The changing carbon cycle at Mauna Loa Observatory

Buermann et al. 10.1073/pnas.0611224104.

Supporting Information

Files in this Data Supplement:

SI Text
SI Figure 6
SI Figure 7
SI Table 1
SI Figure 8

SI Figure 6

Fig. 6. Zero and lagged correlations of the MLO amplitude time series with gridded fields of warm-season (May to October) land surface temperature for the two 23-yr study periods, 1959-1981 (a-c) and 1982-2004 (d-f). The lagged amplitude-temperature correlations are computed with the amplitude time series fixed at the two study periods and the time series of the gridded temperature fields shifted back in time, accordingly. Contoured are only correlations that are statistically significant at the 90% level (r ≥ 0.28; Student's t test, one-tailed).

SI Figure 7

Fig. 7. Zero and lagged correlations of the MLO amplitude time series with gridded fields of cold-season (November to April) land surface temperature for the two 23-yr study periods, 1959-1981 (a-c) and 1982-2004 (d-f). The lagged amplitude-temperature correlations are computed with the amplitude time series fixed at the two study periods and the time series of the gridded temperature fields shifted back in time, accordingly. Contoured are only correlations that are statistically significant at the 90% level (r ≥ 0.28; Student's t test, one-tailed).

SI Figure 8

Fig. 8. Standardized anomalies in MLO amplitude (black) and spatial averages of mean warm-season (May to October) NDVI (green) and moving-window correlations between these two time series. (Upper) The NDVI spatial means encompass 30°N to 80°N and 60°E to the eastern coast for Eurasia. Nonvegetated areas were masked out in the spatial averaging. All standardized anomalies are relative to the satellite period 1982-2004. Plotted are both annual values (triangles connected by dashed lines) and a smoothed curve based on a five-point binomial filter (thick solid curves). One tick mark on the y scale corresponds to 1 SD. (Lower) The window length for the moving correlations is 11 yr, and the corresponding correlation is plotted in the middle of each interval (diamonds). Moving correlations between amplitude and NDVI are plotted for zero (green) and 1-yr (blue) and 2-yr (red) lags, with the amplitude always lagging. Nonshaded correlations are statistically significant at the 95% level (r ≥ 0.52; Student's t test, one-tailed).

SI Text

Warm-Season Influence.In their pioneering analysis, Keeling et al. (1) identified a lagged relationship between the Mauna Loa Observatory (MLO) amplitude and annual terrestrial temperatures averaged over the northern hemisphere (NH) from 30°N to 80°N. To shed more light on possible mechanisms underlying these lags, we also performed lagged spatial correlations between the NH warm-season normalized difference vegetation index (NDVI), 6-month standardized precipitation index (SPI6), and Palmer drought-severity index (PDSI), as well as temperatures and MLO amplitude for the two study periods, 1959-1981 and 1982-2004. The results show that only during the earlier period (1959-1981), and in the case of temperature, significant and spatially extensive positive 1-yr-lagged correlations over the mid-to-high latitudes of eastern Eurasia are evident (SI Fig. 6).

Cold-Season Influence. Spatial correlations. Variability in cold-season heterotrophic respiration related to variations in temperature and litter variability over several years may also influence the MLO CO₂ amplitude. Consequently, we computed correlations between NH cold-season temperatures at each gridbox and the MLO CO₂ amplitude at zero and 1- and 2-yr lags for the two study periods, 1959-1981 and 1982-2004. For 1959-1981, variations in the MLO amplitude are positively, albeit weakly, correlated with high-latitude North American cold-season temperatures at zero lag and not significantly at greater lags (SI Fig. 7a). In contrast correlations with Eurasian cold-season temperature for 1959-1981 are positive and significant over a vast zonal band from 40°N to 70°N at 1-yr lag and over portions of central Europe and eastern Asia at 2-yr lag (SI Fig. 7 b and c). During the more recent 1982-2004 period, the 1- and 2-yr-lagged positive correlations over Eurasian mid-to-high latitudes are again significant, although shifts in their spatial extent are evident (SI Fig. 7 e and f): the 1-yr-lagged correlations are more localized over eastern Asia, whereas the 2-yr-lagged correlations encompass almost the entire Eurasian mid-to-high latitudes.

Temporal Evolution. To document the difference in the frequency behavior of the persistent 1- and 2-yr-lagged correlations between eastern-Eurasian cold-season temperatures and MLO amplitude, we computed correlations by using both the full-time series and the residual or strictly interannual time series after subtracting a 5-yr smoothed curve from the original time series (see SI Table 1). With the persistent 1-yr-lagged correlations from 1970-1991, the interannual variations explain only 18% of the correlations of the overall time series, indicating substantial low-frequency contributions (>80% of overall variance explained). In contrast, the more recent persistent 2-yr-lagged correlations from 1979-2000 is dominated by substantial interannual contributions (>80%). Zooming into the later 22-yr period (1979-2000) indicates that particularly in the last decade (1990-2000) the 2-yr-lagged correlations between temperature and MLO amplitude were almost entirely caused by interannual variability (>90% of overall variance explained), suggesting little contribution to the declining trend in the MLO amplitude during this period.

Lagged respiration response. To test the hypothesis of a temperature-related biomass buildup at year 0 and a subsequent respiration response related to refractory litter in the following years that is captured in the MLO amplitude, we constructed a warm-season NDVI time series for eastern Eurasia averaged over the same regions as in the temperature case (30°N-80°N and 60°E to the eastern coast) and computed corresponding moving-window correlations with the MLO amplitude (SI Fig. 8). The results indicate that moderate-to-significant 1- and 2-yr-lagged correlations between MLO amplitude and NDVI do exist during the first half of the satellite period 1982-1993, the time of common upward trends in eastern-Eurasian cold- and warm-season temperatures (see main text), as well as warm-season NDVI. For this period, separating the two time series into their low-frequency and interannual components (five-point binomial filter) indicates that most of the overall variance explained was caused by the common trends; for the period 1984-1993, the fraction of overall variance explained by interannual amplitude and NDVI components is <30% for the 1-yr lag and <10% for the 2-yr lag. During the second half of the satellite period (1994-2004), the moving-window correlations of the amplitude and NDVI time series at 1- and 2-yr lags become insignificant, which could be explained, in part, by more frequent summer drought stress in this period (see main text). In summary, these results suggest that the MLO amplitude registers eastern-Eurasian cold-season decomposition related to the production of refractory litter only when the trends in cold- and warm-season temperatures, as well as the warm-season NDVI, are coherent.

Data Analysis: Extraction of the CO₂ Seasonal Cycle. The 46-yr (1959-2004) time series of the MLO CO₂ record shows a steady increase, related largely to the burning of fossil fuels, and a superimposed seasonal cycle. To extract the CO₂ seasonal cycle from this monthly record, we applied the curve-fitting procedures (the CCGVU software) developed at the National Oceanic and Atmospheric Administration Climate Monitoring and Diagnostics Laboratory (2). In a first step, this algorithm uses a combination of a trend function and a series of annual harmonics to fit the long-term variations and the seasonal component in the monthly CO₂ record. Thereafter, the residuals are spectrally analyzed (fast Fourier transform filtering) and assigned to the trend function or the seasonal component or are discarded (high-frequency noise). Specifically, we used a quadratic polynomial for the trend function, a four-yearly harmonics for the seasonal component, and long/short-term cutoff values of 650 days/80 days in the fast Fourier transform filtering, respectively, to extract the CO₂ seasonal cycle.

1. Keeling CD, Chin JFS, Whorf TP (1996) Nature 382:146-149.

2. Thoning KW, Tans PP, Komhyr WD (1989) J Geophys Res Atmos 94:8549-8565.