Abrupt changes in air temperature and precipitation: Do they matter for water chemistry?



[1] We analyzed 120 years long time series of air temperature and precipitation from 29 respective 44 sites distributed all over Sweden and determined abrupt changes by using three methods. For air temperature we found significant changes in 1930 and 1989 and for precipitation in 1920, 1979, and 1998. Analyzing more than 30 yearlong time series of ice cover (333 sites), discharge and watercourses chemistry (87 sites), we observed abrupt changes in 1977, 1989, and 1998 for discharge but first in 1998 for watercourses chemistry, most pronounced for organic matter and sulfate concentrations. We suggest that the abrupt increase in air temperature in 1989 liberated more easily mobilized organic matter in the catchments, which, for water chemistry, was first detected in 1998 as a consequence of increased discharge. We conclude that increases in air temperatures can make ecosystems more sensitive to further changes in precipitation.

1. Introduction

[2] Ecosystems can change in many ways. These changes can either be expected or surprising. Especially the surprising changes are a challenge for ecosystem management as there has not been any time to prepare for them. Surprising changes are often associated with abrupt changes. They can lead to catastrophic events [Scheffer and Carpenter, 2003], explaining why their importance is increasingly recognized [e.g., Andersen et al., 2006; Beaugrand, 2004; Blenckner et al., 2004; Chang, 2004; Hargeby et al., 2006; Kingston et al., 2006; Kishi et al., 2005; Yonetani and Gordon, 2001]. Abrupt changes can be reversible or irreversible. Irreversible changes that create alternative stable states have been observed for a variety of ecosystems, mainly with focus on biological processes [Carpenter, 2003]. Recently, climate-induced abrupt changes have been pointed out as a major threat to ecosystem services, since abrupt changes in physical restrictions of the environment such as temperature, flow regimes and durations of seasons seem to have a profound impact on terrestrial and aquatic ecosystems [Alheit et al., 2005; Carpenter and Brock, 2006].

[3] In this study we used 120 yearlong time series of air temperature and precipitation to detect abrupt changes. We hypothesized that those abrupt changes in air temperature and in precipitation cause abrupt changes in water chemical variables, i.e., water chemistry responds linearly to changes in meteorological variables. However, we know from an earlier study that air temperature changes can also lead to nonlinear responses in water systems. Weyhenmeyer et al. [2004] showed that the timing of ice breakup responds in a nonlinear way to changes in air temperatures with significant responses only in the warmer geographical regions [Weyhenmeyer et al., 2005]. Since a variety of water chemical variables are assumed to be related to the timing of ice breakup we further hypothesized that water chemical variables driven by ice cover dynamics will show abrupt changes only in the warm southern part of Sweden. To test the hypotheses we used long-term data series from 87 watercourses and 333 lakes distributed all over Sweden.

2. Methods

2.1. Data

[4] From the NORDKLIM-database (SMHI 2003, http://www.smhi.se/hfa_coord/nordklim/index.htm) we used yearly mean air temperature data from 29 Swedish sites and yearly sum precipitation data from 44 Swedish sites from 1880–2001 [Tuomenvirta et al., 2001]. In addition, we used monthly mean air temperature data (72 sites, 1961–2005), monthly sum precipitation data (277 sites, 1961–2005), daily discharge data (87 sites, 1965–2004) and yearly dates of ice breakup (333 lakes, 1960–2004) from the Swedish Meteorological and Hydrological Institute (SMHI) (Table 1). Depending on the length of the time period considered different results might be obtained, a reason why we included the long time series from 1880 in our study.

Table 1. Used Variables From Swedena
VariablesExplanationsSitesTime PeriodSampling Frequency
Tlongair temperature, long time291880–2001yearly means
Tmanyair temperature, short time721961–2005monthly mean
Plongprecipitation, long time441880–2001yearly sum
Pmanyprecipitation, short time2771961–2005monthly sum
Ice breakice breakup3331960–2004Julian day from 1 Jan
Qdischarge871965–2004monthly mean
AbsFabsorbance 420 nm, 0.45 μm871975–20051 sample per month
AbsF:COD 871975–20051 sample per month
Turbidityabsorbance unfiltered - AbsF871975–20051 sample per month
BC*base cations Σ(Ca, Mg, K, Na) − 1.111 × Cl871975–20051 sample per month
Clchloride871975–20051 sample per month
C:NCOD per Organic-N871975–20051 sample per month
CODchemical oxygen demand determined using KMnO4871975–20051 sample per month
Feiron491975–20051 sample per month
Mnmanganese491975–20051 sample per month
InorgNNO2−N + NO3−N + NH4−N871975–20051 sample per month
OrgNKjeldahl-N − NH4-N871975–20051 sample per month
pH 871975–20051 sample per month
PO4phosphate-phosphor871975–20051 sample per month
Other-Pnon reactive P: Tot-P − PO4-P871975–20051 sample per month
Sisilica871975–20051 sample per month
SO4sulfate871975–20051 sample per month

[5] The catchments of the 87 investigated watercourses are generally dominated by forest, and to a lesser degree covered by lakes, arable fields and wetlands (Table 2). The watercourses were sampled once per month during 1975–2005 (Table 1) (data from: http://info1.ma.slu.se/db.html). All watercourses chemical analyses were performed by a certified laboratory according to EN or ISO standards [Wilander et al., 1998]. Whatman GF/C glass fiber filters (0.45 μm) were used to determine the filtered absorbance at 420 nm in a 5 cm quartz cuvette. Turbidity is estimated by the difference between unfiltered and filtered absorbance. The Chemical Oxygen Demand (COD) was determined by using KMnO4 as an oxidizing agent. The pH was measured on unaerated samples. Calcium, magnesium, sodium and potassium were measured with ICP-OES. Chloride and sulfate were measured using ion chromatography and nitrate using cadmium reduction with a segmented flow analyzer. From 1975–1983 sulfate was measured with the Mackereth method [Mackereth, 1955]. Approximately 100 Swedish surface water samples were run in parallel using the two sulfate methods [A. Wilander, personal communication, 2006], and the data before 1984 were adjusted by linear regression to the ion chromatography method. Calculations of non-marine salts are based on chloride concentrations and ion ratios for seawater [Umweltbundesamt, 1996].

Table 2. Catchment Size (km2) and Land-Use Coverage (%) for the 87 Swedish Watercourses Catchments
Latitude, outlet59°N 55′66° 27′55° 47′
Longitude, outlet16°E 17′24° 14′11° 44′
Catchment size24534703523
Lake surface6350
Densely built-up area0110
Open field4130
Arable field6760
Mountain forest0220

2.2. Abrupt Change Detection

[6] With abrupt changes we mean a sudden stepwise change in the time series. We used three methods to detect abrupt changes in long-term time series, the Excel add-in “Sequential Regime Shift Detection version 3.2” (SRSD) (http://www.beringclimate.noaa.gov, see also Rodionov and Overland [2005], Rodionov and Martin [1999], and Rodionov [2004, 2006]), the software Change-Point Analyzer (version 2.3, Taylor Enterprises Inc, http://www.variation.com) and the manually performed CUSUM/Pettit-test [Buishand, 1982; Lanzante, 1996; Pettitt, 1979]. Since all three methods revealed similar results we focused on the first method. The SRSD is easy to use and many sites can be analyzed simultaneously using yearly mean values. The SRSD assumes no a priori of when the abrupt change should occur, but the time series must be continuous. In the very few cases (0.06%) where a yearly mean value was missing (only for water chemistry), the mean from the antecedent year and the following year was used. The differences in the mean between two time periods were tested with sequential Student's t-test, assuming that the variances for both periods are the same [Rodionov, 2004]. Serial correlation was modeled by the first order autoregressive model (AR1) [Rodionov, 2004]. The AR1 coefficient in this study was estimated by MPK (Marriott, Pope and Kendall) [Kendall, 1954; Marriott and Pope, 1954]. The other two methods of estimating the AR1 coefficient in SRSD [Rodionov, 2006], ordinary least squares (OLS) and inverse proportionality with 4 corrections (IP4), were initially also tested but they seem to give the same results as MPK. The Huber [2005] weight parameter reduces the effect of outliers by weighing them inversely proportional to their distance from the mean. The regime shift index (RSI) is the cumulative sum of the normalized anomalies [Rodionov, 2004]. The higher the RSI value is the stronger is the abrupt change. The time series were checked visually to confirm that the change was abrupt, which was the case for all abrupt changes.

[7] The shown values in results and discussion are based on MPK and the parameters were set to a significance level of 0.05, a cut-off length of 8 years and a Huber's weight parameter of 6. The residuals were checked for shifts in variance by SRSD, but no shifts were detected in the variance. Regional pattern of the abrupt change in 1988/89 were compared by grouping each site by rounding the latitude to integers. Principal Component Analysis (PCA) was used to evaluate which variables showed similar pattern during the abrupt changes, by using the software SIMCA-P (version 10.5, Umetrics AB).

3. Results

[8] For the long-time series (1880–2001) two abrupt changes in the yearly mean air temperature were detected in 1930 (at 24% of the 29 sites, RSI = 2.3) and in 1989 (at 45% of the sites, RSI = 2.4) (Figure 1a). Three abrupt changes in the yearly mean sum of precipitation were detected in 1920 (at 9% of the 44 sites, RSI = 0.9), in 1979 (at 7% of the sites, RSI = 0.9) and in 1998 (at 36% of the sites, RSI = 11) (Figure 1b). For the shorter long-term time series (1961–2005), for which more sites were available, abrupt changes in the yearly mean air temperature were detected in 1988 (at 25% of the 72 sites in 1988, RSI = 3.5), in 1989 (at 51% of the sites, RSI = 4.7), in 1999 (at 7% of the sites, RSI = 1.4) and in 2000 (at 6% of the sites, RSI = 1.9) (Figure 1c). Abrupt changes were found for the 277 precipitation sites in 1987 (0.7% of the sites, RSI = 0.1), in 1997 (1.1% of the sites, RSI = 0.2), and in 1999 (0.7% of the sites, RSI = 0.4) and in 2000 (0.4% of the sites, RSI = 0.6) (Figure 1d). Lake ice breakup showed abrupt changes in 1988 (at 19% of the 333 lakes, RSI = 70), in 1989 (at 7% of the sites, RSI = 13), in 1998 (at 2% of the sites, RSI = 0.7) and in 1999 (at 7% of the sites, RSI = 7.4) (Figure 1e). For discharge we detected abrupt changes in 1977 (at 7% of the 87 sites, RSI = 2.7), in 1989 (8% of the sites, RSI = 1.8), and in 1998 (13% of the sites, RSI = 2.3) (Figure 1f).

Figure 1.

Z-score values of physical parameters in Sweden; air temperature ((a) referring to 29 sites from 1880 to 2001 and (c) referring to 72 sites from 1961 to 2005), precipitation ((b) referring to 44 sites from 1880 to 2001 and (d) referring to 277 sites from 1961 to 2005), lake ice breakup date ((e) referring to 333 sites from 1960 to 2004), and discharge ((f) referring to 87 watercourses sites from 1965 to 2004). Grey vertical lines are the years when significant abrupt changes were identified using SRSD, and the percentage is the number of significant sites. Please notice that the scale of the axis differs. See section 2 for more details, for abbreviations see Table 1.

[9] The strong abrupt change in air temperature in 1988/89 and the abrupt change in discharge in 1989 was hardly reflected as an abrupt change in the water chemical time series (Figure 2). The abrupt change in precipitation in 1998/99, however, coincided with an abrupt change in discharge (Figure 1f) and in the watercourses chemistry, in particular in absorbance (in 1998 17% of the sites, RSI = 10; in 1999 5% of the sites, RSI = 2.4), COD (in 1998 10% of the sites, RSI = 6.2; in 1999 7% of the sites, RSI = 5.9) and sulfate (in 1998 9% of the sites, RSI = 9.9; in 1999 22% of the sites, RSI = 18) (Figure 2). The abrupt change in discharge correspond also to changes in the ratio between absorbance and COD (in 1998 5% of the sites, RSI = 1.6; in 1999 3% of the sites, RSI = 0.8), turbidity (in 1998 4% of the sites, RSI = 2.4; in 1999 1% of the sites, RSI = 0.5), organic nitrogen (in 1998 5% of the sites, RSI = 3.1; in 1999 8% of the sites, RSI = 6.0) and chloride (in 1998 4% of the sites, RSI = 1.8; in 1999 none) (data not shown for the last three parameters in Figure 2). An abrupt change in sulfate was observed when abrupt changes of air temperature or precipitation were absent and was most probable due to the change in the analysis method (in 1984 14% of the sites, RSI = 8.7) (Figure 2).

Figure 2.

Z-score values of (a) absorbance, (b) COD, (c) ratio absorbance per COD, (d) C:N, (e) sulfate and (f) iron in 87 Swedish watercourses sites (1975–2005). Grey vertical lines are years when significant abrupt changes were identified. Percentage is the number of significant sites. See section 2 for more details, for abbreviations see Table 1.

[10] The abrupt change in air temperature in 1988/89 occurred throughout Sweden but the abrupt change in lake ice breakup during the same years occurred mainly only in southern Sweden, most pronounced at latitudes around 56 to 59°N (Figure 3). Discharge showed an abrupt change at 7 sites in 1988/89 mainly at the latitude 61°N (Figure 3). At this latitude also COD and C:N showed abrupt changes but these changes occurred not in the 7 watercourses for which abrupt changes had been detected. First in 1998/99, the spatial pattern of the abrupt change in discharge corresponded to the spatial pattern of the abrupt change in watercourses chemistry with most pronounced changes at latitudes around 58 and 65°N. 12 sites had a significant increase in discharge in 1998/99, and of these 12 watercourses 9 had a significant increase in absorbance, 6 in COD, 4 in the ratio absorbance and COD, and 8 sites with a significant decrease in sulfate.

Figure 3.

Gradient along Sweden during 1988/89. Large circle indicate that many of the sites at that latitude had a significant abrupt change using SRSD. Upper values show the percentage of significant sites of totals sites for that parameter and “+” denotes an increase. Air temperature and precipitation are based on 1961–2005 time series. The sites were compared by grouping each site by rounding the latitude to integers. See section 2 for more details, for abbreviations see Table 1.

[11] Using Principal Component Analysis (PCA) on the 1998/99 SRSD results for water chemistry and discharge revealed that absorbance, COD, ratio absorbance and COD, organic nitrogen, sulfate, turbidity, iron and silica were grouped together along the first principal component (PC) axis together with discharge (data not shown). Opposite to this group, along the same PC axis, were inorganic nitrogen, C:N, other-P, manganese and pH. The second PC axis was made up of phosphate, chloride and base cations. All remaining variables were located on the other end of the second PC axis.

4. Discussion

[12] With the method used in this study we achieved the same results as found earlier in Scandinavia, i.e., an abrupt increase in air temperature in 1988/1989 [Busuioc et al., 2001; Fealy and Sweeney, 2005; Hagen and Feistel, 2005; Troell et al., 2005]. We were, however, not able to identify many abrupt changes in the watercourses chemistry during the same years, although even water discharge had significantly changed at some sites (Figure 3). The lacking abrupt changes in water chemistry (low RSI at only a few sites) in 1988/1989 were unexpected. Instead we found a strong agreement between abrupt changes in the yearly sum of precipitation and consequently the yearly mean discharge and abrupt changes in the watercourses chemistry later in the time period in 1998/99 (Figures 1 and 2). There are several possibilities for the late water chemical responses. One possibility is that the change in air temperature itself was not enough to cause an abrupt change in watercourses concentration. In this case we would assume a linear response of water chemistry to changes in air temperature. Another possibility is that the response of water chemistry to changes in air temperature is nonlinear, i.e., a gradual change in temperature will have an effect on watercourses chemistry first after a threshold is reached. A third possibility for the late water chemical response could be that the abrupt air temperature increase strongly affects biological and chemical processes in the catchment that first become apparent in the watercourses when the discharge increases. This possibility would explain why we see a much more pronounced effect of an abrupt increase in precipitation and discharge in 1998/99 after the extreme air temperature increase than during 1988/89 when the air temperature increase occurred. Present climate change scenarios for Sweden determine an increase in air temperature while precipitation and discharge are expected to increase or decrease depending on the region in Sweden. In addition, an increase in the temperature and precipitation variability is expected [Andréasson et al., 2002; Rummukainen et al., 2003]. According to these scenarios the catchments runoff chemistry could become more sensitive to changes in discharge in the future.

[13] The methods of Rodionov [2004, 2006] have worked well in many different environmental assessment studies [Friedland et al., 2007; Hunt and Elliott, 2005; King and McFarlane, 2006; Notaro et al., 2006; Wilson et al., 2007]. When comparing Rodionov's methods with other methods we found consistent results, e.g., the homogeneity test used by Alexandersson [2002] for Swedish meteorological data from 1860 to 2001 identified the same years for abrupt changes as we found in our study.

[14] Point-sources had noticeably impacts on Swedish environmental watercourses sites. Many point-sources were reduced during the 1970s, with clear effects on the water chemistry. The recovery from anthropogenic impacts was longest for phosphorus concentrations that stabilized first in the early 1980s [Wilander and Persson, 2001]. Air deposition of sulfate peaked in Sweden in the 1970s [Mylona, 1996], while nitrogen deposition stabilized during the 1990s [Wilander, 2001]. We used time series from 1975 onwards when recovery from point-sources was at its end, i.e., we consider that our time series are suitable for climate impacts studies. However, abrupt changes in the evaluated variables could also be due to changes in analytical methods. Sulfate analytical method changed in 1984 (from Mackereth to ion chromatography) which coincide with the detected abrupt change in 1984.

[15] Of all chemical variables tested, absorbance and COD showed the clearest response to an abrupt change in precipitation and discharge. Absorbance and COD are two methods of estimating the concentration of organic matter. Organic matter has a large impact on chemistry [Schwarzenbach et al., 2003; Stumm and Morgan, 1996], and biota [Wetzel, 2001], as well as on human society [Eikebrokk et al., 2004]. Increasing concentrations of organic matter in surface waters have been observed in many regions and several causes are debated, among them runoff processes [Erlandsson et al., 2008], atmospheric deposition [Evans et al., 2005, 2006; Monteith et al., 2007], and climate [Forsberg and Petersen, 1990]. During the vegetation period and especially after that period large amounts of organic matter are stored in the soil [Moore, 2003]. Droughts during the vegetation period will build up even large amounts of organic carbon in the soil [Worrall et al., 2006]. When the discharge increases, usually in the autumn in Sweden, large amounts of organic carbon is flushed out to the nearby surface waters [Bishop et al., 2004; Laudon et al., 2004]. Allochthonously derived organic matter is more colored, it has a higher ratio between absorbance and COD, and it has more carbon per nitrogen (C:N) than autochthonous derived [Campbell et al., 2000; Easthouse et al., 1992; Lovett et al., 2000]. The increase in air temperature in 1988/89 could liberate more and more easily mobilized organic matter in the system, making the system more sensible to changes in precipitation [Freeman et al., 2001; Tranvik and Jansson, 2002]. The concurrent abrupt increase in precipitation and discharge with absorbance, COD and sulfate in 1998/99, and to a lesser degree the ratio between absorbance and COD, turbidity, organic nitrogen and chloride (Figures 1 and 2) could be interpreted as a higher groundwater table activating new soil layers with more colored organic matter and more water that dilutes sulfate.

[16] Our hypothesis that abrupt changes in air temperature cause abrupt changes in the timing of ice breakup only in the southern part of Sweden due to a nonlinear response of ice breakup to changes in air temperatures [Weyhenmeyer et al., 2004] could be confirmed in this study. However, we were not able to find any relation between changing patterns of the timing of lake ice breakup and the yearly mean watercourses chemistry. The reason for this might be that lake ice breakup might be an inappropriate indicator for the yearly mean watercourses chemistry that seems to be mainly driven by catchment characteristics and discharge. The strong impact of discharge on watercourses chemistry was confirmed by our PCA and the fact that abrupt changes in discharge corresponded to abrupt changes in water chemistry in 1998/99.

[17] From our results we suggest that abrupt changes in air temperature will strongly change biogeochemical processes in the catchment that first become apparent in watercourses when discharge patterns change. Although discharge is likely to be the predominat driver for watercourses chemistry we found a stronger effect of discharge on water chemistry after an abrupt air temperature increase compared to the effect when there was no preceding warming event. From this we conclude that abrupt changes in air temperatures can change the relationship between discharge and water chemistry, an issue that becomes relevant for water quality management.


[18] The authors wish to thank the Faculty of Natural Resources and Agricultural Sciences at the Swedish University of Agricultural Sciences, the Swedish Research Council (621-2005-4335) and the Royal Swedish Academy of Sciences (Knut and Alice Wallenberg Foundation) for financial support of this research. The authors are thankful to the Swedish Environmental Protection Agency and the IMA laboratory for financing, sampling and analyzing numerous water samples. Data were kindly received from the Swedish Meteorological and Hydrological Institute.