Trend and climate signals in seasonal air concentration of organochlorine pesticides over the Great Lakes



[1] Following worldwide bans or restrictions, the atmospheric level of many organochlorine pesticides (OCPs) over the Great Lakes exhibited a decreasing trend since the 1980s in various environmental compartments. Atmospheric conditions also influence variation and trend of OCPs. In the present study a nonparametric Mann-Kendall test with an additional process to remove the effect of temporal (serial) correlation was used to detect the temporal trend of OCPs in the atmosphere over the Great Lakes region and to examine the statistical significance of the trends. Using extended time series of measured air concentrations over the Great Lakes region from the Integrated Atmospheric Deposition Network, this study also revisits relationships between seasonal mean air concentration of OCPs and major climate variabilities in the Northern Hemisphere. To effectively extract climate signals from the temporal trend of air concentrations, we detrended air concentrations through removing their linear trend, which is driven largely by their respective half-lives in the atmosphere. The interannual variations of the extended time series show a good association with interannual climate variability, notably, the North Atlantic Oscillation (NAO) and the El Niño–Southern Oscillation. This study demonstrates that the stronger climate signals can be extracted from the detrended time series of air concentrations of some legacy OCPs. The detrended concentration time series also help to interpret, in addition to the connection with interannual variation of the NAO, the links between atmospheric concentrations of OCPs and decadal or interdecadal climate change.

1. Introduction

[2] The effectiveness evaluation of the Stockholm Convention on Persistent Organic Pollutants (POPs) under Article 16 (available at refers to the need for monitoring data for the evaluation of spatial and temporal trends and regional and global environmental transport of POPs. Understanding the long-term variability of organochlorine pesticides (OCPs) is important to predict their fate and cycling in multicompartmental environments, their long-range atmospheric transport, and their source-receptor relationships. A declining trend of many legacy OCPs over the last two decades, notably, in the Great Lakes and Arctic regions, has been monitored from field measurements [Cortes et al., 1998; Hung et al., 2002, 2005; Sun et al., 2006a, 2006b]. The declining trend of legacy OCPs has been partly attributed to decreasing emissions following their worldwide ban or restriction [Li and Bidleman, 2003]. Other factors, such as compounds' reemission or volatilization associated with their physical and chemical characteristics, location of sources, atmospheric activities, and the residence times (or half-lives) in the atmosphere and soils, also contribute to their long-term temporal trends [Hung et al., 2002]. Therefore, to understand the long-term changes in these toxic chemicals, these factors need to be taken into consideration.

[3] Using a multivariate regression model, Ma and Li [2006] have shown that the annual global reemission, the atmospheric half-lives, and the spring sea surface temperature anomaly in the tropical mideast Pacific Ocean explain 78% of the spring mean air concentration variance of α-hexachlorocyclohexane (HCH) in the 1990s over the Great Lakes. This suggests that the strong interannual variability of atmospheric concentrations of this toxic compound was associated not only with its global reemission and residence time but also with climate variation. Using the measured air concentration of POPs around the Great Lakes region and Canadian high Arctic over the 1990s, Ma et al. [2003a, 2004a, 2004b] found that the year-to-year fluctuation of many POPs in the atmosphere was also correlated with Northern Hemispheric climate variability, notably, the North Atlantic Oscillation (NAO) and the El Niño–Southern Oscillation (ENSO) during winter and spring seasons. Association of interannual variation of toxic chemicals in the atmosphere with the NAO and the ENSO suggests that changes in air temperature and large-scale wind systems associated with these two strong internal phenomena in the global climate system have profound impacts on the environmental fate of POPs.

[4] These results anticipate that the residence time and climate variability will likely be two dominant factors affecting the temporal trend for these legacy OCPs for many years to come. The reported declining trends reflect, in fact, the decadal or interdecadal variation of OCPs, which largely responded to their atmospheric half-lives [Ma and Li, 2006]. For a sufficiently long time series of measured air concentrations of OCPs that exhibits a strong declining trend, the interannual variation might become a “noise,” which can be overwhelmed by the processes causing decadal and interdecadal decreasing trends. In a recent study to reassess the linkage between interannual climate variability and atmospheric concentration of α-HCH and γ-HCH over the Great Lakes, it was found that the correlation of the two HCHs’ time series with climate variabilities was less significant compared with previous findings [Ma et al., 2003a, 2004a, 2004b]. While concerns are raised for the effect of climate variability on interannual changes in OCPs, a question is, Can we more effectively extract interannual climate signals from their decadal or interdecadal trend? Further, the previous studies [Ma et al., 2003a, 2004a, 2004b] identified only the link in the interannual changes between OCPs and climate by using relatively short time series (about 10 years) of OCP data. While extended time series of observed OCPs (16 years) over the Great Lakes are now available, these data may provide useful information to assess the potential connection between the trend of OCPs over the Great Lakes and decadal climate change.

[5] To address these questions, we first establish a statistically significant temporal trend of seasonal air concentration of OCPs in the Great Lakes region. We then reconstruct annual residual time series of the mean spring and summer air concentrations by removing their respective linear trends and the effect of serial correlation (a lag 1 autoregressive process). A nonparametric Mann-Kendall (MK) statistical test is undertaken to examine the statistical significance and robustness of reconstructed trends and time series of OCPs. These detrended time series are compared with interannual and interdecadal climate variabilities to assess the impact of climate change on the fate of OCPs in the Great Lakes region.

2. Materials and Methods

2.1. Data

[6] The Integrated Atmospheric Deposition Network (IADN) was founded in 1990 to study the atmospheric transport and loadings of POPs to the Great Lakes. Atmospheric concentration data, including α-HCH, lindane (γ-HCH), hexachlorobenzene (HCB), and α-endosulfan, are selected in this study. Among these four chemicals, α-HCH and HCB were banned some 30 years ago in North America, γ-HCH has been used in Canada until 2004, and endosulfan is a currently used pesticide. Because of their higher volatility as defined by their smaller value of logKOA, HCHs and HCB have been demonstrated to respond more strongly to climate variability when compared with other OCPs [Ma et al., 2003a; Becker et al., 2008]. They are therefore used as benchmark substances to study the association between OCPs and climate variability. Three IADN sampling sites, which have the longest time series of measured air concentrations from 1992 to 2007, were selected in the present study. These three sites are Eagle Harbor on Lake Superior (EGH, 47°27′47″N, 88°08′59″W), Sleeping Bear Dunes on Lake Michigan (SBD, 44°45′40″N, 86°03′31″W), and Sturgeon Point on Lake Erie (STP, 42°41′35″N, 79°03′18″W). The sites are considered to be representative of the upper lakes (Lakes Superior and Michigan) and one of the lower lakes (Lake Erie). These three sites are operated by the U.S. Environmental Protection Agency (EPA). The sampling details and analytical methods are given in previous publications [Hoff et al., 1996; Cortes et al., 1998; Sun et al., 2006a, 2006b]. The vapor phase concentrations of α-HCH and γ-HCH, HCB, and α-endosulfan have been collected every 12 days at these stations. This investigation focused on spring (March–May) and summer (June–August) seasons. During winter and spring seasons, interannual climate signals are most prominent, and hence, the response of OCPs to climate variation is stronger than during summer and autumn seasons [Ma et al., 2003a]. However, the lower-atmospheric level of OCPs in winters makes little contribution to the annual change in OCPs, whereas, because of strong long-range atmospheric transport and revolatilization from previously contaminated soils during springs, the response of OCPs to climate variation in the spring season may exert influence on the environmental fate of OCPs during the summertime [Ma and Li, 2006]. In the summertime, long-range atmospheric transport seldom occurs; local atmospheric circulations contribute to the atmospheric transport, environmental exchange, and fate of OCPs. Both direct (e.g., for endosulfan) and secondary emissions of legacy and currently used OCPs become dominant. For these reasons, the air concentration data are averaged over spring and summer seasons, respectively.

[7] Climate indices, the modes of variation in the climate system, were collected from the Earth System Research Laboratory of the U.S. National Oceanic and Atmospheric Administration (NOAA) (available at These include the NAO index and the multivariate ENSO index (MEI). The NAO index, defined as the difference of the sea level pressure between Lisbon, Portugal, and Stykkisholmur, Iceland, is derived from the first rotated principal component based on monthly 700 hPa geopotential height anomalies [Hurrell, 1995]. The MEI is defined by the first unrotated principal components based on six main observed variables over the tropical Pacific. These six variables are as follows: sea level pressure, zonal and meridional components of the surface wind, sea surface temperature, surface air temperature, and total cloudiness fraction of the sky [Diaz et al., 2001]. Daily and mean surface air temperature (SAT) data averaged over the Great Lakes were obtained from the U.S. National Center for Environmental Prediction reanalysis [Kalnay et al., 1996].

2.2. Trend Analysis

[8] Extensive OCP trend analyses over the Great Lakes and other regions have been carried out previously [Cortes et al., 1998; Sun et al., 2006a, 2006b; Hung et al., 2005]. Because the time series of the IADN-measured OCPs are still relatively short as compared with climate records, their trend may exhibit large variability and a lack of robustness. It is necessary to reexamine these trends by using a more robust nonparametric technique, e.g., the MK statistical test [Mann, 1945; Kendall, 1975]. The MK test requires that the observations that are to be analyzed are the realizations of a collection of independent random variables. Under the null hypothesis, the standardized MK statistic Z follows the standard normal distribution with mean of zero and variance of one. If ∣Z∣ > Z1−α/2, a trend is statistically significant at a level of 1 − α/2 (for further details, see Cao [2008]). If the test is applied to serially correlated data, the trend detection results may not be reliable because the test may then reject the null hypothesis of no trend (H0) more often than specified by the significance level [von Storch, 1995]. Buehler et al. [2004] have found and discussed the autocorrelation in the IADN OCP data. Matalas and Langbein [1962] have demonstrated that positive serial correlation within a time series would reduce the effective sample size of the series compared with independent data and increase probability. To reduce the effect of serial correlation on trend analysis, von Storch [1995] suggested removing the AR(1) (a lag 1 autoregressive process) from the time series through a prewhitening procedure, namely,

equation image

where X is a time series and r1 is the lag 1 serial correlation coefficient of sample data, estimated from sample data by an autocorrelation function as defined by Salas et al. [1980, p. 38]. Yue and Wang [2002] noticed that this prewhitening also removes a portion of the trend. For trend detection, we therefore use a trend-free prewhitening approach [e.g., Yue and Wang, 2002; Cao, 2008; Cao and Ma, 2009] prior to applying the MK test, so that the true trend is preserved and it is no longer influenced by the effect of autocorrelation (serial correlation). In the AR(1) removal procedure, we first detrend air concentration anomaly time series through the removal of the linear trend of normalized (by standard deviation) concentration anomalies. We then remove the first autocorrelation of the detrended (or residual) time series by using equation (1). This procedure enables us more effectively to extract climate signals from the times series of OCPs, as will be presented in section 3.

3. Temporal Trend

[9] Figure 1 illustrates the seasonal mean air concentrations of HCB at the three IADN sites from 1992 to 2007 in springs (Figures 1a1c) and summers (Figures 1d1f) before and after removing AR(1). Except for the elevated atmospheric concentration of HCB in 1998, likely due to a strong El Niño event occurring from the autumn of 1997 to spring 1998 [Ma et al., 2003a, 2004a], the compound exhibits a decreasing trend over the 16 year period. After removing the AR(1), the linear downward trends of HCB are still clearly visible. The declining trend of HCHs over the same periods at the three IADN sites can be also demonstrated by Table 1, which shows Z values and the statistical significance of the reconstructed OCP time series. Most Z values are negative in either spring or summer, indicating a downward trend [Yue and Wang, 2002], except for spring α-endosulfan at EGH and STP. The absolute values of the computed Z statistics (|Z|) are greater than Z1−α/2 in most cases, indicating that the downward trend of the mean air concentrations of OCPs at the three IADN sites is statistically significant at the 95% level. The magnitude of negative Z values indicates the rate of decrease in air concentrations (or downward trend). The atmospheric concentration of OCPs at SBD and STP with larger negative Z values exhibited a stronger downward trend as compared with that at EGH, the remote far north site of the IADN. The atmospheric half-life of the gas phase OCPs at EGH is longer than that at the other two sites [Sun et al., 2006a]. For instance, the half-life of HCB at EGH is 18 years, whereas the half-life of this chemical at SBD and STP is 12 and 15 years, respectively. Further, compared with HCB, larger negative Z values of HCHs also suggest that the mean air concentration of HCHs underwent a stronger decreasing trend during the spring and summertime (except for summer γ-HCH at STP), though a significant increase in air concentration of HCHs in the spring of 1998 is also clearly visible (not shown), which has again been attributed to the strong El Niño event in that year [Ma et al., 2004a]. These decreasing trends are significant at the 95% level, as shown in Table 1. The spring and summer averaged air concentrations of HCHs at the three IADN sites are illustrated in Figure S1. Similar to HCB, both α-HCH and γ-HCH show a declining trend from 1992 to 2007.

Figure 1.

Time series of HCB for 1992–2007. (a–c) Mean spring air concentrations at three IADN sites before and after removing AR(1). (e–f) Mean summer air concentrations before and after removing AR(1).

Table 1. Z Statistics in Mann-Kendall Test for Trend Detection of Mean Spring Air Concentration of α-HCH, γ-HCH, HCB, and α-endosulfana
  • a

    Statistical significance (rejection of H0) is indicated at Z1−α/2 = 1.96 for α = 0.05.


[10] Data Sets S1 and S2 list the means and standard deviations of the spring and summer air concentrations, respectively, of the three compounds for 1992–2007. Compared with HCB, whose spatial distribution was fairly uniform [Shen et al., 2005], the standard deviations of α-HCH and γ-HCH are similar to or only slightly lower than their means, and in one case (α-HCH at EGH), the standard deviation is even greater than the mean. This suggests strong interannual variability in elevated HCHs. We further examined the daily air concentrations of HCHs measured every 12 days by the IADN program during the same period. For α-HCH at SBD (451 samples from 1992 to 2007), for instance, the mean daily air concentration was 60 pg m−3, and the standard deviation was 72.3 pg m−3. The daily air concentrations of HCHs at the other IADN sites exhibit similarly high variability. This example indicates that despite being statistically significant, the trends of HCHs are of high variability. The high variability may be the result of other processes at work, which could result in less robust trends of HCHs. These processes may include changes in air temperature, atmospheric transport, and emission from sources. Given large uncertainties and insufficient knowledge on direct and secondary emission sources of HCHs, it is not straightforward to fingerprint and quantify the individual contribution of these processes to the high variability of HCHs. It was found that the spring SAT averaged over the Great Lakes region from 1992 to 2007 has a mean value at 4.5°C and standard deviation at 1.2°C, showing less interannual variability as compared with the spring averaged HCHs' time series.

[11] No statistically significant trends were detected for α-endosulfan from the spring data (Table 1). This agrees with Sun et al.'s [2006a] annual trend results. However, a significant linear trend of this currently used pesticide is observed from the summer data at SBD and STP (Table 1) but not at EGH, where the standard deviation of the atmospheric concentration of α-endosulfan is not substantially lower than its mean (Figure S1) and the ∣Z∣ value is smaller than Z1−α/2 (Table 1). Figure 2 illustrates these trends in mean summer air concentration of α-endosulfan at SBD and STP before and after removing the AR(1) process, showing significant decline in air concentrations.

Figure 2.

Time series of α-endosulfan for summers of 1995–2007 at (a) SBD and (b) STP before and after removing AR(1).

[12] Higher air concentrations of α-endosulfan have been observed for the summer season as compared with the springtime and other seasons [Sun et al., 2006a], especially at the remote site EGH. This has been attributed to the timing of the agricultural use of endosulfan in the United States and to its short half-life [Sun et al., 2006a]. Because the seasonal and interannual changes in a currently used pesticide are driven primarily by its usage and subsequent emission, the statistically significant trends of α-endosulfan during the summertime at SBD and STP sites, as shown in Table 1 and Figure S1, are somewhat surprising. These trends appear to suggest a decreasing application and changes in regulation for endosulfan in the United States since the mid-1990s, based on measurements taken nearest the time of application (i.e., summer). Endosulfan in the United States has been used extensively in cotton (14.2%), cantaloupe (13.2%), tomatoes (12.2%), and potatoes (8.15%) [EPA, 2002]. The crop fields that may exert influence on contamination of the Great Lakes are primarily in the southeast United States. The data from the same source suggest declining use of endosulfan in U.S. agriculture. Nevertheless, it is important to note that, though the statistically significant trend of α-endosulfan was only found in summer seasons, this trend likely reflects a status of the application and environmental fate of this currently used pesticide from the 1990s to the 2000s. The lack of a statistically significant trend of summer α-endosulfan at EGH is likely due to its remote location, far away from the U.S. agricultural source regions.

4. Climate Signals in the Trend of Air Concentration

[13] Sun et al. [2006a] have derived half-lives of OCPs measured at the IADN sites from the partial pressures of OCPs by using sampled air concentration data and air temperatures from 1992 to 2005. The declining trend of HCHs and HCB in air was driven largely by their respective residence times. This dominant process may overwhelm climate signals, causing variability in the temporal trend of chemicals. Indeed, as shown in Table 1, α-HCH exhibits the lowest Z values (linear declining trend) for the period of 1992–2007, which can be explained by its removal from the atmosphere due to the half-life of the chemical. The half-life suggests that there are likely reservoirs within and outside the Great Lakes region from which the slow leakage and subsequent transport from these sources control the temporal trend in the atmospheric concentration. Reemission from these sources associated with increasing air temperature under climate variations may also contribute to half-lives.

[14] Table 2 provides correlation coefficients for the mean spring air concentration of HCHs and HCB with the NAO index and MEI, before and after removing AR(1) at the three IADN sites. As expected, a weak correlation is observed between γ-HCH and the NAO (R = 0.48 − 0.50) because lindane (γ-HCH) was still used in Canada until 2004 and the fresh emissions of the substance would dominate its interannual trend [Ma et al., 2004a, 2004b]. The correlation between α-HCH and the NAO is weaker still (R = 0.4 − 0.56), likely because of the decreasing trend driven by its removal process (as indicated above). The relationships between atmospheric concentrations and the NAO and MEI are less significant compared to those reported previously by Ma et al. [2004a, 2004b] but are improved by removing the AR(1), especially in the case of HCB with the MEI (Table 2). It can be also seen from Table 2 that removing AR(1) not only improves the correlation between the selected OCPs and the climate indices but also appreciably enhances the statistical significance of the correlations. Compared with HCB (with half-lives ranging from 12 to 18 years), the less persistent HCHs with half-lives ranging from 3.8 to 5 years at the three IADN sites [Sun et al., 2006a] may be removed from the air more quickly, as verified by the steeper rates of decline and the larger negative Z values (Table 1) for atmospheric HCHs as compared with that for HCB.

Table 2. Correlation Coefficients Between Mean Spring Air Concentrations of HCHs and NAO Index, and Mean Spring HCB and MEI, Before and After Removing the AR(1)a
  • a

    The correlations are statistically significant at the 95% level.

α-HCH versus NAO0.560.0250.400.1260.430.096
Removing AR(1)0.500.0250.560.0480.520.040
γ-HCH versus NAO0.480.0570.500.0500.440.086
Removing AR(1)0.510.0490.510.0430.480.061
HCB versus MEI0.590.0160.660.0050.480.062
Removing AR(1)0.620.0100.830.000050.630.011

[15] Detrending a time series enables greater focus to be placed on the fluctuations in the data about the trend. The AR(1) removal process provides a detrending procedure that may help to detect more subtle interannual climate signals. Correlation coefficients between the detrended atmospheric concentrations of HCHs and the NAO index, and HCB and the MEI in springs, are provided in Table 3. By comparing these values with those in Table 2, the detrended HCHs in the air are now more strongly associated with the NAO, especially for α-HCH, the chemical with the most steeply declining slopes (Table 1). This indicates that the interannual variability is related to large-scale atmospheric conditions and climate, as will be further elaborated below.

Table 3. Correlation Coefficients Between the Detrended Atmospheric Concentrations of HCHs and HCB With the NAO Index and MEI, Respectivelya
  • a

    The correlations are statistically significant at the 95% level.

α-HCH versus NAO0.610.0110.680.0040.620.010
γ-HCH versus NAO0.500.0460.550.0290.560.025
HCB versus MEI0.450.0770.770.00050.460.075

[16] Figure 3 shows detrended (by removing linear trend and AR(1)) and normalized daily air concentration anomalies of α-HCH, averaged over the three IADN sites from 1992 to 2007. Because of similarities in the temporal change in the detrended time series of α-HCH at different IADN sites, as will be illustrated in Figure 5 (the similarities are also seen in the measured data as shown in Figure S1), the detrended daily concentrations at individual IADN sites are not shown in this paper. The linear trend of the daily mean air temperature from 1992 to 2007 averaged over the Great Lakes is also plotted in Figure 3. In contrast to the declining trend observed in the mean spring air concentrations alone, the detrended time series of α-HCH show an increase after 1995. This increasing trend appears to be consistent with the increasing trend of air temperature throughout the 1990s and the 2000s. While the stronger correlation between the detrended time series of atmospheric concentrations of α-HCH and the NAO reveals the connection of the chemical with interannual variability of the NAO, the increasing trend of the detrended time series since 1995 suggests that the change in this chemical in the atmosphere is very likely also associated, in addition to interannual climate variability [Ma et al., 2004a], with decadal or interdecadal variability of the NAO. Figures 4a and 4b plot the spring NAO index from 1950 to 2009 and the detrended spring mean α-HCH time series from 1992 to 2007 averaged over the three IADN sites and the trend of the NAO and detrended α-HCH (Figure 4b). Following a significant decadal increasing trend over the 1980s through the early 1990s [Hurrell, 1995], and a shorter-term decrease from 1995 to 1998, the NAO index tends to undergo another cycle of interdecadal increasing since the late 1990s. As also shown in Figure 4, the detrended spring mean α-HCH nicely corresponds to the NAO index with a correlation coefficient at r = 0.64 (p = 0.006). On the other hand, the spring mean α-HCH concentration in the atmosphere is not strongly associated with local seasonal mean air temperature. This indicates that the interannual changes in α-HCH over the Great Lakes were caused by a large-scale climate phenomenon and long-range atmospheric transport [Ma et al., 2004b]. Although the measured time series of OCPs over the Great Lakes is less than 20 years, the increase in the detrended α-HCH air concentrations after 1995 is likely related to the interdecadal variability of the NAO. As seen from Figures 3 and 4, the detrended α-HCH time series agrees very well with the upward trend of NAO since 1996 and mean air temperature over the Great Lakes. Given that the increasing trend of the interdecadal variation of the NAO has been linked to global warming [Hoerling et al., 2001], the current time series of the measured OCPs in the Great Lakes by the IADN program are likely already to contain signals of global climate change.

Figure 3.

Detrended daily α-HCH time series (solid blue line) and daily α-HCH concentration anomalies (departure from mean, dashed line) normalized by its standard deviation. Both time series are averaged over the three IADN sites. Linear trend of 12 day running averaged air temperature averaged over the Great Lakes region is also plotted.

Figure 4.

(a) Spring NAO index from 1950 to 2007 and detrended spring α-HCH from 1992 to 2007 averaged over the three IADN sites. (b) Time series of detrended spring mean α-HCH concentration in the atmosphere, averaged over the three IADN sites, and mean spring NAO index. Dashed lines indicate fourth-order polynomial fit of detrended spring mean α-HCH and NAO index.

[17] Figure 5 plots the time series of seasonal HCHs and HCB at the three IADN sites after removing their respective linear trends and AR(1). These detrended time series at different IADN sites exhibit similarities in their temporal variation. The detrended time series of α-HCH and γ-HCH also show an increase after 1995, especially for α-HCH. The difference in the detrended time series between α-HCH and γ-HCH is caused by different emission sources and patterns. Canada (especially the Canadian Prairies) has been demonstrated to be the major source of γ-HCH to the Great Lakes before its phaseout in 2004 [Ma et al., 2003b; Ma et al., 2004b]. Direct emission following its application in Canadian agriculture largely drove the temporal and spatial distribution of γ-HCH. The chemical α-HCH was banned in the United States and Canada in 1978. Its temporal and spatial pattern has been attributed to the secondary emission from its historically contaminated reservoirs across North America and other parts of the Northern Hemisphere [Hoff et al., 1996; Shen et al., 2005; Li and Bidleman, 2003]. For HCB, which is more persistent in the environment, the detrended time series is, once again, characterized by high concentration in 1998 because of the El Niño event [Ma et al., 2004a] following its decline from the early 1990s (see also Figure 1). In addition to the good correlation between detrended OCPs and climate indices as shown in Table 3, the detrended spring HCHs and HCB atmospheric concentrations are also associated with the spring air temperature (stippled line) averaged over the Great Lakes (Figure 5), especially before 1996. This demonstrates again that the detrended OCPs' time series link strongly with climate on either an interannual or decadal scale when compared with the measured time series, which merely exhibit a decreasing trend from the 1990s to the 2000s.

Figure 5.

Detrended spring air concentration time series for (a) α-HCH, (b) γ-HCH, and (c) HCB after removing their linear trend and AR(1).

[18] We also calculated the yearly anomaly of the detrended spring α-HCH concentrations in the atmosphere as the departure from their mean. We then sorted the anomalies of the detrended α-HCH in an ascending order. There are eight positive anomaly years (1992, 1994, 2007, 2006, 1998, 2003, 2005, and 2002) and seven negative anomaly years (1995, 1993, 1999, 1997, 1996, 2001, and 2004). Although the mean anomaly in 2000 averaged over the three IADN sites is also positive, it was negative at STP in the same year and, hence, is not included. The statistical difference between two means of these two groups of the detrended α-HCH anomalies during positive and negative year periods can be determined using a t test through computation of statistic t [Wilks, 1995]. On the basis of the calculation, the statistic t is equal to 4.05 (>t0.01 = 3.01) with the statistically significant level of 99%, indicating that the two groups of data during the positive and negative anomaly years are statistically different.

[19] Among the eight positive anomaly years of α-HCH, five years strongly correspond to positive NAO years with the spring NAO index >0.55, and the negative NAO index occurred in 1998, 2005, and 2006. The lowest spring value (−1.127) of the NAO index from 1992 to 2007 occurred in 2005, whereas the detrended α-HCH anomaly in this spring is positive (0.095). A strong reversal of the NAO from a positive to negative phase took place from the wintertime (the NAO index equals 0.89) to spring (−1.127) in 2005. It is likely that the lag response of the chemical to the winter NAO slowed down the increase in air concentration over the Great Lakes corresponding to the positive phase of NAO in the following spring [Ma and Li, 2006]. The positive to negative phase reversal of the NAO also occurred from winter to spring 2006, during which the NAO index changed from 0.107 (winter) to −0.393 (spring). Although the NAO index was also negative in the spring of 1998, the increase in air concentration of α-HCH (and other OCPs) over the Great Lakes has already been linked to the strong El Niño event [Ma et al., 2004a]. To elucidate how the detrended α-HCH concentration anomalies are linked with large-scale atmospheric conditions (and the NAO), we consider a simple composite of the difference of the spring surface air temperature between the eight positive anomaly years and the seven negative anomaly years. Results are demonstrated in Figure 6a. The magnitude of the composite difference ranges from 0.5°K to 2°K in western and southwestern North America and the Midwest United States, indicating that the positive anomaly years of the detrended α-HCH time series are more subject to warm springs than the negative anomaly years. The warmer springs would, in turn, enhance volatilization of OCPs from reservoirs accumulated from past use.

Figure 6.

(a) Composite difference of the spring surface air temperature (°K), derived from the mean surface air temperature averaged over positive anomaly years minus the mean surface air temperature averaged over negative anomaly years of detrended α-HCH time series. Image is produced from the NOAA–Cooperative Institute for Research in Environmental Sciences (CIRES) Climate Diagnostic Center through their Web site, (b) Correlation coefficients between detrended spring mean α-HCH air concentration averaged over the three IADN sites (EGH, SBD, and STP) and surface air temperatures. Also indicated are positive correlations (solid black lines) and negative correlations (dashed black lines). White contours encircle regions where the regression is significant at greater than or equal to 90% confidence level.

[20] The influence of the air temperature on the detrended OCPs time series can be further highlighted by a spatial linear regression analysis between the detrended spring averaging α-HCH and SAT time series. This was done by regressing the time series of detrended mean spring α-HCH (averaged over three IADN sites) against the SAT values over the region between 40°W–140°W and 20°N–70°N at a 2.5° × 2.5° grid resolution from 1992 to 2007. Results are presented in Figure 6b. The spatial distribution of the correlation coefficients indicates a statistically significant relationship (≥90% confidence) between higher α-HCH air concentrations over the Great Lakes and the increase in the SATs in the region extending from the west (the Midwest United States to the Great Lakes). The interannual change in the atmospheric level of α-HCH can be linked with reemission from its reservoirs in the contaminated terrestrial surfaces driven by the change in the SAT. Under strong positive correlations (also indicated by solid black lines in Figure 6b) between detrended α-HCH air concentration over the Great Lakes and the SATs as well as prevailing westerly wind regime during springs across the west (Midwest United States to the Great Lakes), these regions are possibly a secondary emission source of the substance measured in the Great Lakes region.

[21] The detrended time series of HCB air concentration are more weakly associated with either MEI or NAO as compared with the nondetrended time series (Tables 2 and 3). The causes are not clear for the weak response of the detrended HCB air concentrations to the two climate indices. This is likely due to its greater persistence and slower interannual variation in the environment compared with the HCHs. Hence, the residue time series of HCB might not be strongly fluctuated under climate variation.

[22] There are no significant correlations between the three toxic chemicals and climate indices in the summer period. However, correlations between the indices of spring NAO and MEI and mean summer air concentrations of α-HCH and HCB are observed. The spring MEI and summer α-HCH air concentrations are correlated at r = 0.57, 0.65, and 0.66 at the three IADN stations, EGH, SBD, and STP, respectively. This confirms the lag response of summer OCPs to spring climate variations as reported previously by Ma and Li [2006]. Given that interannual climate signals are most prominent during winter and spring seasons [Ma et al., 2004a], these lag correlations may carry significant implications for the impact of climate variations on OCP air concentrations over the Great Lakes region. The lag response of atmospheric concentrations of OCPs to those dominant climate variabilities in the Northern Hemisphere may provide a basis for forecasting the temporal trends of these toxic chemicals over the region.

[23] In examining the connections between atmospheric concentrations of OCPs monitored over the Great Lakes region and tropical Pacific sea surface temperature anomalies, Ma and Li [2006] found that the interannual variability in the atmospheric concentration of α-HCH observed in the Great Lakes during the summer season strongly correlates with cooling in the western tropical Pacific during the previous winter and warming in northeast Asia, which is the major source of α-HCH [Li and Bidleman, 2003]. This suggests that the year-to-year fluctuation of α-HCH air concentrations in the Great Lakes region during warm seasons may also display some climate signals in the Asia-Pacific region.

[24] Because it is currently still in use, interannual variation in α-endosulfan is expected to be primarily related to its agricultural application but presumably not to interannual climate variability. Indeed, the mean spring air concentration, after removing the AR(1), was not significantly correlated with climate indices at two of the three IADN sites. However, the decreasing trend of α-endosulfan in the summertime (see Figure 2) before 2005 appears to correspond well to the negative phase of the spring NAO index. Correlations were observed between the spring NAO index and the mean summer air concentration of α-endosulfan (r = −0.63 at EGH, r = −0.36 at SBD, and r = −0.74 at STP), indicating a somewhat surprising negative relationship between these variables. Between 1995 and 2004, the spring NAO index was negative in 1995, 1996, 1998, and 2001. The composite anomaly of the surface air temperature in these four years indicated cold springs over the Great Lakes region (Figure 7), which could be attributed to the lower levels of α-endosulfan in the air [Burgoyne and Hites, 1993]. It is not clear if the agricultural application of endosulfan during the summertime could be affected by spring air temperatures. The relationship between air temperatures, the NAO index, applications of α-endosulfan, and its air concentrations is worthy of further study. The summer air concentration of α-endosulfan at SBD and STP increased dramatically in 2005, though the spring mean air temperature dropped slightly in 2005 compared with 2004. The reason for the sharp increase in the air concentration of α-endosulfan is not clear yet but likely resulted from changes due to regulations.

Figure 7.

Composite anomaly of the spring surface air temperature in 1995, 1996, 1998, and 2001. The temperature anomalies are computed as the departure from the mean surface air temperature from 1968 to 1995. Image is produced from the NOAA-CIRES Climate Diagnostic Center through their Web site,

5. Conclusions

[25] Using extended time series of OCPs from 1992 to 2007 across the Great Lakes, we have examined and established more robust trends for air concentrations of HCHs, HCB, and α-endosulfan by using the MK statistical test with a process to remove a lag 1 autoregressive response. The MK test provides not only the statistical significance for the air concentration trends but also the degree of decreasing trends of OCPs air concentrations at each IADN monitoring site. Using these extended time series of air concentrations, we have also revisited the relationships between OCPs in air over the Great Lakes and major interannual climate variabilities in the Northern Hemisphere, namely, the North Atlantic Oscillation (NAO) and the El Niño–Southern Oscillation (ENSO). Compared to the previous findings, which used relatively shorter time series of atmospheric concentrations [Ma et al., 2003a, 2004a, 2004b; Ma and Li, 2006], present results indicate that, under the declining trend of OCPs in the Great Lakes, stronger climate signals may be extracted from detrended (residual) time series of the atmospheric concentration of OCPs. Considering that the influence of global warming may change (increase) the rates and quantities of OCPs reemitted from their reservoirs, their declining in the atmosphere is likely slowed down. Given that the detrended time series of OCPs exhibits a decadal or interdecadal trend, detrending OCPs time series may shed light on the detection of the association of OCPs with interdecadal climate variation and global climate warming.


[26] The authors thank the Indiana University IADN team and Ron Hites for the use of U.S. atmospheric OCP concentration data and constructive comments on the paper and Helena Dryfhout-Clark for providing the IADN data.