As pointed out by Bösch and others, validations of XCO2 retrieved from satellite data are essential because clouds, aerosols, and the surface can introduce potential errors. In this work, we use a validation method similar to that of previous studies (Bösch et al., 2006; Griffith et al., 2010). The collocated FTS observations from the same bands are utilized as the ground truth of XCO2 to validate the space-based XCO2. We have assembled XCO2 retrieved from OCO-2 spectra according to nadir, target, and glint modes and from correlative ground-based FTS at up to eight TCCON sites.
XCO2 values retrieved from OCO-2 and ground-based FTS are not measured exactly at the same position. The first step in the validation procedure was therefore to select the OCO-2 orbits that just pass or point towards the TCCON sites. For the validation of SCIMACHY XCO2, Reuter et al. (2011) selected those pixels within a distance up to 350 km around an FTS site. For GOSAT XCO2, Morino et al. (2011) used GOSAT data within about a ±0.5–1.5° rectangular area centered at an FTS site. For OCO-2 XCO2, Wunch et al. (2017) chose a box centered on an FTS site that spans 5° in latitude and 10° in longitude. In this study, we generally select XCO2 of OCO-2 in nadir and glint modes that pass over TCCON stations within a distance range of ±1° latitude/longitude, which is a tight condition compared with those comparison criteria adopted by GOSAT, SCIMACHY, and OCO-2. Our strict criteria were obtained from range calculation (see Fig. 1). FTS can usually observe the atmosphere when the solar elevation angle is greater than a threshold, for example, angle of 15°. The top of atmosphere is assumed to be 30 km. Therefore, FTS can represent the mean concentration in about ±111 km, i.e., approximately ±1° latitude/longitude. Generally, when OCO-2 passes over this ±1° area, these measurements can be selected to compare with FTS results in our study.
The opportunities of overpasses just over a site in nadir or glint observing mode are very scarce due to the small field-of-view size of OCO-2. Hence, for some scenes, the distances have to be extended to ±2° to find suitable OCO-2 overpasses. However, if a TCCON site is selected as a target by OCO-2, distances of 0.5° are adequate because all of the OCO-2 observations point towards this site and the footprint size will not exceed a 10-km2 area even at large viewing angles (approximately 60°). OCO-2’s instrument collects up to 8 soundings along its ground cross-track swath (approximately 10 km) at 3.0 Hz, yielding 24 soundings per second and up to 390 soundings over each 1° latitude along its orbit track. In all observing modes, clouds in the atmosphere are the main reason for reducing the number of available CO2 measurements over sites. All the OCO-2 measurements located in the selected areas were averaged to result in an XCO2 value. The standard deviations (1σ) of these measurements were taken as the precision in OCO-2 XCO2.
Secondly, the difference regarding observing time should be taken into account for comparison between the ground-based measurements and the space-based measurements. TCCON data within ±30 min centered at the time of OCO-2 overpasses were averaged as the truth value of XCO2. These processes allow product comparisons between OCO-2 and TCCON. Biases in OCO-2 XCO2 can be calculated against the average of TCCON XCO2.
It should be noted that not all TCCON FTS XCO2 data were available and only the data from eight sites are available when this study was carried out. The available data was delayed by about six months. Figure 2 displays the distributions of TCCON sites used to validate OCO-2 XCO2. The detailed information of these TCCON sites is shown in Table 1.
Site Lamont Lauder Park Falls Sodankylä Ascension Island Edwards Caltech/Pasadena Reunion Island Loc. (°) 36.60N 45.04S 45.95N 67.37N 7.93S 34.96N 34.14N 20.90S 97.49W 169.68E 90.27W 26.63E 14.42W 117.88W 118.13W 55.49E Alt. (m) 320 370 442 188 32 700 230 87
Table 1. TCCON sites used in this study
We use the random error as precision, and bias error as accuracy. The uncertainty error in XCO2 with the FTS observation is about 0.8 ppm. For detailed comparison and analysis of TCCON FTS, please see Wunch et al. (2010) and Griffith et al. (2010). Here, the XCO2 from FTS is simply treated as the truth in the comparison. In the comparison, if the OCO-2 measurements have biases greater than 10 ppm as well as standard deviations larger than 15 ppm, they are treated as outliers and excluded from the comparison.
In OCO-2’s nadir observing mode, we obtained about 1－4 overpasses suitable for comparison except at the Ascension site where no collocated data was found in the six months selected. A typical example of OCO-2’s orbit in nadir mode in the region of Lamont can be seen in Fig. 3. The region of ±1° centered at the Lamont site is represented by the rectangle where the OCO-2’s footprint of orbit 02891 passed on 16 January 2015. The length of time for OCO-2 passing the region is about 2 min. During this short period, there are only one or two FTS XCO2 measurements available that are approximately 399 ppm, because the time resolution of FTS is lower than that of OCO-2. As mentioned earlier, OCO-2 has a higher temporal resolution of 0.33 s and more spatial samplings in the region, which can both lead to the XCO2 measurement being greater than those of TCCON XCO2. In this case, there were approximately 20 times of OCO-2 data relative to TCCON data. The average of OCO-2 XCO2 was 398.22 ppm with a standard deviation of 0.85 ppm, showing a good quantitative agreement between OCO-2 and FTS at the Lamont site.
Figure 3. (a) Map of the OCO-2 orbit (dotted line) in nadir mode in the region of Lamont (rectangle: 1° x 1°; asterisk: FTS position). (b) Time series of XCO2 retrieved from OCO-2 nadir measurements and FTS spectra at Lamont on 16 January 2015.
In order to characterize the temporal variability of XCO2, Fig. 3 also shows the FTS XCO2 within ±0.5 h of the OCO-2 overpass time. During this 1-h period, FTS XCO2 varies with a range of about 2 ppm. The average and the standard deviation of FTS XCO2 are about 399.01 and 0.71 ppm, respectively, which demonstrates a minor variability of approximately 0.2% in XCO2 during this period.
Figure 4 demonstrates all the comparisons of XCO2 products between OCO-2 and seven TCCON FTSs. The accuracy and precision in OCO-2 XCO2 are clearly different among these TCCON sites. The biases range from 1.2 to 3.5 ppm except at Sodankylä, Finland. Overall, OCO-2 XCO2 has a systematic negative bias at most of the TCCON sites except the sites of Park Falls and Sodankylä. The systematic biases may be attributed to the retrieval algorithm and need further investigation. Table 2 lists statistical results of the difference between OCO-2 and TCCON. Among all the selected TCCON sites, OCO-2 XCO2 at Lamont shows the best agreement with that of FTS. Errors in XCO2 at the Sodankylä site at latitude 67.37°N are very suspicious due to only one collocated data point.
Figure 4. Comparison of XCO2 retrieved from OCO-2 in nadir mode and TCCON FTS from January to June 2015. The error bars show 1σprecision in OCO-2 XCO2. The short-dashed lines and the long-dashed lines represent 1 and 4 ppm, respectively.
Site Num. of colloc. data Bias (ppm) 1σ (ppm) Edwards 3 –1.87 3.16 Lamont 14 –1.66 1.81 Lauder 6 –0.71 5.31 Pasadena 8 0.70 2.50 Park Falls 12 –1.89 2.77 Reunion Island 3 –2.62 6.20 Sodankylä 3 5.39 6.17
Table 2. Statistics of errors in XCO2 retrieved from OCO-2 in nadir mode at seven TCCON sites
Target mode is designed to observe specific surface targets when the satellite flies over these sites. If one site is selected as the target, this mode will provide a large number of measurements towards this site. Figure 5a illustrates the number of OCO-2 XCO2 measurements generated by using this mode at Reunion Island on 18 May 2015. It can be seen that there are a large number of XCO2 and a few FTS XCO2 measurements.
Figure 5. (a) Number of XCO2 measurements retrieved from OCO-2 in target mode and FTS at the Reunion site on 18 May 2015. (b) Comparison of XCO2 retrieved from OCO-2 and TCCON FTS from January to June 2015. The error bars show 1σ precision in OCO-2 XCO2. The short-dashed lines and the long-dashed lines represent 1 and 4 ppm, respectively.
The number of collocated XCO2 data in this mode over TCCON sites should be obviously greater than that in nadir or glint mode. But at some sites, biases as large as 29 ppm in OCO-2 XCO2 have been found and treated as clear outliers such as days 25 and 137 in 2015 at the Lamont site. At Ascension Island, two overpasses of OCO-2 met the spatial and temporal coincidence criteria. But the one on 24 February 2015 was flagged due to larger biases and standard deviations. These data have been excluded from the comparison with TCCON FTS. Figure 5 presents the final comparisons of XCO2 retrieved from TCCON FTS and OCO-2 in target mode. For this case in Fig. 5a, FTS XCO2 at the Reunion site has minor variation during one hour, while OCO-2 XCO2 shows large variation. This site is a lower-latitude site located on small islands remote from large landmasses, but the topography significantly varies. The surface properties and elevations can introduce errors in the OCO-2 XCO2 retrieved from target mode (Wunch et al., 2017). In Fig. 5b, negative biases in OCO-2 XCO2 over most TCCON sites have been found. Table 3 gives the error statistics in XCO2 from OCO-2 in target mode. The negative biases range from –0.35 to –4.9 ppm.
Site Num. of colloc. data Bias (ppm) 1σ (ppm) Ascension 1 –4.9 10.23 Edwards 5 –1.16 1.39 Lamont 7 –1.42 2.48 Lauder 4 –0.12 3.86 Park Falls 4 –0.02 3.63 Pasadena 6 –1.74 2.78 Reunion Island 4 –0.49 6.11 Sodankylä 2 0.48 3.09
Table 3. Statistics of errors in XCO2 of OCO-2 in target mode at eight TCCON sites
The very low albedo of ocean surfaces degrades precision in space-based measurements. Therefore, glint mode is designed to provide valuable measurements which can provide a much higher signal to noise ratio when OCO-2 flies over oceans. In this mode, the satellite instrument points towards the brightest region, where solar radiation is specularly reflected. Because this mode is continuously operated in one complete orbit, there are also large numbers of observations available over land.
Figure 6 displays an example of the spot measured by OCO-2 in glint mode in the vicinity of the Ascension Island site, as well as a time series plot of XCO2 retrieved from the measurements on 6 April 2015. It can be found from Fig. 6a that the OCO-2’s pixels on the ground just pass the edge of the region around the FTS site (shown by the asterisk). FTS XCO2 varies from 397.5 ppm to near 401 ppm, while OCO-2 XCO2 is below 399 ppm. The range of variation in XCO2 from OCO-2 is similar to that of FTS, which is about 4 ppm. The average of OCO-2 XCO2 variation is 1 ppm (approximately 0.25%), systematically lower than that of FTS XCO2.
Figure 6. (a) Map of the OCO-2’s orbit 04053 (dotted line) in glint mode in the region of Ascension Island on 6 April 2015 (rectangle: 1°×1°; asterisk: FTS position). (b) Time series of XCO2 retrieved from OCO-2 in glint mode and FTS at Ascension Island.
Figure 7 reports a comparison between XCO2 retrieved from OCO-2 and FTS XCO2 at all available TCCON sites from January to June 2015 when glint mode was used by OCO-2. All the compared XCO2 values retrieved from OCO-2 show good qualitative agreement with those of TCCON FTS. Overall, no obvious negative biases are found. Table 4 reports the statistics of errors in XCO2 retrieved from OCO-2 in glint mode. No biases larger than 4 ppm are found and most of the biases range from –2.58 to 1.82 ppm. Variability in precision in OCO-2 XCO2 is about 0.2%–0.9%. The finding that OCO-2 XCO2 in glint mode over all the sites has much better agreement with TCCON XCO2 may imply that glint mode is suited for measuring atmospheric CO2 columns over not only ocean but also land.
Figure 7. Comparison of XCO2 retrieved from OCO-2 in glint mode and FTS from January to June 2015. The error bars show 1σ precision of OCO-2 XCO2. The short-dashed lines and the long-dashed lines represent 1 and 4 ppm, respectively.
Site Num. of colloc. data Bias (ppm) 1σ (ppm) Ascension 4 0.06 1.05 Edwards 3 1.93 2.47 Lamont 14 0.14 1.85 Lauder 3 0.88 5.61 Park Falls 12 0.33 2.71 Pasadena 9 0.70 2.38 Reunion Island 8 0.90 1.13 Sodankylä 4 4.15 1.74
Table 4. Statistics of errors in XCO2 retrieved from OCO-2 in glint mode over eight TCCON sites
Glint mode was originally designed to observe CO2 over oceans. The uniformity of oceanic surfaces results in specular reflection of sun light. Because the use of this mode covers a whole orbit, XCO2 can also be retrieved over land. We think that the better agreement between TCCON and OCO-2 in glint mode is probably attributed to the uniformity of the land surface where FTSs are located. To some extent, this uniformity of land surface can easily generate the reflections that deviate from the diffusion of Lambertian surfaces.
The FTS operates automatically every day, including tracking the sun, to record data. Figure 8 shows the photo of the FTS. The time series of FTS XCO2 in Beijing from April to August 2016 is shown in Fig. 9. The seasonal variation of FTS XCO2 can be clearly seen from these measurements. XCO2 decreased slowly from 405 ppm in April to about 395 ppm in August. The daily variation of FTS XCO2 is mostly below 1 ppm.
In total, we have found three overpasses over this site during 2016. Only nadir and glint modes were available to compare. These data are listed in Table 5. Large biases greater than 3 ppm can be found between OCO-2 XCO2 and FTS XCO2 on 6 May and 26 August. OCO-2 XCO2 agrees well with FTS XCO2 on 16 June. The large biases can be attributed to the large distance between FTS site and OCO-2 footprints. Due to the complexity of the urban environments in Beijing, including CO2 emissions and transport in the atmosphere, the CO2 in the Beijing area may have remarkable variations with location. Figure 10 shows the distance between FTS site and OCO-2 footprints in the comparison. On 26 August, the distance between them is much greater than that on 16 June. FTS measurements cannot reflect the real CO2 concentration over the OCO-2 footprints on 26 August. In the future, when more OCO-2 passes over this area, we plan to decease the distance between FTS and satellite footprints. In this region, more collocated data are needed to validate OCO-2 XCO2.
Date (yy-mm-dd) Mode Num. of samples Mean of OCO-2 (ppm) Mean of FTS (ppm) Bias (ppm) 1σ of OCO-2 (ppm) 2016-05-06 Nadir 4 405.05 401.60 3.45 2.39 2016-06-16 Glint 3 396.84 396.82 0.02 2.44 2016-08-26 Glint 3 397.70 394.13 3.57 1.60
Table 5. Comparison of XCO2 between OCO-2 and FTS in Beijing
|Site||Lamont||Lauder||Park Falls||Sodankylä||Ascension Island||Edwards||Caltech/Pasadena||Reunion Island|