1. Introduction
Satellite-derived measurements of solar backscattered radiation are commonly used to retrieve information about the earth’s atmosphere as well as the surface [e.g., the Solar Backscatter Ultraviolet (SBUV) instrument, the Total Ozone Mapping Spectrometer (TOMS), the Global Ozone Monitoring Experiment (GOME), the Moderate Imaging Spectrometer (MODIS), and the Scanning Imaging Absorption Spectrometer for Atmospheric Cartography (SCIAMACHY)]. The Measurements of Pollution in the Troposphere (MOPITT) remote sensing instrument launched on board the Earth Observing System (EOS) Terra satellite in December 1999 is a gas correlation radiometer designed to retrieve the atmospheric column amount of CO and methane from measurements of reflected solar radiation in the near-IR, and to retrieve the vertically resolved CO mixing ratio from radiance measurements in the thermal IR. In this paper, we concentrate on the retrieval of the total column amount of methane from MOPITT radiance measurements in the 2.2–2.3-μm spectral range.
MOPITT is the first instrument trying to measure tropospheric methane concentrations with a global coverage. At the current stage of the MOPITT processing, there is no operational methane product available. Various issues, such as geophysical noise (Francis et al. 2002) over land and water surfaces, and polarization effects and a low signal-to-noise ratio (as a consequence of the low surface reflectivity) over water surfaces, complicate the methane retrieval significantly. The main problem in the retrieval of the methane column from signals measured over land is that the reflectance of most land surface types varies with wavelength over the spectral width of the instrument’s bandpass filter. In the ideal case of a constant surface reflectance, the impact of the surface term would cancel out in the MOPITT methane retrieval, but as the analysis presented here shows, this is not the case if the reflectance varies with wavelength.
The structure of the paper is as follows. In section 2 we present an overview of the MOPITT instrument and the retrieval of the methane column amount. Section 3 discusses the spectral reflectance characteristics of various land surface types, and these datasets are used in the theoretical studies described in section 4. Section 5 concentrates on measurements of the MOPITT instrument and looks into the range of observed biases. We summarize and conclude our findings in section 6.
2. The MOPITT experiment
The MOPITT instrument is a nadir-looking eight-channel gas correlation radiometer with a spatial resolution of 22 km by 22 km at nadir. The detector arrays are arranged as a line of four adjacent pixels. Cross-track scanning directs the pixel array to 14 rectangular locations on either side of nadir for a swath coverage of about 640 km (in total 29 stares for one track). The scan pattern allows a complete global coverage in about 3 days.
MOPITT has four channels in the thermal spectral range (at 4.7 μm, channels 1, 3, 5, and 7) used to retrieve the vertical distribution of CO with an accuracy of about 10% (Emmons et al. 2004), and four channels in the near-IR (referred to as solar channels herein). Two of the solar channels (at ∼2.3 μm, channels 2 and 6) are designed to detect the total amount of CO, and two (at ∼2.2 μm, channels 4 and 8) are used for retrieving the total column amount of methane. The two methane channels are very similar in their design, but differ slightly in their broadband filters. At the present time, an operational CO data product is available only from the thermal channels, and not yet for the solar channels.
A maximum a posteriori retrieval approach (Rodgers 2000; Deeter et al. 2003) and the fast radiative transfer model of MOPITT (MOPFAS; Edwards et al. 1999) are used to invert the measured radiances into atmospheric trace gas concentrations. MOPFAS has been implemented through a hierarchy of radiation codes: the line-by-line model GENLN2 and the optical depth lookup-table model MOPABS. The latter is used in the theoretical studies described later. The retrieved quantity (e.g., the methane column) is basically a weighted average of the a priori profile (climatological value) and the information derived from the measured radiances with the weighting determined by the instrument measurement sensitivity. For more information about the MOPITT experiment we refer to Drummond and Mand (1996), Deeter et al. (2003), and Emmons et al. (2004), and references therein.
a. The methane channels
In a gas correlation radiometer (Pan et al. 1995), the incoming radiation passes through a cell containing the same gas (target gas) as the atmospheric gas that is being measured (i.e., methane). In the case of the MOPITT solar channels, the filtering characteristics of the cell are modulated by varying the cell length (length modulated cell, LMC). The modulation of the cell length results in a modulation of the opacity of the cell at frequencies corresponding to the spectral lines of the target gas, while the cell opacity at other frequencies remains constant. It should be mentioned that MOPITT is the first airborne instrument that employs the LMC technique and cannot revert to any previous experiences (Tolton and Drummond 1997).
The modulation of the cell length results in a short (S) and a long (L) paths of the radiation through the cell with a higher transmission for the short path. The S and L signals are combined into two synthetic signals: an average (A) [A = (L + S)/2] and a difference (D) (D = S − L) signal. The A signal is dominated by the spectral regions between the target gas absorption lines and carries mainly information about the background radiance. The D signal has a high response near the center of the absorption lines and is more sensitive than the A signal to the atmospheric concentrations of the target gas.
In Fig. 1 we depict the spectral response function for the A and D signals from channel 4, and the spectral A response for channel 8. The spectral response for 8D is similar to the 4D response and is not included in the figure to maintain clarity. The graphs reflect the distribution of the methane absorption lines over the channel bandpass, that is, positions where the D response function is large and the A function drops indicate the location of absorption lines.
b. The retrieval of methane
Methane is the most abundant hydrocarbon in the atmosphere and, as an infrared active gas, contributes to the greenhouse effect. It has a lifetime of ∼8 yr (e.g., Spivakovsky et al. 2000) and as a result is well mixed in the troposphere (e.g., Dlugokencky et al. 1994). The global variability in the atmospheric column amount (normalized to the standard surface pressure) of methane is within about ±10%. This is indicated by various datasets such as in the aircraft data used to derive the MOPITT a priori methane profile. This dataset includes 525 profile measurements and is composed of data from various campaigns such as the Stratospheric Ozone Experiment (STRATOZ), the Tropospheric Ozone Experiment (TROPOZ), the Transport and Atmospheric Chemistry near the Equator—Atlantic (TRACEA) experiment, the Pacific Exploratory Mission in the Western Pacific Region (PEMWEST), etc. The entire dataset extends over the years 1984–98 and covers a latitude range of 70°S–80°N (Deeter et al. 2003). Also simulations with the global Model of Ozone Research in the Troposphere (MOZART; Horowitz et al. 2003) indicate a similar variability.
Since methane is well mixed in the troposphere, column measurements are sufficient to identify surface sources. However, because of the nearly homogeneous mixing, a high precision in the column measurements is needed to detect spatial and temporal variations (Drummond and Mand 1996). Even if a large methane source produces a substantial (∼10%) variation in the methane concentrations near the surface, this may translate into a variation of only 1% or less in the total column above the source region. For this reason, the required precision for measured methane column amounts is high (1%). Our sensitivity studies with the radiative transfer model MOPABS show that a 1% change in the atmospheric methane column results in a change in D/A of 0.15%. This implies that the radiances need to be measured with a precision of 0.15% or better. The high standard is challenging with regard to the retrieval technique as well as the instrumental performance.
The MOPITT methane channels detect solar radiation within the filter bandpass that passes down through the atmosphere, is reflected at the surface, and passes a second time through the atmosphere before it reaches the detector. On both passes through the atmosphere the radiation is subject to absorption by mainly methane and water vapor. There are no other trace gases in the atmosphere with significant absorption in the chosen channel bandpass. Minor impact might also be caused by undetected clouds (a retrieval is feasible for cloudless pixels only) and aerosols.
The retrieval in the solar channels does not make use of the D and A signals separately, because both are directly proportional to the surface reflectance (Pan et al. 1998; Smith 1997). If the surface reflectance is independent of wavelength over the spectral width of the instrument’s bandpass filter, the surface reflectance term as well as any other signal that is spectrally flat over the channel bandpass (e.g., aerosols to the first order) will cancel out by forming the ratio D/A. However, as we will see, the assumption of a spectrally constant surface reflectance is typically not justified over land, and the impact of a spectral surface reflectance term on D/A is nonnegligible in the retrieval.
3. The spectral variability in the reflectance of land surface types
Most surface types do not show a constant surface reflectance over the bandpass of the MOPITT methane channels, but do show a variation in the reflectance with wavelength. As an example, we depict in Fig. 2 measurements provided by the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Spectral Library (information online at http://speclib.jpl.nasa.gov/) showing the spectral reflectance of various surface types. Over the 2.2–2.3-μm range, vegetation types typically feature 1) a lower reflectance compared to soil surface types and 2) a smooth variation in the reflectance with wavelength. Soil types typically feature a higher reflectivity and a more structured wavelength dependence. The spectral variation of the reflectance of water surfaces (not shown in Fig. 2) is very small and not subject of the current studies. As mentioned earlier, MOPITT experiences other serious problems (polarization effects and high instrumental noise contribution) over water surfaces that overshadow the impact of a spectral surface term in the retrieval.
As a result of the spectrally varying reflectance of most surface types, the surface reflectance term does not constitute a spectrally flat term in the radiative transfer equation and cannot be removed easily by taking the ratio of D/A. The result is a bias in D/A dependent on the type of the underlying surface and its conditions (e.g., soil moisture, roughness of the ground). Because the D/A ratio must be known with high precision, the spectrally varying component, although small, is still a significant source of uncertainty.
4. Theoretical studies
We performed simulations with the radiative transfer model MOPABS (Edwards et al. 1999) to ascertain how the A and D signals are influenced by a spectrally varying surface reflectance. For this purpose, we have adapted MOPABS to include a spectral surface reflectance and performed simulations using the spectral reflectance data from the ASTER Spectral Library. In addition, we have calculated three reference cases using a constant value for the surface reflectance (case A, 0.2; case B, 0.1; case C, 0.01). The first two data columns in Table 1 show the calculated biases for various surface types in reference to case A. Results are given for channels 4 and 8. The different biases evident for channels 4 and 8 are explained by the slightly different bandpass filters for these channels and illustrate the sensitivity to the filter position.
We also include results for cases B and C in Table 1. A spectrally flat surface reflectance term cancels out in D/A, and, hence, the biases with reference to case A are close to zero. The biases for case C are slightly higher but are mostly due to numerical errors. The amount of reflected radiation for a surface reflectance of 0.01 is extremely small, and in the course of the radiative transfer calculations some values might be close to the numerical precision. The results for the ASTER Spectral Library surface types show distinct biases and it is apparent that, typically, soil types (Nos. 3–12) have a positive or a small negative bias, while vegetation types (Nos. 13–16) cause a negative bias.
By interpreting Table 1 in combination with Fig. 3, where the relative variation in the reflectance as a function of wavelength is illustrated, two main facts are evident: 1) surface types with the largest spectral variability in the reflectance cause the largest biases, and 2) most soil types show an increase in the reflectivity with wavelength and cause positive biases, while vegetation types exhibit a decrease in the reflectivity with wavelength and cause a negative bias.
In Fig. 3 we include the spectral A radiances to indicate the channel bandpass and the location of the methane absorption lines. It is apparent that the methane lines are not evenly distributed over the range of the methane bandpass, but are more concentrated toward the long-wavelength side. For this reason, the A signal will basically get more weight from radiation at short wavelengths, while the D signal is more weighted toward radiation at long wavelengths. In the case of vegetation, the A signals are more impacted by radiation reflected at short wavelength where also the surface reflectance is higher, while the D signals are more affected by long wavelengths where the surface reflectance is lower. This explains the negative bias (spectral reflectance case < reference case), because in relative values A is increased and D decreased; that is, D/A is smaller compared to simulations with a constant surface reflectance. The relation between D/A and the retrieved methane column is such that a decrease in D/A signifies a 6–7 times stronger increase in the methane amount. A higher amount of methane in the atmosphere causes a stronger absorption of the solar radiation; that is, the D signal is smaller in the case of high methane concentrations, while the A signal being significantly less sensitive to atmospheric methane concentrations than the D signal will roughly remain constant. Hence, D/A will decrease with increasing atmospheric methane concentrations. The negative bias over vegetation implies that the retrieval will overestimate the methane column amount in this case.
Soil types are more complicated to interpret as many of them show a very structured wavelength dependence. For most types, though, the A signals are more impacted by radiation at short wavelengths where the surface reflectance in these cases is lower, while the D signal gets more weight from long wavelengths where the surface reflectance is higher. This causes a positive bias (spectral reflectance case > reference case), and subsequently the retrieved methane column amount will be underestimated.
To investigate the sensitivity of the bias on the measurement conditions, we performed similar simulation studies over a wide range of viewing geometries and atmospheric conditions (e.g., variable water vapor and methane amounts, surface pressure, etc.). Typically, the corresponding changes in the biases are about 0.1% or below, and in the case of MOPITT, might be neglected to first order.
The impact of a spectral variation in the surface reflectance on the measurements of backscattered solar radiation can be minimized by narrowing the width of the blocking filter and/or by selecting a spectral range where surface types show a small variability and where, in the case of MOPITT, the methane absorption lines are more evenly distributed. We performed simulations similar to those mentioned above, but integrated only over a part of the spectral range of the methane bandpass filter. The resulting biases for channel 4 for two different limitations in the spectral range are included in Table 1.
If the integration is limited to the range 2.25–2.28 μm, the biases are reduced to 0.5% and below for all the surface types specified in Table 1. If the spectral range is limited to 2.22–2.25 μm, the biases are less strongly reduced, and some of the soil types (e.g., Nos. 1 and 2) still show a bias in the order of 5%. This different behavior is partly explained by the relative change in the reflectance over the selected spectral range. As can be seen in Fig. 3, most of the surface types included in Table 1 possess a larger variability over the range 2.22–2.25 μm compared to the range 2.25–2.28 μm. In addition to the relative variation in the reflectance, the distribution of the absorption lines over the spectral band plays an important role. The distribution of the spectral lines over 2.22–2.25 μm is similar to the distribution over the entire range, that is, a concentration of lines toward the long wavelength side, while over the 2.25–2.28-μm range the methane absorption lines are more evenly distributed, and rather seem to possess slightly more weight toward the short wavelength side (cf. to Fig. 1).
The biases between the two spectral subranges have opposite signs, which can be explained by the different distributions of the absorption lines in the two spectral ranges (see, e.g., Fig. 1). The surface reflectance actually shows a similar behavior for both subranges and is not the reason for the sign change. The reflectance of most soil types, for example, is increasing with wavelength in both subranges, but the bias changes from positive to negative. Likewise, vegetation types show a decrease in their reflectance over the 2.22–2.25- as well as the 2.25–2.28-μm ranges.
By reducing the width of the bandpass, it has to be considered that the signal strength is reduced and consequently the signal-to-noise ratio might decrease. A change in the bandpass filter might also cause a change in the instrument’s sensitivity to the atmospheric methane concentrations and in the water vapor interference (Smith 1997). For the actual MOPITT filter a 1% change in the atmospheric methane column causes a change in D/A as observed by MOPITT of about 0.15% [slightly increasing (decreasing) with increasing (decreasing) pathlength through the atmosphere]. Sensitivity studies with MOPABS show that the above-used limitations in the spectral range do not impact the instrument sensitivity significantly. Limiting the bandpass to 2.22–2.25 μm somewhat reduces the sensitivity to about 0.11%, while a limitation to 2.25–2.28 μm slightly increases the sensitivity to about 0.16%. The water vapor interference in all cases remains small; D/A changes by less than 0.005% for a 1% change in the atmospheric water vapor column.
5. Evidence in the observations
In this section we describe to what extent the impact of a spectrally varying surface reflectance is noticeable in the observed radiances and the retrieved methane mixing ratio (methane column amount divided by the air mass; this normalization helps to minimize the impact of surface elevation/pressure differences). For the analyses, we combined the MOPITT data with land coverage data from the International Geosphere–Biosphere Program (IGBP; Loveland et al. 2000) and also with surface reflectance data provided by ASTER. The latter instrument is part of the instrumentation of the Terra satellite.
We used MOPITT data for September and December 2000. Data for these months have been reprocessed to store information not saved in the operational retrieval such as the pixel cloud coverage used to extract pixels that are 100% cloud free. It has been shown that the 5% cloud coverage allowed in the retrieval in the thermal channels (Warner et al. 2001) might affect the radiances in the solar channels more than is acceptable. In addition, we have stored the first-guess model radiances in the reprocessing. The first-guess radiances are calculated with the operational radiative forward model MOPFAS by using ancillary data to define the state of the atmosphere, a constant value for the surface reflectance, and a climatological methane column amount from the National Center for Atmospheric Research (NCAR) global chemistry model MOZART. The first-guess radiances are used as a reference for estimating the bias in D/A caused by the actual spectral variation in the surface reflectance.
We are aware that deviations between the actual methane amount and that used in MOPFAS induce uncertainties in the calculation of the reference radiances and consequently in the biases. However, these uncertainties will be independent of the type of surface and, further, are expected to be small compared to the observed biases. True variations in the global methane column amounts are assumed to be within the range of about ±10% as mentioned earlier, which in turn has an effect on D/A of less than ±1.5%.
a. Analysis using the IGBP land cover information
In contrast to the theoretical studies, we have to deal in actual observations with the complication that over the dimension of one MOPITT pixel various surface types might exist, and the spectral reflectance of a combination of surface types might differ essentially from values found in the literature. Factors like the wetness of the ground or the roughness are expected to have an additional impact on the reflectivity.
However, by combining the MOPITT data with land cover information, for example, from IGBP, a clear relation between the type of surface and the bias in D/A is apparent. We used the IGBP land coverage information that provides a global classification into 17 different types at 1 km by 1 km resolution, and collocated these data globally with selected MOPITT pixels for September and December 2000.
In Fig. 4 we show histograms of the relative difference between measured and modeled D/A (channel 4) for various land classes. A minimum of 150 data points per class and month is used. For this plot we have combined all forest types (IGBP separates five different forest types) into one class, the IGBP types savanna and woody savanna into savanna, and the two IGBP classes cropland and cropland/natural vegetation mosaic into cropland/natural vegetation mosaic. We used only 100% cloud-free pixels with a low geophysical noise contribution (Francis et al. 2002) and where the surface pressure is larger than 960 hPa. The latter restriction is needed because of limitations in MOPFAS.
In Fig. 4 we see again a different behavior for pixels that are covered by vegetation compared to pixels that are mostly free of vegetation. The sign of the biases is in agreement with those estimated from theoretical studies, that is, a positive bias for soil/sparsely vegetated surface types, and a negative bias for vegetation types. The magnitude of the biases for soil/sparsely vegetated areas is smaller compared to the findings from theoretical studies, probably a result of the mixture of soil types and/or the existence of some amount of vegetation within a MOPITT pixel.
The width of the distribution is due to at least three factors: uncertainties in the modeled radiances, the above-mentioned mixture of surface types within one MOPITT pixel, and the fact that the IGBP classification does not provide information about the temporal and seasonal variations in the type and conditions of the surface. To some extent we see seasonal dependencies by comparing the biases for September and December, whereas the seasonal differences seem most pronounced for the land class cropland. To suppress the impact of hemispheric differences in season we include in this graph pixels for the Northern Hemisphere only. A further restriction to a certain region would lead to a better insight into the seasonal variability, but the resulting dataset would be too small for reasonable statistical analysis.
As mentioned earlier, the D/A ratio and the retrieved column amounts are inversely proportional; that is, an overestimation of D/A causes the methane amounts to be too low and vice versa. As a result, we expect the methane concentrations retrieved from the MOPITT radiances to be too high over vegetation and too low over soil/sparsely vegetated areas.
b. Analysis using ASTER spectral reflectance data
In addition to the IGBP land coverage data we used surface reflectance data derived from the ASTER instrument as ancillary information to identify the land surface type. The ASTER spectral bands covering the range of the MOPITT methane bandpass are band 6 (2.185–2.225 μm), band 7 (2.235–2.258 μm), and band 8 (2.295–2.365 μm). Our investigations showed that the slope between band 6 and band 7 is an effective indicator to separate vegetation (negative slope) from soil/sparsely vegetated areas (positive slope). The spectral reflectances depicted in Fig. 3 reflect this behavior.
ASTER has a very high spatial resolution (30 m × 30 m), and the collocation with MOPITT pixels involves a huge data volume. For this reason, we concentrate in the preliminary studies we present herein on a selected region only. We collocated 23 cloud-free and low-elevation MOPITT pixels (surface pressure in the range 990–1010 hPa) over Australia with surface reflectance data from ASTER. The ASTER surface reflectance from band 5 to band 9 (extending over a range of about 2.1 to 2.4 μm) has been spline interpolated to the wavenumbers used in MOPABS, and these data have subsequently been used in the retrieval (ASTER corrected MOPABS retrieval). Even though the ASTER shortwave infrared bands experience crosstalk contamination, we expect the effect to be fairly minor for the ASTER bands used herein (Tonooka and Iwasaki 2003).
Figure 5 shows the original (constant surface reflectance employed in the retrieval) as well as the corrected (interpolated ASTER surface reflectance employed in the retrieval) methane mixing ratio as a function of the ASTER band 7 minus ASTER band 6 reflectance. Regions where (band 7 − band 6) < 0 are typically covered by vegetation. The combination with the IGBP land classification shows that most of these pixels, mainly those where the difference in the reflectance of the two bands is large, belong to the class “savanna.” For these pixels, the retrieved methane concentrations are clearly higher than the concentrations retrieved from pixels where (band 7 − band 6) > 0. In the latter case, most pixels belong to the IGBP type “open shrubland.”
Including some amount of spectral information about the surface reflectance in the retrieval clearly improves the derived methane column amount as can be seen from Fig. 5. The original methane concentrations vary by as much as 16% over these 23 pixels, the corrected ones by only 6%. However, the spectral information about the surface reflectance is not sufficient to cancel out the biases completely. While over vegetation the bias is clearly reduced, over most sparsely vegetated areas the ASTER-corrected MOPABS retrieval leads to no improvement. This is very likely due to the fact that most soil/sparsely vegetated surface types have a more structured wavelength dependence than vegetation types, and the spectral resolution of the surface reflectance employed in the radiative transfer calculations does not sufficiently represent the actual variations.
6. Conclusions
We have presented studies about the influence of a spectral surface reflectance on measurements of backscattered solar radiation. Our investigations discuss in particular the remote sensing gas correlation radiometer MOPITT, and the retrieval of the atmospheric column amount of methane from radiance measurements in the spectral range 2.2–2.3 μm. Various issues such as geophysical and instrumental noise or polarization effects seriously effect the retrieval of methane from MOPITT. Over land, the main problem is a surface-type-dependent bias resulting from the use of a spectrally flat surface reflectance term in the retrieval instead of the true spectrally varying reflectance. The observed radiances D and A are affected differently by a spectral surface reflectance, and the surface reflectance term does not cancel out by forming the ratio D/A as it would be in the case of a spectrally flat reflectance.
Results from theoretical studies are in good agreement with observed biases and indicate a positive bias in the radiance ratio D/A (measured D/A > modeled D/A) over areas with no or sparse vegetation, and a negative bias (measured D/A < modeled D/A) over areas with vegetation. In the case of MOPITT, the observed uncertainties in the radiances are in the range of about ±2%–3%. This causes errors in the retrieved methane concentrations of about ±15%–20% with the methane amount being overestimated over vegetated areas and underestimated over soil/sparsely vegetated surfaces. The biases are independent of viewing geometry and atmospheric conditions to the first order. Our studies have shown that the errors in the retrieved methane column amounts are reduced by including some spectral information about the surface reflectance in the retrieval. However, with the currently available information an accuracy/precision of 1% for the retrieved methane column cannot be reached by MOPITT.
The theoretical studies presented in this paper reveal that the impact of a spectral surface reflectance on measurements of backscattered solar radiation might be minimized by narrowing the channel bandpass and/or by shifting the bandpass to a spectral region where surface types show the least variation in the reflectance. In the case of MOPITT, a shift of the bandpass to spectral regions where the spectral absorption lines of methane are more evenly distributed is advantageous too.
Our findings might be of value for other remote sensing instruments that are sensitive to the amount of backscattered solar radiation. The impact of a spectral surface reflectance on the retrieval from solar radiation measurements needs to be carefully investigated, especially for instruments in the planning phase. In the case of an unacceptably large effect, proper actions, for example, a change in the instrument’s bandpass filter, should be taken.
Acknowledgments
Funding for this work is provided by the Erwin-Schrödinger Fellowship of the Austrian Science Foundation and by the NCAR Visiting Scientist Program. The NCAR MOPITT project is supported by the National Aeronautics and Space Administration (NASA) Earth Observing System (EOS) Program. The National Center for Atmospheric Research (NCAR) is sponsored by the National Science Foundation. The authors thank Steven Massie and Alyn Lambert for helpful discussions.
REFERENCES
Deeter, M. N., and Coauthors, 2003: Operational carbon monoxide retrieval algorithm and selected results for the MOPITT instrument. J. Geophys. Res., 108 , D14,. 4399–4410.
Dlugokencky, E. J., Steele L. P. , Lang P. M. , and Masarie K. A. , 1994: The growth rate and distribution of atmospheric methane. J. Geophys. Res., 99 , D8,. 17021–17043.
Drummond, J. R., and Mand G. S. , 1996: The Measurements of Pollution in the Troposphere (MOPITT) instrument: Overall performance and calibration requirements. J. Atmos. Oceanic Technol., 13 , 314–318.
Edwards, D. P., Halvorson C. M. , and Gille J. C. , 1999: Radiative transfer modeling for the EOS Terra satellite Measurement of Pollution in the Troposphere (MOPITT) instrument. J. Geophys. Res., 104 , D14,. 16755–16775.
Emmons, L. K., and Coauthors, 2004: Validation of Measurements of Pollution in the Troposphere (MOPITT) CO retrievals with aircraft in situ profiles. J. Geophys. Res., 109 .D03309, doi:10.1029/2003JD004101.
Francis, G. F., Gille J. C. , Edwards D. P. , and Ziskin D. , 2002: Influence of surface reflectivity variability on MOPITT 2.2–2.3 μm channel radiances and the retrieval of CO and CH4 IEEE Int. Geoscience and Remote Sensing Symp. (IGARSS), Toronto, ON, Canada, IEEE, 1094–1096.
Horowitz, L. W., and Coauthors, 2003: A global simulation of tropospheric ozone and relates tracers: Description and evaluation of MOZART, version2. J. Geophys. Res., 108 .4784, doi:10.1029/2002JD002853.
Loveland, T. R., Reed B. C. , Brown J. F. , Ohlen D. O. , Zhu J. , Yang L. , and Merchant J. W. , 2000: Development of a global land cover characteristics database and IGBP DISCover from 1-km AVHRR data. Int. J. Remote Sens., 21 , 1303–1330.
Pan, L., Edwards D. P. , Gille J. C. , Smith M. W. , and Drummond J. R. , 1995: Satellite remote sensing of tropospheric CO and CH4: Forward model studies of the MOPITT instrument. Appl. Opt., 34 , 30,. 6976–6988.
Pan, L., Gille J. C. , Edwards D. P. , Bailey P. L. , and Rodgers C. D. , 1998: Retrieval of tropospheric carbon monoxide for the MOPITT experiment. J. Geophys. Res., 103 , D24,. 32277–32290.
Rodgers, C. D., 2000: Inverse Methods for Atmospheric Sounding: Theory and Practice. World Scientific, 238 pp.
Smith, M., 1997: Method and results for optimizing the MOPITT methane bandpass. Appl. Opt., 36 , 18,. 4285–4291.
Spivakovsky, C. M., and Coauthors, 2000: Three-dimensional climatological distribution of tropospheric OH: Update and evaluation. J. Geophys. Res., 105 , 8931–8980.
Tolton, B. T., and Drummond J. R. , 1997: Characterization of the length-modulated radiometer. Appl. Opt., 36 , 5409–5420.
Tonooka, H., and Iwasaki A. , 2003: Improvement of ASTER/SWIR crosstalk correction. Proc. SPIE, 5234 , 168–179.
Warner, J. X., Gille J. C. , Edwards D. P. , Ziskin D. C. , Smith M. W. , Bailey P. L. , and Rokke L. , 2001: Cloud detection and clearing for the Earth Observing System Terra satellite Measurements of Pollution in the Troposphere (MOPITT) experiment. Appl. Opt., 40 , 1269–1284.
Spectral A and D response functions for the MOPITTmethane channel 4, and spectral A response function for channel 8.
Citation: Journal of Atmospheric and Oceanic Technology 22, 5; 10.1175/JTECH1721.1
Spectral reflectance for various surface types. Data are from the ASTER Spectral Library. The MOPITT methane bandpass is indicated by the shaded area.
Citation: Journal of Atmospheric and Oceanic Technology 22, 5; 10.1175/JTECH1721.1
Relative change in the reflectance of various surface types over the range 2.2–2.3 μm. The spacing of the data points reflects the spectral resolution of the data. The spectral variation (not to scale) of the A signal (channel 4) is indicated by the gray line.
Citation: Journal of Atmospheric and Oceanic Technology 22, 5; 10.1175/JTECH1721.1
Frequency distribution of the relative difference between measured and first-guess D/A for different IGBP land classes. Data for Sep (black) and Dec (gray).
Citation: Journal of Atmospheric and Oceanic Technology 22, 5; 10.1175/JTECH1721.1
Original (employing a constant surface reflectance in the retrieval) and corrected methane amounts (employing the ASTER surface reflectance data in the retrieval). IGBP surface type savanna (triangles) and open shrubland (squares).
Citation: Journal of Atmospheric and Oceanic Technology 22, 5; 10.1175/JTECH1721.1
Bias (%) in D/A for channels 8 and 4 for different surface types relative to case A (constant reflectivity of 0.2). Biases are calculated as (x reference)/reference. Results are given for the entire wavelength range and, in the case of channel 4, also when the radiance integration is performed over a limited range only (2.22–2.25 μm, 2.25–2.28 μm).
