Three different sets of quality control filters were applied to OCO-2 high-latitude retrievals in this study and are defined in Table  (see the Supplement or for definitions of quality control parameters). Two of these three sets of quality control (QC) filters are recommended by the OCO science team for ACOS B8 retrievals (B8 QC) and ACOS B9 retrievals (B9 QC) and are summarized by the binary variables xco2_quality_flag_b8 and xco2_quality_flag in the OCO-2 B9 Lite files . Methods for selecting quality thresholds and details on the B8 QC filters are discussed by . Improvements in pointing accuracy in ACOS B9 , and a careful reevaluation of quality control parameters, allowed for intentionally more permissive quality thresholds in B9 QC than in B8 QC; this resulted in a substantial increase in data throughput over regions, such as the boreal forest and high latitudes in general, that have been sparsely represented under past OCO-2 ACOS spectral fitting and quality control regimes. The third set of quality control filters (boreal QC) were determined by evaluating quality control histograms like those presented by with “truth” as the NNG observations from the three boreal forest sites (see Appendix Fig. ). Scatterplots of bias in XCO2 (ΔXCO2≡ OCO-2 – NNG) against various retrieval parameters were also considered as a way to search for groupings of bias outliers that could be eliminated with small changes in quality control thresholds. The boreal QC were set with the goal of maximizing data throughput for high-latitude boreal forest sites in spring and autumn while maintaining acceptable ranges of bias at boreal forest sites. Additional retrieval parameters, not used in B8 or B9 QC, were also considered for the boreal QC that relates to challenges in high-latitude observations, including the difference between retrieved and a priori temperature (deltaT), solar zenith angle (sza), XCO2 retrieval uncertainty (xco2_uncertainty), and total column water vapor (tcwv).

Changes to thresholds for albedo in the strong CO2 band (albedo_sco2), the quality of the spectral fit in the weak CO2 band (rms_rel_wco2), and the standard deviation in surface elevation in the satellite field of view (altitude_stddev) were major contributors to the increase in passable high-latitude retrievals with B9 QC relative to B8 QC (compare Figs.  and ). In the boreal QC, ranges of acceptable values are expanded from those in the B9 QC for the ratios of single-band retrievals of CO2 (co2_ratio) and H2O (h2o_ratio) and the quality of the spectral fit in the weak CO2 band (rms_rel_wco2). Albedo in the strong CO2 band (albedo_sco2) is not used as a QC filter in the boreal QC because it seemed that problematic data with low albedo_sco2 could be screened by other QC filters, and there was no evidence that low albedo_sco2 was explicitly correlated to larger OCO-2 biases at boreal forest sites (see quality control plots in Appendix ). In fact, increases in bias and retrieval standard deviation were more often associated with high albedo in the strong CO2 band rather than low values. More conservative thresholds were placed on the slope of albedo in the strong CO2 band given by the continuum fit (albedo_slope_sco2) than were previously used in the B8 QC or the B9 QC due to observed increases in the standard deviation of retrievals and larger negative biases specifically associated with more negative albedo slopes. One possible explanation for this observation is that certain surface types that are more prevalent in the boreal forest are not correctly modeled by the ACOS B9 algorithm, and this could be related to snow-covered surfaces. We expect that introducing a polynomial fit to the albedo in each band, rather than a linear fit, could improve the accuracy of modeled surface albedo in future ACOS versions and potentially result in reduced high-latitude biases. Thresholds for the difference between retrieved surface pressure and a priori surface pressure at the pointing location of the O2A band (dp_o2a) remained the same in boreal QC as in B9 QC, while thresholds for the difference between retrieved surface pressure and a priori surface pressure at the pointing location of the strong CO2 band (dp_sco2) were made marginally more conservative. discuss the pointing errors and other long-term challenges with surface pressure bias in OCO-2 retrievals that lead to the addition of the dp_o2a and dp_sco2 parameters in which there is one retrieved surface pressure and a separate a priori surface pressure defined for each band. The aerosol optical depth (aod) parameters are mostly the same in the boreal QC as in the B9 QC, with the exceptions that total aod (aod_total) and the combined dust, water, and sea salt aod (dws) were removed in the boreal QC because these seemed superfluous after applying other aod filters. While the range of acceptable values for the difference between retrieved and a priori vertical CO2 gradient (co2_grad_del) is nearly the same in the B9 QC as in the boreal QC, the range of values is shifted up. This choice was made based on the distribution for co2_grad_del for the boreal forest sites, and the difference may be attributed to the use of a regional data set for boreal QC rather than a global data set for B9 QC. As previously mentioned, several parameters were used to define quality control filters in the boreal QC that were not included in the parameters for B8 QC and B9 QC. A threshold for sza was introduced in the boreal QC and was chosen to restrict data furthest north to the months of March through November. Potential challenges with data at high sza are discussed in the introduction of this paper, and high sza was found to be correlated with larger negative OCO-2 biases at boreal forest sites. The sza threshold in the boreal QC only screens approximately 0.5 % of retrievals that manage to move through the other boreal QC filters. The thresholds placed on the difference between retrieved and a priori temperature (deltaT) and total column water vapor (tcwv) were chosen because very low atmospheric water vapor or large differences between retrieved and modeled temperatures are likely to correspond with cold weather and snow cover. In particular, before the application of quality control filters, large negative biases in OCO-2 retrievals were found to be associated with low values of tcwv (discussed in more detail in Sect. ). Although the majority of biased retrievals with low tcwv are screened out by other quality filters, this filter helped to remove a small number of outliers that pass the other QC filters (see Fig. ). Finally, the uncertainty in retrieved XCO2 (xco2_uncertainty) was included arbitrarily in the analysis and found to be effective in eliminating a small number of outliers that made up less than 0.05 % of retrievals not screened by other filters.

For each set of QC filters, all retrievals over land north of 50∘ N latitude in OCO-2 Lite files were evaluated to determine how many failed to meet the quality thresholds for each parameter in each month. The results in Figs. , , and show the number of soundings flagged as bad for each individual QC parameter, so larger values suggest that more data are being removed during quality filtering. One caveat is that for a sounding removed in QC filtering there could be multiple parameters for which the retrieval was flagged as bad, or there could be as few as one parameter for which the retrieval is flagged as bad. Figures  and show there is a clear seasonality to triggered quality filters, with the majority of soundings flagged as bad occurring in spring and early summer. This seasonality is slightly diminished with the B9 QC (Fig. ) relative to the more conservative B8 QC (Fig. ) and is only marginally manifested in the boreal QC (Fig. ). The reduction in the number of filtered soundings in spring with boreal QC is largely attributable to less conservative bounds on the spectral fit quality in the weak and strong CO2 bands (rms_rel_wco2, rms_rel_sco2) and the ratios of single-band retrievals of CO2 (co2_ratio) and H2O (h2o_ratio). In all three sets of quality control filters, the parameters which most often cause soundings to be flagged as bad, resulting in the removal of data points, are the spectral fit quality (rms_rel_wco2, rms_rel_sco2), the ratios of single-band retrievals of CO2 (co2_ratio) and H2O (h2o_ratio), and differences between the retrieved and various a priori surface pressures (dp_sco2, dp_o2a, dp). The fact that these parameters account for a greater abundance of retrievals being flagged as bad in spring and autumn suggests that there could be seasonal effects related to these retrieval parameters that need to be accounted for in high-latitude measurements. In particular, there has been speculation that spring snow cover would result in low surface albedo in the 1.61 and 2.06 µm bands, and patchy snow cover or snow-free vegetation protruding from the snow pack could cause variability in albedo within the satellite's field of view . However, after matching Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover data to coincident OCO-2 retrievals at our boreal forest sites, there was no clear connection found between snow cover and increased magnitudes of OCO-2 bias with or without QC filtering. It may still be the case that an incongruous spatial resolution between MODIS and OCO-2 is masking the effects of snow cover on bias or that OCO-2 is only biased by snow in combination with certain other effects of cold weather conditions that are more frequently occurring in spring.

Additional data gained from applying boreal QC rather than B9 QC can be visualized as an increase in spatial coverage of terrestrial high-latitude regions. Figures  and show the difference in the number of soundings passed as “good” by boreal QC relative to B9QC in each 1∘ latitude by 1∘ longitude geographic grid cell. These maps point to substantial increases in spatial coverage of the boreal forest (∼50–70∘ N latitude band) in the spring and autumn months with boreal QC. This improvement in coverage is an important advantage of the boreal QC for selecting OCO-2 retrievals with the goal of evaluating longitudinal trends in seasonal cycles for the boreal forest. The maps in Figs.  and also show some areas where boreal QC provides less OCO-2 data throughput than B9 QC. First, the sza threshold on the boreal QC removes most soundings that manage to pass the other quality filters in winter (November through February), but winter observations are nearly absent from the OCO-2 Lite files before any QC filtering is applied, either because they are removed by prescreening or the data are not collected in the first place. It, therefore, seems reasonable to exclude these data points to the greatest extent possible, and, in general, analyses on satellite-based high-latitude XCO2 continue to be confined to the period between March and October. Second, Figs.  and show that there are some regions where boreal QC yields noticeably less throughput than B9 QC in June, July, and August; however, these months are when there is already an abundance of data and these grid cells with decreased boreal QC throughput are often adjacent to grid cells with an increase in throughput. Overall, the boreal QC provides increased spatial coverage, and this is complemented by an increase in total throughput for soundings over land north of 50∘ N (see Fig. ).

The B9 QC filters succeed in tripling the number of passed retrievals over land at high latitudes, relative to the B8 QC, and the boreal QC allow nearly double the number of retrievals allowed by B9 QC (see Fig.  and the right column of Fig. ). An important result of the boreal QC is the increase in passed retrievals in May, August, and September relative to the B9 QC. While the more relaxed B9 QC allows more high-latitude retrievals than B8 QC, the relative throughput between any two months remains roughly unchanged, and the shapes of the histograms in Fig. a and b are very similar despite scaling. By plotting monthly snow extent in the Northern Hemisphere, as reported by NOAA , alongside monthly average sza and monthly fraction of soundings north of 50∘ N passed by QC filters, Fig.  provides further evidence that some combination of sza and snow cover could be playing a role in high-latitude data removal. If solar zenith angles (sza) were the primary driver of seasonality in high-latitude data throughput, one would expect to obtain approximately the same quantity of passed retrievals in May as in July, but Fig.  indicates that nearly twice as many high-latitude retrievals pass QC filters in July. As mentioned previously in this section, additional analysis did not lead us to the conclusion that snow is the culprit in itself, but some effects from snow or differences between fresh and melting snow cannot be entirely excluded either. It remains unclear how combinations of radiative transfer effects may be contributing to increased data removal at high latitudes in spring.

High-latitude regions experience a higher degree of seasonality in many climate and environment variables than midlatitude regions, and one of our primary motivations in this study is expanding our ability to evaluate CO2 seasonality in the boreal forest. Considering the total average and standard deviation in biases for all coincident soundings, as in Fig. , can obscure seasonal variability in biases that contribute to uncertainty in characterizing seasonal cycles of CO2 obtained from satellite observations. Figure  shows the monthly average biases and monthly standard deviation in biases considered for each site and each set of QC filters. Under all three sets of QC filters we observe seasonal trends in biases, which are more pronounced at East Trout Lake and Sodankylä than at Fairbanks. The observed seasonal variability is characterized by more positive biases in mid to late summer and more negative biases in spring and autumn, which may cause satellite-based estimates of seasonal amplitude and timing to differ from ground-based estimates of these seasonal parameters. For most months the boreal QC does not result in substantial increases in the absolute value of the average monthly biases relative to B9 QC, but in April the boreal QC yields larger negative biases than B9 QC at Fairbanks, and at Sodankylä large negative biases are observed in the boreal QC without any counterpart in the B8 or B9 QC to compare them to. In Fairbanks the average April bias drops from -0.59 ppm, with the B9 QC, to -1.15 ppm, with the boreal QC. Figure a, shows that the difference in the absolute values of the average April bias in Fairbanks is the largest difference in absolute monthly biases when comparing the boreal and B9 QC methods. At Sodankylä April data are only allowed by the boreal QC, while B8 QC and B9 QC filter out all coincident OCO-2 retrievals, but the negative average bias obtained with boreal QC at Sodankylä in April represents the maximum absolute monthly bias of any month, site, or QC method. Figure  also shows larger negative biases with the boreal QC at Fairbanks in August, September, and October in which the boreal QC yields average monthly biases descending from -0.46 to -0.72 ppm in consecutive months, while the B9 QC yields average biases descending from -0.08 to -0.43 ppm. At East Trout Lake and Sodankylä, the monthly biases are only marginally different between the boreal QC and B9 QC, and the boreal QC resulted in slightly smaller monthly biases in March, June, and November at East Trout Lake and in August at Sodankylä (see Figs.  and a).

Monthly bias distributions are visualized with box plots for each site and set of QC filters in Fig.  to further elucidate potential seasonal trends in OCO-2 biases. Figure a, b, and c show that East Trout Lake has the most pronounced seasonal variability in biases and the trends observed are similar for all three sets of QC filters. Figure f suggests there is a slight seasonal trend at Sodankylä with the boreal QC that appears when March and April soundings are included. Overall, the monthly bias distributions also serve to emphasize the similarity in results from the different QC methods.

The seasonal dependence of ΔXCO2 described in the previous section was found to be largely induced by the OCO-2 bias correction and is not apparent in ΔXCO2 calculated with un-bias-corrected OCO-2 retrievals (see Fig. ). In OCO-2 B9 retrievals, the B9 bias correction (B9 bc) for soundings over land is defined by as follows: 4XCO2,corrected=XCO2,raw-foot+0.9(dpfrac)+9.0(dws)+0.029(co2_grad_del-15)0.9954, with a footprint bias-correction term, foot, an overall divisor to agree with TCCON, and parameter-dependent terms adjusting based on a modified parameterization of the retrieved surface pressure bias defined by dpfrac, the sum of dust, water, and sea salt aod's (dws), and the difference between retrieved and a priori vertical gradients in the CO2 profile (co2_grad_del). Of the terms in the B9 bc, dpfrac was the only one found to have seasonal variability at boreal forest sites that was similar to that observed in ΔXCO2 with the OCO-2 bias correction (see Figs.  and ). As will be discussed in Sect. , all versions of the residual in retrieved surface pressure relative to a priori surface pressure (dpfrac, dp, dp_o2a, and dp_sco2) have seasonal variability that can be at least partially attributed to temperature dependence, so we propose new OCO-2 bias corrections with a term for temperature at 700 hPa (T700) to correct for the temperature dependence in dpfrac and dp. To calculate the temperature-dependent modification to the B9 bc, we consider the linear regressions for dpfrac as a function of T700 in each of the satellite-viewing modes for soundings over land north of 50∘ N that pass boreal QC (see Fig. ). Then the regression coefficients for the different viewing modes are combined into average slope and average y intercept with weighting by the fractional abundance of retrievals in that mode to obtain an alternative B9 bias correction (B9 abc), as follows: 5XCO2,corrected=XCO2,raw-foot+0.9(dpfrac-(0.068(T700)-19.03))+9.0(dws)+0.029(co2_grad_del-15)0.99546=XCO2,raw-foot+0.9(dpfrac)-0.0612(T700-279.9)+9.0(dws)+0.029(co2_grad_del-15)0.9954.

An alternative B8 bias correction (B8 abc) was also constructed using linear regression terms for the difference between the retrieved and a priori surface pressure from GEOS5 forward processing for instrument teams (GEOS5-FP-IT; dp) as a function of T700 in Fig.  as follows: 7XCO2,corrected=XCO2,raw-foot+0.36(dp-(0.165(T700)-45.84))+8.5(dws)+0.029(co2_grad_del-15)0.99588=XCO2,raw-foot+0.36(dp)-0.0594(T700-277.8)+8.5(dws)+0.029(co2_grad_del-15)0.9958.

Applying a modification of the B8 bias correction is consistent with the fact that spectroscopy and most aspects of the radiative transfer model remained the same when the ACOS version was updated from B8 to B9. Ideally, a global bias correction would be constructed to include temperature as a component of a broader analysis that considers contributions to the OCO-2 bias in a more holistic context. In this way, the bias correction would be more uniform and widely applicable, while the effects of potential parameter covariance or other influences that we are unable to control for in this regional analysis can be mitigated. That being said, the results that follow suggest that correcting for temperature dependence could be effective in reducing seasonality in OCO-2 bias over the boreal forest.

The second column of Fig. b, f, and j, in addition to results in Figs. –, show that seasonally dependent variability in biases is reduced when the dpfrac term is removed from the B9 bc, but both monthly and overall standard deviations in biases are increased. Without the dpfrac term in the B9 bc, monthly biases in March and April at Sodankylä and in April at Fairbanks are substantially reduced, and month-to-month variability at East Trout Lake is also reduced. Replacing the dpfrac or dp term with a T700 modification, as in B9 abc and B8 abc (Eqs.  and ), results in lower monthly standard deviations in biases than those obtained in the B9 bc with the dpfrac term removed and that are nearly equivalent to those obtained with the standard B9 bc (Eq. ). While some of the seasonal shape is reintroduced with the B9 abc and the B8 abc, biases are still reduced in spring and autumn relative to the B9 bc (see Fig. ). The combined results of Fig. , with the total average biases and total standard deviations in biases shown in Fig. , suggest that the B8 abc is slightly more effective than B9 abc in reducing seasonal variability in bias, reducing total average bias, and reducing total standard deviations in biases. Figure  demonstrates that, for all sites and viewing modes, most of the total average biases with the B8 abc are within ±0.5 ppm. In particular, the B8 abc results in reduced average bias in target mode soundings at all three sites, including nadir soundings at Sodankylä and glint soundings at East Trout Lake and Fairbanks. The B8 abc did result in slight increases in total average biases in nadir soundings at East Trout Lake and in glint soundings at Sodankylä (see Fig. ). However, with the B9 abc, average biases in all modes at Sodankylä, and in nadir and target retrievals at East Trout Lake, are nearly doubled relative to the standard B9 bc.

Overall, the B8 QC is the most conservative set of QC filters, the B9 QC allows for more relaxed thresholds in the QC parameters, and the boreal QC is the most permissive set of QC filters. It was observed in Fig.  that the total standard deviations in biases for all coincident soundings at East Trout Lake and Sodankylä gradually increase from ∼1.3 and ∼1.4 to ∼1.6 ppm ascending, with the increase in throughput obtained from relaxing QC filters (B8 QC < B9 QC < boreal QC). A similar trend at Fairbanks is reflected by an increase from ∼1.1 and ∼1.3 to ∼1.4 ppm. Figure  also demonstrates this increase in standard deviation with different QC filters, but there does not appear to be a seasonal trend in the monthly standard deviation in biases at East Trout Lake and Fairbanks. The anomalously high standard deviation in biases in June at Sodankylä remains to be reconciled, and it represents a potential complication that would perpetuate midsummer uncertainty even if some method of correcting seasonal trends in monthly bias were devised and implemented. Additionally, there is a substantial increase in standard deviation in biases at East Trout Lake in March with boreal QC compared to the B9 QC. While this increase in standard deviation is concerning, the availability of OCO-2 retrievals in the boreal forest in March remains insufficient for a representative sample of northern regions and is not likely to be included in seasonal studies of the boreal forest at this time.

The largest difference in the absolute values of monthly bias between boreal QC and B9 QC is 0.56 ppm in April at Fairbanks (see Fig. a). Boreal QC also results in a 0.18 ppm larger absolute bias than B9 QC in April at East Trout Lake. In July through October, boreal QC results in monthly biases at Fairbanks that are 0.1 to 0.4 ppm larger than with B9 QC, while in May and June there is no change in average monthly biases between the two QC methods. At East Trout Lake and Sodankylä boreal QC produces some monthly biases that are smaller, by up to 0.4 ppm, than B9 QC. Despite some increases in monthly biases with boreal QC relative to B9 QC, it is clear from Fig.  that the modifications in QC filters do not always result in larger monthly biases, and the effects should be weighed against the potential advantages of increasing passable retrievals and spatial coverage. We conclude that the differences between boreal QC and B9 QC are not likely to be a major source of seasonal variability in bias because seasonal dependence is observed with both QC methods in Figs.  and .

The co2_ratio refers to the ratio of XCO2 retrieved by the 2.06 µm band to that retrieved by the 1.61 µm band. Recall that measured low and differing reflectance for snow in the 1.61 and 2.06 µm bands. Systematic departure from unity in the co2_ratio could result from spectroscopic inaccuracies in either band that are characteristic of the instrument or the line list used in the retrieval algorithm. Anomalous departures from unity in the co2_ratio can arise from low signal-to-noise ratio in either or both CO2 bands, which can be due to cloud and aerosol interference or the low reflectivity of snow- and ice-covered surfaces . Patchy snow cover or vegetation protruding through the snow may also cause discrepancies in signal intensity between the weak and strong CO2 bands as a result of variable surface reflectivity in the satellite field of view. In all months at midlatitudes and in May through October at high latitudes, terrestrial retrievals have a systematic departure from unity in the median co2_ratio, with the data approximately normally distributed around 1.012 (see Fig. ). There is an even greater departure from unity in the co2_ratio for high-latitude retrievals in the winter months, namely November through April, with the data approximately normally distributed around 1.020. Figure  demonstrates that there is seasonal variation in the median and distribution of co2_ratio at latitudes north of 50∘ N that is not observed at latitudes from 10 to 50∘ N. This monthly difference in the distribution of retrieved co2_ratio at high latitudes may be a symptom of the effects of snow albedo, or it may be attributable to some other factor, but it warrants some attention because it may be associated with radiative transfer effects that contribute to negative biases in spring at the boreal forest sites.

The parameter tcwv refers to total column water vapor, which is calculated as the product of a scaling factor determined by the full physics retrieval and the a priori tcwv from the European Centre for Medium-Range Weather Forecasts (ECMWF). Atmospheric water vapor is expected to be seasonal, and the seasonality of tcwv at the three boreal forest sites is illustrated in the box plots in Fig. . Large amounts of atmospheric water vapor can suggest that there may be more cloud cover degrading the quality of both satellite-based and ground-based measurements. Even in the absence of clouds, water vapor is a strong infrared absorber in all three bands used by OCO-2, and water vapor is identified in as the most important absorbing gas interfering with line fitting in OCO-2 retrievals. In selecting QC filters for the boreal QC, large negative biases (OCO-2 retrievals reporting lower values of XCO2 than NNG) were correlated to low tcwv, prompting the introduction of quality thresholds for tcwv in the boreal QC (see Fig.  and Table ). Figure  shows the additional retrievals cut by the lower bound on tcwv at 3 kg m-2 (data left of the black dashed line) in the boreal QC which are not cut by other QC filters, and an overall downward trend persists in these removed data. One possible explanation is path shortening resulting from atmospheric scattering, which could result in retrieved spectral radiance that has failed to penetrate atmospheric layers near the surface. This would cause all retrieved gases to be underestimated so that total column water vapor and the total CO2 column are both erroneously low. However, Fig.  shows that while path shortening may explain some instances of negative biases and low tcwv, the relationship persists between a priori tcwv from ECMWF reanalysis and negative XCO2 biases. Because water vapor is a strong infrared absorber, it would be reasonable to expect retrieval errors when tcwv is high, but low atmospheric water vapor is also associated with cold fronts and snow cover. Figure  illustrates the relationship between tcwv and midtropospheric temperature (T700), at 700 hPa, in boreal forest coincident OCO-2 retrievals. There is a distinct maximum for tcwv at a given atmospheric temperature that is defined by the condensation temperature of water, and Fig.  shows that most of the retrievals with tcwv below 3 kg m-2 are also those with midtropospheric temperature (T700) below approximately 250 K. Therefore, it is reasonable to conclude that negative OCO-2 biases are also occurring at low temperatures, which is demonstrated by the correlations between ΔXCO2 and T700 in Fig. .

The dp_o2a and dp_sco2 variables are the residuals of retrieved and a priori surface pressure at the pointing locations of the O2A and strong CO2 bands, respectively. These two retrieval parameters were first included in B9 following the discovery of a pointing error that caused systematic inaccuracies in retrieved surface pressure . Before the release of ACOS B9, only a single dp variable (the difference between retrieved and a priori surface pressure from GEOS5-FP-IT) was used as a quality control and bias-correction parameter. In the analysis by , an additional parameterization of surface pressure residuals (dpfrac) was introduced for use in the OCO-2 B9 bc. The inclusion of dpfrac and dp in the OCO-2 bias correction is not the only reason that surface pressure residuals are important; accurate surface pressure measurements are essential for calculating XCO2, which is defined as the ratio of the total CO2 column to the total column of dry air. These terms are essential components of quality control and bias-correction methods because even small inaccuracies can translate to unacceptable errors in XCO2. While the effects of removing the dp term from the bias correction are considered in Sect. , it is probably inadvisable to remove this term entirely from bias correction or to loosen quality thresholds on dp variables without careful consideration of the impacts on XCO2. Furthermore, in attributing the causes and effects of trends in surface pressure residuals, there may be many competing factors. The seasonal box plots in Fig.  show that there is a seasonality in all four of the variants on surface pressure residuals (dpfrac, dp_o2a, dp_sco2, and dp) at the three boreal forest sites that are similar to the seasonality in bias-corrected ΔXCO2 (compare to Fig. ). Similar seasonality may be a result of multiple seasonal parameters that equally effect dp and XCO2. Figures  and show that both dpfrac and dp also exhibit linear dependence on T700, with greater linearity than the correlations between ΔXCO2 and T700, given either B9 QC or boreal QC in Fig. c and d. Not only is temperature clearly seasonal and correlated to other seasonal parameters, but rates and directions of atmospheric transport are also seasonal, and T700 has been found to link plumes in the free troposphere . show that systematic biases in dp are characterized by a positive trend close to the Equator and a negative trend at higher southern and northern latitudes, and we claim that this could also be a manifestation of temperature dependence.