1.33
1.344
1.879
1.283
2011.05
1.403
1.368
1.725
1.347
2011-06
1.575
1.557
2.154
1.531
2011-07
1.435
1.486
2.081
1.444
2011-08
1.439
1.409
1.926
1.394
2011-09
1.215
1.217
1.9
1.19
2011-10
1.296
1.29
1.869
1.269
2011-11
1.248
1.267
1.674
1.241
2011-12
0.991
0.978
1.355
2012-01
1.124
1.1
2012-02
1.395
2012-03
1.56
2012-04 2012-05
2011-11
1.248
1.35
1.808
1.284
2011-12
0.991
1.172
1.325
1.085
0.995
2012-01
1.124
1.104
1.519
1.102
1.682
1.062
2012-02
1.395
1.448
2.157
1.446
1.406
1.999
1.36
2012-03
1.56
1.995
2.004
1.482
1.502
1.867
1.442
2012-04
1.392
1.233
1.666
1.163
1.392
1.42
1.847
1.37
2012-05
1.234
1.167
1.913
1.145
1.234
1.204
1.833
1.198
2012-06
1.447
1.475
2.048
1.413
2012-06
1.447
1.458
2.051
1.444
2012-07
1.452
1.297
2.032
1.293
2012-07
1.452
1.442
2.211
1.395
2012-08
1.603
1.549
1.884
1.491
1.547
1.419
1.996
1.286
2012-08
1.603
1.582
1.909
1.551
2012-09
2012-09
1.547
1.588
2.014
1.496
2012-10
1.104
1.119
1.898
1.097
2012-11
1.205
1.584
2.1
1.553
2012-12
1.242
1.481
2.386
1.405
2013-01
1.089
1.172
1.289
1.086
2013-02
1.358
1.488
1.748
1.445
2013-03
1.255
1.496
1.92
1.439
2013-04
1.349
1.464
1.487
1.41
2013-07
1.392
1.14
1.838
1.091
1.224
2.069
1.191 1.067
2012-10
1.104
1.182
1.658
1.115
2012-11
1.205
1.217
1.73
1.205
2012-12
1.242
1.186
1.76
1.17
2013-01
1.089
1.154
1.429
1.073
2013-02
1.358
1.346
1.833
1.247
2013-03
1.255
1.254
1.94
1.245
2013-04
1.349
1.331
1.993
1.323
2013-08
1.27
2013-07
1.392
1.41
2.026
1.374
2013-09
1.197
1.09
1.662
2013-08
1.27
1.273
2.059
1.234
2013-10
1.043
0.988
1.303
0.969
2013-09
1.197
1.22
1.891
1.153
2013-11
1.119
1.188
1.662
1.133
2013-10
1.043
1.186
1.375
1.041
2013-12
0.953
0.882
1.392
0.83
2013-11
1.119
1.195
1.367
1.118
2013-12
0.953
0.978
1.445
0.946
the discussions aremainly focused on the HASM simulation results. Accuracy 4.2 verification with in situ data Currently, there are fewer than twenty TCCON sites around the world. Considereing the time range discussed in this paper, and whether there are valid retrieval data points at the TCCON sites, this paper selects six TCCON sites to verify the simulation accuracy of the HASM. The loacations of each
sites are Bialystok (53.23°N, 23.025°E), Bremen (53.10°N, 8.85°E), Garmisch (47.476°N, 11.063°E), Izana (28.3°N, 16.5°W), Lamont (36.604°N, 97.486°W), and Parkfalls (45.945°N, 90.273°W). Figure 2 shows the change trends of XCO2 at the six TCCON sites. The blue solid circle represents the in situ value, the red solid circle represents the retrieval value of GOSAT, and the green solid circle represents the simulation value of HASM. The horizontal axis represents the months, spanning from July 2009 to August 2012. Because the observation data of the TCCON sites are not continuous in time, there are gaps in the in situ data curve.
148
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
A Figure 1 comparision of the two methods’ results and the GOSAT-XCO2 for three years. (a), (b) 2011; (c), (d) 2012; (e), (f) 2013.
Generally, the simulated XCO2 of HASM is consistent with the TCCON site observations. However, the comparison between the simulated XCO2 and the satellite retrieval values shows differences at the six TCCON sites. At some sites, such as Bialystok and Garmisch, the simulated XCO2 of HASM shows better consistency than the satellite retrieval value. At other sites, the HASM removes some data that have significant differences with the in situ data; for example, at Bremen and Izana, there are some satellite retrieval data that have significant differences with the in situ data, while the simulated result of HASM removed these differences. Finally, there are some sites, such as Lamont and Parkfalls, which the satellite retrieval value and the simulated result of HASM show minimal differences. Taking the TCCON observation values as the ture values, the MAE of the satellite retrieval values and the HASM values are calculated for the six TCCON sites. The HASM simulation MAE is 1.688 ppmv, and the satellite retrieval MAE at the same position is
2.147 ppmv, which indicates that the HASM improves the satellite retrieval data. Although the HASM simulation is based on the satellite retrieval data, the simulated results show better consistency with the TCCON observation data, i.e., the HASM not only performs spatial interpolation of the XCO2, but also optimizes the existing satellite retrieval data. We can explain this conclusion using Figure 3, which shows the satellite retrieval value, and the simulated value, with the in situ values of the six TCCON sites. Because there are missing satellite retrieval data, there are fewer scatter points in Figure 3b than in Figure 3a. The simulated values have better consistency than the satellite retrieval value. The superiority of HASM can be explained as follows: the retrieval value of CO2 from the spectrum information is conducted in a local area. In other words, the CO2 concentration information from other locations has no influence on the CO2 concentration value at aspecific location. Although
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
149
Comparison Figure 2 of TCCON-XCO2, GOSAT-XCO2, and HASM-XCO2. (a) Bialystok; (b) Bremen; (c) Garmisch; (d) Izana; (e) Lamont; (f) Parkfalls.
Goodness Figure 3 of fit between TCCON-XCO2 and GOSAT-XCO2, and HASM-XCO2. (a) GOSAT-XCO2; (b) HASM-XCO2.
the published satellite retrieval XCO2 has been subjected to quality control, there may still be some bad data points for unknown reasons. However, the advantage of HASM is that the model can simulate the CO2 distribution regularity over
a large area. When the global CO2 concentration distribution has some regularities, HASM can make full use of these regularities to estimate the values in the data void areas while simultaneously optimizing the existing value. Therefore, the
150
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
simulated XCO2 values are more accurate than the satellite retrieval value, which is also the important reason why we employ HASM to simulate the distribution of XCO2 on the global scale.
China and Japan have higher values than other locations. In addition, the Persian Gulf region, West Asia, Central Africa, Western Europe, North America, and eastern and central parts of South America also have higher values. There is a relatively low value of XCO2 in Southern South America, Southern Africa, Oceania and Northern Siberia. The global distribution of CO2 shows obvious seasonal characteristics. Therefore, this paper divides the year into four seasons, MAM (March, April, May), JJA (June, July, August), SON (September, October, November) and DJF (December, January, February). The average values of the four seasons’ XCO2 are calculated for the four years (Figure 6). For MAM, the XCO2 value varies from 385.467 to 395.057 ppmv, and the average value is 391.072 ppmv. East and Northeast Asia, and Northeast North America have relatively high XCO2 values, and large portions of South America, Africa and Oceania have low values. For JJA, XCO2 varies from 384.466 to 392.599 ppmv, and the average is 388.125 ppmv. In JJA, most of Northern Asia XCO2 values are relatively low, and high XCO2 values are distributed in the central part of Africa, the central and northern parts of South America, and the southwestern region of North America. The XCO2 value in SON varies from 387.052 to 392.804 ppmv, and the average value is 389.504 ppmv. The central part of South America has high XCO2 values in SON. For DJF, the XCO2 value varies from 385.335 to 394.554 ppmv, and the average value is 390.472 ppmv. The global distribution pattern of XCO2 is consistent with MAM. Figure 7 shows the change tendency of the monthly average value of XCO2 during the research period. For each year, the XCO2 concentration gradually increases from January to May, and then begins to decline, reaching the lowest value in August, before beginning to increase. The difference (HASM-XCO2 minus GOSAT-XCO2) between the monthly mean value of GOSAT-XCO2 and the HASM results on a global scale is shown in Figure 8. The difference has a similar
Linear 4.3 correction of the HASM-XCO2 Compared with GOSAT-XCO2, the HASM interpolated XCO2 has a stronger linear correlation with the in situ XCO2. Therefore, it is feasible to correct the HASM-interpolated results using the linear correlation, to improve the accuracy of the XCO2. Considering the sparse distribution of the TCCON sites on the global scale, in the linear correction test, each TCCON site is regarded as the test site while the others are regarded as the training site (Figure 4). For example, in Figure 4a, the linear correlation between the HASM-interpolated XCO2 and the in situ XCO2 excludes Bialystok. Table 4 shows the MAE and RMSE of the HASM interpolated XCO2 and the linear-corrected HASM XCO2. Using the Bialystok site as the example, the absolute error of the HASM-interpolated XCO2 is 1.668 ppmv, and the absolute error declines to 1.561 ppmv when using the linear correction. For the selected TCCON sites, the linear correction improves the accuracy by 0.493 ppmv overall, indicating that the linear correction of the HASM-interpolated XCO2 improves the quality of the XCO2. Spatial-temporal 4.4 analysis of the global mainland XCO2 During the study period, from January 2010 to December 2013, the 46-month average global mainland XCO2 distribution is shown in Figure 5. XCO2 varies from 386.852 ppmv to 392.653 ppmv, and the mean value is 389.806 ppmv. XCO2 has a distribution pattern of high values in the northern hemisphere and low values in the southern hemisphere. Eastern
Table 4 MAE, RMSE of the HASM-XCO2 and the corrected XCO2 at the studied TCCON sites Site
HASM-MAE
Corrected MAE
HASM-RMSE
Corrected RMSE
Bialystok
1.668
1.561
1.594
0.962
Bremen
1.960
1.324
1.494
1.011
Darwin
2.250
1.069
1.015
0.727
Garmish
0.948
1.316
1.009
0.823
Izana
1.574
0.736
0.992
0.693
Lamont
2.719
1.597
0.850
0.839
Lauder
1.615
1.074
1.149
0.927
Nyalesund
1.652
1.263
1.448
1.049
Orleans
1.527
1.008
1.056
0.882
Parkfalls
1.305
1.184
1.246
0.889
Sodankyla
1.582
1.248
1.475
1.084
Mean
1.709
1.216
1.212
0.899
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
151
Goodness Figure 4 of fit between TCCON-XCO2 and HASM-XCO2 at each TCCON site. (a) Bialystok; (b) Bremen; (c) Darwin; (d) Garmish; (e) Izana; (f) Lamont; (g) Lauder; (h) Nyalesund; (i) Orleans; (j) Parkfalls; (k) Sodankyla.
152
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
The Figure 5 average monthly global XCO2 distribution (2010-01–2013-12).
The Figure 6 average seasonal global XCO2 distribution (2010-01–2013-12). (a) MMA, (b) JJA, (c) SON, (d) DJF.
tendency as the monthly mean value of XCO2. The difference in 2011 is used as an example, the monthly mean value of GOSAT-XCO2 is higher than the HASM result, in May, June, July, August, and September, whereas the monthly mean value of GOSAT-XCO2 is lower than the HASM result in the remaining months. In other words, compared with the HASM interpolation results, the limited GOSAT-XCO2 data overesimates the global mean value in some months and underetimates the mean value in some months. Taking 2010 as an example, the average monthly HASM
interpolation is lower than the average GOSAT-XCO2 values between May and September. In the remaining months, the HASM-interpolation average month value is higher than the average GOSAT-XCO2 value because it is summer in the northern hemisphere from May to September, and most of the northern hemisphere has a lower XCO2 values than the southern hemisphere. In addition, the land area of the northern hemisphere is greater than that of the southern hemisphere, which decreases the average value of the golabl mianland. Table 5 shows the variation in the annual and seasonal
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
153
averagevalues of XCO2 in the four years (2010–2013). The change tendency of the four seasonal average values is in accordance with that of the annual average value. For each year, the average value of XCO2 in JJA and SON is lower than the annual average value, whereas the average value in MMA and DJF is higher. During the four years, the annual average value increases 6.878 ppmv, and the increases in the four seasonal values are 6.580, 6.558, 7.002 and 6.988 ppmv. Figure 9 shows the annual average increase rate (%) of the global mainland between 2010 and 2013. The XCO2values of the mainland all increase from 2010 to 2013. The maximum growth rate is 0.741, the minimum growth rate is 0.437, and the average growth rate of the global mainland is 0.592. On the global scale, large parts of Central Asia, West Asia, the Arabian Peninsula, and Africa have higher annual growth rates. For Northern Asia, Northwestern North America, and Northern South America, the average annual growth rate is relatively low.
HASM-XCO2 is 1.216 ppmv, and the RMSE is 0.899 ppmv, demonstrating higher accuracy than other related studies. Compared with the simulation methods, the quality and quantity of the retrieval data have greater influence on the XCO2 surface simulation. Therefore, more advanced greenhouse gas detection satellites and XCO2 retrieval algorithms
Conclusion 5.
Variation Figure 7 tendency of the monthly mean XCO2 during the study period.
In this study, HASM is employed to simulate the XCO2 surface based on GOSAT retrieval data. Sufficient tests show that this method can obtain a more accurate XCO2 surface compared with the classic spatial interpolation methods. As introduced in the previous section, HASM was founded on the fundamental theorem of surfaces (Yue et al., 2007), and can reflect the overall trend of the study parameters in space. Therefore, HASM not only can simulate the XCO2 values at locations where there are no validation data, but also can correct the XCO2 values of existing data points. Furthermore, linear correction is performed on for the HASM-interpolated results because the HASM-XCO2 has a strong linear relation with TCCON-XCO2, and this correction can further improve the accuracy of the results. In this study, the TCCON-XCO2 was taken as the true values. The MAE of the corrected
Difference Figure 8 between the monthly mean value of GOSAT-XCO2 and HASM-XCO2 for each month.
XCO Figure 9 2 annual increase rate (%) of the global mainland from 2010 to 2013.
154
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
Table 5 Variation of the annual and seasonal average XCO2 during the research period Year
Annual
MAM
JJA
SON
DJF
2010
386.755
388.193
385.411
386.395
388.551
2011
388.667
390.447
387.127
387.66
390.632
2012
390.806
392.1
389.266
390.558
393.153
2013
393.633
394.773
391.969
393.397
395.539
are needed. Fortunately, many countries are developing greenhouse gas detection satellites. For example, Orbiting Carbon Observatory-2 was launched in 2014, and TanSat, will be launched in 2016 (Liu et al., 2013). It is essential to validate the XCO2 surfaces to understand the qualities of data sets from different satellites. The ideal objective of our research is to develop a simulation system for XCO2 surfaces on the basis of HASM (simply termed as the HASM system), which takes satellite remote sensing data as its driving field and ground observation data as its optimum control constraints (Zhao M et al., 2014). However, the Total Carbon Column Observing Network has only provided an essential validation resource to assess the accuracy of XCO2 surfaces from the satellites in this paper because ground-based Fourier transform spectrometers are too sparsely scattered over the Earth’s surface to act as optimum control constraints for the simulation system for XCO2 surfaces. Therefore, the TCCON must be improved to provide sufficient observation data to obtain more accurate conclusions. Acknowledgements We thank the members of the GOSAT Project (JAXA, NIES and Ministry of the Environment, Japan) for providing GOSAT Level 2 data products. TCCON data were obtained from the TCCON Data Archive, operated by the California Institute of Technology from the website at http://tccon.ipac.caltech.edu/, and we thank all the Pis’ support and helpful suggestions about the paper. This work was supported by the National High-Tech R&D Program of the Ministry of Science and Technology of China (Grant Nos. 2013AA122003, 2011AA12A104-3), the Research Projects of Chuzhou University (Grant No. 2015QD08), the Key Program of National Natural Science Foundation of China (Grant No. 91325204), the National Fundamental R&D Program of the Ministry of Science and Technology of China (Grant No. 2013FY111600-4), the European Commission’s Seventh Framework Programme “PANDA” (Grant No. FP7-SPACE-2013-1), the Public Industry-Specific Fund for Meteorology (Grant No. GYHY201106045) and the 4th and 5th GOSAT/TANSO Joint Research Project.
References Baker D F, Bösch H, Doney S C, O’Brien D, Schimel D S. 2010. Carbon source/sink information provided by column CO2 measurements from the Orbiting Carbon Observatory. Atmos Chem Phys, 10: 4145–4165 Buchwitz M, de Beek R, Noël S, Burrows J P, Bovensmann H, Schneising O, Khlystova I, Bruns M, Bremer H, Bergamaschi P, Körner S, Heimann M. 2006. Atmospheric carbon gases retrieved from SCIAMACHY by WFM-DOAS: Version 0.5 CO and CH4 and impact of calibration improvements on CO2 retrieval. Atmos Chem Phys, 6: 2727–2751
Shim C, Lee J, Wang Y. 2013. Effect of continental sources and sinks on the seasonal and latitudinal gradient of atmospheric carbon dioxide over East Asia. Atmos Environ, 79: 853–860 Chevallier F, Maksyutov S, Bousquet P, Bréon F M, Saito R, Yoshida Y, Yokota T. 2009. On the accuracy of the CO2 surface fluxes to be estimated from the GOSAT observations. Geophys Res Lett, 36: L19807 Golub G H, van Loan C F. 2009. Matrix Computations. Beijing: Posts & Telecom Press Guo M, Wang X, Li J, Yi K, Zhong G, Tani H. 2012. Assessment of global carbon dioxide concentration using MODIS and GOSAT data. Sensors, 12: 16368–16389 Guo M, Wang X F, Li J, Yi K P, Zhong G S, Wang H M, Tani H. 2013. Spatial distribution of greenhouse gas concentrations in arid and semiarid regions: A case study in East Asia. J Arid Environ, 91: 119–128 Hammerling D M, Michalak A M, O’Dell C, Kawa S R. 2012. Global CO2 distributions over land from the Greenhouse Gases Observing Satellite (GOSAT). Geophys Res Lett, 39: L08804 Henderson D W. 1998. Differential Geometry. London: Prentice-Hall Houweling S, Hartmann W, Aben I, Schrijver H, Skidmore J, Roelofs G J, Breon F M. 2005. Evidence of systematic errors in SCIAMACHYobserved CO2 due to aerosols. Atmos Chem Phys, 5: 3003–3013 Kadygrov N, Maksyutov S, Eguchi N, Aoki T, Nakazawa T, Yokota T, Inoue G. 2009. Role of simulated GOSAT total column CO2 observations in surface CO2 flux uncertainty reduction. J Geophys Res, 114: D21208 Lei L P, Guan X H, Zeng Z C, Zhang B, Ru F, Bu R. 2014. A comparison of atmospheric CO2 concentration GOSAT-based observations and model simulations. Sci China Earth Sci, 57: 1393–1402 Liu Y, Yang D X, Cai Z N. 2013. A retrieval algorithm for TanSat XCO2 observation: Retrieval experiments using GOSAT data. Chin Sci Bull, 58: 1520–1523 Liu Y, Wang X, Guo M, Tani H. 2012. Mapping the FTS SWIR L2 product of XCO2and XCH4 data from the GOSAT by the Kriging method—A case study in East Asia. Int J Remote Sens, 33: 3004–3025 Mao J, Kawa S R. 2004. Sensitivity studies for space-based measurement of atmospheric total column carbon dioxide by reflected sunlight. Appl Opt, 43: 914–927 Rayner P J, O’Brien D M. 2001. The utility of remotely sensed CO2 concentration data in surface source inversions. Geophys Res Lett, 28: 175–178 Somasundaram D. 2005. Differential Geometry. Harrow: Alpha Science International Ltd Toponogov V A. 2006. Differential Geometry of Curves and Surfaces. New York: Birkhaeuser Boston Wunch D, Toon G C, Wennberg P O, Wofsy S C, Stephens B B, Fischer M L, Uchino O, Abshire J B, Bernath P, Biraud S C, Blavier J F L, Boone C, Bowman K P, Browell E V, Campos T, Connor B J, Daube B C, Deutscher N M, Diao M, Elkins J W, Gerbig C, Gottlieb E, Griffith D W T, Hurst D F, Jiménez R, Keppel-Aleks G, Kort E A, Macatangay R, Machida T, Matsueda H, Moore F, Morino I, Park S, Robinson J, Roehl C M, Sawa Y, Sherlock V, Sweeney C, Tanaka T, Zondlo M A. 2010. Calibration of the Total Carbon Column Observing Network using aircraft profile data. Atmos Meas Tech, 3: 1351–1362 Wunch D, Wennberg P O, Toon G C, Connor B J, Fisher B, Osterman G B, Frankenberg C, Mandrake L, O’Dell C, Ahonen P, Biraud S C, Castano R, Cressie N, Crisp D, Deutscher N M, Eldering A, Fisher M L, Griffith D W T, Gunson M, Heikkinen P, Keppel-Aleks G, Kyrö E, Lindenmaier R, Macatangay R, Mendonca J, Messerschmidt J, Miller C E, Morino I, Notholt J, Oyafuso F A, Rettinger M, Robinson J, Roehl C M, Salawitch R J, Sherlock V, Strong K, Sussmann R, Tanaka T, Thompson D R, Uchino O, Warneke T, Wofsy S C. 2011. A method for evaluating bias in global measurements of CO2 total columns from space. Atmos Chem Phys, 11: 12317–12337 Xu Y, Ke C, Wang J, Sun J, Liu Y, Harris W, Kou C. 2013. Satellite-derived estimations of spatial and seasonal variation in tropospheric carbon dioxide mass over China. Ecol Evol, 3: 4310–4325
Sci China Earth Sci January (2017) Vol. 60 No. 1
Zhao M W, et al.
Yoshida Y, Ota Y, Eguchi N, Kikuchi N, Nobuta K, Tran H, Morino I, Yokota T. 2011. Retrieval algorithm for CO2 and CH4 column abundances from short-wavelength infrared spectral observations by the Greenhouse gases observing satellite. Atmos Meas Tech, 4: 717–734 Yoshida Y, Kikuchi N, Morino I, Uchino O, Oshchepkov S, Bril A, Saeki T, Schutgens N, Toon G C, Wunch D, Roehl C M, Wennberg P O, Griffith D W T, Deutscher N M, Warneke T, Notholt J, Robinson J, Sherlock V, Connor B, Rettinger M, Sussmann R, Ahonen P, Heikkinen P, Kyrö E, Mendonca J, Strong K, Hase F, Dohe S, Yokota T. 2013. Improvement of the retrieval algorithm for GOSAT SWIR XCO2 and XCH4 and their validation using TCCON data. Atmos Meas Tech, 6: 1533–1547 Yue T X, Zhao M W, Zhang X Y. 2015. A high-accuracy method for filling voids on remotely sensed XCO2 surfaces and its verification. J Cleaner Prod, 103: 819–827
155
Yue T X. 2011. Surface Modeling: High Accuracy and High Speed Methods. New York: CRC Press Yue T X, Du Z P, Song D J, Gong Y. 2007. A new method of surface modeling and its application to DEM construction. Geomorphology, 91: 161–172 Zeng Z C, Lei L P, Guo L J, Zhang L, Zhang B. 2013. Incorporating temporal variability to improve geostatistical analysis of satellite-observed CO2 in China. Chin Sci Bull, 58: 1948–1954 Zhao N, Yue T X, Zhao M W, Du Z P, Fan Z M, Chen C F. 2014. Sensitivity studies of a high accuracy surface modeling method. Sci China Earth Sci, 57: 2386–2396 Zhao M, Yue T, Zhao N, Sun X, Zhang X. 2014. Combining LPJ-GUESS and HASM to simulate the spatial distribution of forest vegetation carbon stock in China. J Geogr Sci, 24: 249–268