The water levels of Lakes Michigan and Huron have been monitored since 1865, and numerous attempts have since been made to connect their variations to potentially predictable large-scale climate modes. In the present study, the levels are analyzed after outflow-related damping effects were removed, increasing the transparency of the lake level fluctuations and potential climate connections. This filtering exposes a large oscillation which is connected to the Atlantic Multidecadal Oscillation (AMO), and a ∼27-yr periodicity that is likely resulting from the intermodulation of two near-decadal cycles originating in the North Atlantic region. While the lake level fluctuations prior to 1980 were predominately driven by changes in precipitation, it is now found that for the first time in our years of record, evaporation has begun to significantly contribute to lake level changes. Summertime evaporation rates have more than doubled since 1980 as a result of increasing water-surface temperatures, which are significantly correlated with decreasing wintertime ice cover.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The water levels of Lakes Michigan and Huron have been of interest for many years due to the large impacts that their changes have on surrounding communities. These impacts range from increased shoreline erosion when the levels are higher than normal, to inflated shipping costs when the levels are too low for cargo ships to carry typical-sized loads. When lake level extremes are not foreseen, the negative impacts are inevitable, as proper preparation is oftentimes not possible. The most reliable lake level fluctuations are those that arise from seasonal changes, resulting in a relatively predictable annual cycle. In the winter months, evaporation is high and precipitation is often stored in snowpack, which results in decreasing water levels, whereas in the summer, the levels tend to rebound.
 While the annual cycles are important, and noticeably contribute to the overall lake level variations, the standard deviation of the monthly mean is actually only about 30% of the annual standard deviation about the long-term mean (see Text S1 of the auxiliary material). Because the interannual extremes are only compounded when also considering the annual cycle, it is important to obtain an improved understanding of these changes which are likely resulting from regional climate fluctuations. Connections to potentially-predicable climate modes can lead to improved lake level forecasting ability, which would result in better preparation and decreased negative effects from these lake level extremes.
Rodionov  linked wintertime Great Lakes-region precipitation changes to differences in cyclone origins associated with the Pacific/North American teleconnection pattern (PNA) while Grover and Sousounis  discovered changes between mid and late-century large-scale flow patterns during fall months, which also resulted in altered precipitation characteristics. Furthermore, Namdar-Ghanbari and Bravo  identified significant correlations between the Great Lakes' water levels and various teleconnection patterns, and Hanrahan et al.  suggested a connection between two near-decadal cycles of the Michigan-Huron system and recurring sea-surface temperature patterns in the North Atlantic region.
 The importance of linking interannual lake level fluctuations to climate is now becoming more urgent due to anticipated changing precipitation and temperature trends resulting from global climate change. The unpredictable historic lake level extremes have already proven problematic, and future uncertain trends will likely only amplify these issues. Our study expands on the work of Hanrahan et al.  through the analysis of updated datasets, and the use of an outflow-removed filtering technique which increases the transparency of interannual lake level fluctuations.
2. Historic Lake Level and Component Data
 Lake level monitoring began for the Michigan-Huron system in 1865, and many analyses have since been conducted on the resulting time series. One of the first attempts to identify cycles to aid in lake level forecasting was made by Cohn and Robinson , who utilized spectral analysis to verify the existence of the annual cycle, but also noted spectral peaks with multidecadal periods. Polderman and Pryor  qualitatively identified an approximate 30-yr lake level periodicity in agreement with the findings of Thompson and Baedke  who studied coastal ridges surrounding Lake Michigan. Until recently, however, the historic record has proven too short to statistically verify the existence of these, or any other, multi-decadal lake level cycles.
 For the present study, monthly average lake levels were obtained from the U.S. Army Corps of Engineers for January 1865 to January 2009. In addition, components that contribute to the lake level fluctuations were obtained online from the Great Lakes Environmental Research Laboratory (GLERL), which are available for various spans of time in the form of total monthly linear units over the lake's surface. These components consist of the net basin supply (NBS), which is made up of the total annual over-lake precipitation Py, runoff Ry, and evaporation Ey. Also included, are the total annual inflows Iy from Lake Superior and the outflows Oy into the lower Great Lakes. The resulting water balance equation is as follows:
where Ly (Figure 1a, thin black line), is the beginning lake level in a given year as estimated by the January average level, and ɛy is the residual error resulting from additional negligible lake level sources, and measurement uncertainties. Assuming no error bias, ɛy should largely cancel out over the entire time series, however, there exists a negative error trend about 0.0018 meters per year after 1960, which can likely be attributed to factors not pertinent to this study (see Text S2). To maintain a focus on climate-related fluctuations, the trend is removed from the lake level time series after removing outflow effects as described in the next section.
3. Outflow-Removed Lake Level and Climate Connections
 The outflow rates, which represent the combined total flow of the St. Clair River and the Chicago Diversion, are directly proportional to the levels of the Michigan-Huron system. Because of this strong dependence of outflow on the levels themselves, a natural damping of the lake level variation occurs. For example, when there exists a duration of increased precipitation, the lake levels tend to rise, and in response, the outflow increases, lessening the extent to which the levels climb. Consequently, the fluctuations which might otherwise result from atmospheric variability are never fully realized, potentially concealing obvious connections. Because of these damping effects, and because the outflow fluxes do not directly respond to external lake level drivers which are the primary interest of this study, it is useful to remove the outflow component from the historic levels. The details of this filtering, and the removal of the error trend, are outlined in Text S2. The resulting detrended lake level curve Ly−0 will hereafter be referred to as the “outflow-removed” level, and is illustrated in Figure 1a (thick black line).
 The most striking feature of the new outflow-removed lake level time series is an apparent low-frequency, large-amplitude, oscillation with a period of about 80 years. Numerous observational and modeling studies have previously identified a similar oscillation in global temperatures, the Atlantic Multidecadal Oscillation (AMO), which is likely associated with oceanic thermohaline circulations [Delworth and Mann, 2000; Gray et al., 2004; Knight et al., 2005]. Kravtsov and Spannagle  derived an AMO loading pattern (Figure 1a, gray line) by forming the difference between observed surface temperatures and ensemble averages of twentieth-century climate simulations, which stemmed from the World Climate Research Programme's (WCRP) Coupled Model Intercomparison Project phase 3 (CMIP3). To address possible errors associated with the calculation of this index, a bootstrap analysis had been conducted which verified the significance of the negative values in the 1920s and 1980s, and the positive values around 1950. There exists a significant correlation of r2 = 0.37 (p < 0.001) between this index, and the outflow-removed lake level.
 In addition to the multidecadal climate connection, there is also evidence that near-decadal climate signals are being transmitted to the water levels. Hanrahan et al.  identified statistically significant 8 and 12-yr periodicities in the annually-averaged lake level time series, which they believe to be connected to the North Atlantic region, where two similar modes have also been identified through spatial EOF analysis of sea-surface temperatures [Deser and Blackmon, 1993; Moron et al., 1998]. It was then speculated that a larger ∼30-yr lake level cycle [Thompson and Baedke, 1997] may result from the intermodulation of these two near-decadal cycles (see Text S3). The existence of this periodicity has now been confirmed by multi-taper method spectral analysis of the high-pass filtered, outflow-removed lake level curve (see Text S2 and Figure S3).
4. Precipitation as a Historic Lake Level Driver
 Historically, precipitation-related fluxes have been the primary drivers of interannual lake level fluctuations [Brinkmann, 2000; Hanrahan et al., 2009]. To determine whether this is also the case with the identified multidecadal lake level oscillation, average basin precipitation and runoff totals are compared during the significant phases of the AMO index, which are concurrent with the most extreme outflow-removed lake levels of opposite sign. This analysis concludes that the average annual totals were about 18 cm higher during the negative phases of the AMO (1916–1924 and 1973–1991), than during the positive AMO years (1932–1967). A t-test was conducted to determine the probability of equal means during the different phases, which concludes that the null hypothesis can be rejected at the 0.01% level. Therefore, similar to interannual lake level variations, interdecadal fluctuations have also primarily resulted from changes in precipitation.
5. Recent Evaporative Effects
 While evaporation is a primary driver of seasonal cycles, its interannual variance has historically been relatively small and has therefore had little influence on interannual lake level changes. Analysis of recent annual evaporation data, however, reveals a positive trend that began around 1980 (Figure 2c), which is now having a significant impact on the long-term water levels. This evolving relationship is clear when examining the time series of lake level anomalies calculated by integrating the sum of the lake level components. Because these component curves have varying trendlines which depend on the time interval under consideration, the pre-1985 linear trends are removed from the full time series (through 2008) as seen in Figure 1b. This period was chosen because Ly−0, and evaporation-excluded levels LyP+R+1, consistently match up until then, after which, the curves and their respective trendlines begin to diverge. Due to increasing precipitation, LyP+R+1 shows a continued increase through the present time, but a separation of Ly−0 occurs in the late 1980s, as it responds to the changing behavior of evaporation LyE (see also Text S4).
 Total annual evaporative losses have increased nearly 25% since 1980, and about half of this increase is evenly distributed throughout the fall, winter and spring months. The remaining and most drastic evaporative changes, however, have occurred solely during the summer months (Figure 2d). This is counterintuitive, since summertime evaporation is typically suppressed due to cool water temperatures relative to those of the overlying air. Therefore, in order to further study the recently increasing summertime evaporation, an analysis of regional surface temperatures is also conducted. Monthly averaged water-surface and over-lake air temperatures were obtained online from GLERL, for the summer months between 1948 and 2004. A recent warming trend can be seen in the region's annual air temperatures, but the greatest temperature increase is observed in the lakes themselves (see Figure S5). The annually averaged water-surface temperatures have increased about 2°C between 1980 and 2004, and while all seasons have seen an increase in water temperatures over the last few decades, summertime temperature increases were the most significant with a jump of about 4°C as seen in Figure 2e (black line). It follows that the differential heating rates between Lakes Michigan and Huron, and their overlying atmosphere, have resulted in record-high water loss through evaporation, primarily in the summer months. This is also substantiated by the strong positive correlation between summertime water-surface temperatures and summertime evaporation totals with r2 = 0.57 (p < 0.001), as illustrated in Figure 2a.
 These findings are consistent with those of Austin and Colman  who found that Lake Superior's summertime water-surface temperatures are also increasing faster than regional air temperatures. They primarily attribute this discrepancy to decreasing wintertime ice cover, which is resulting in earlier annual stratification that subsequently leads to longer periods of radiative warming during the summer months. To determine whether this is also the cause of the increasing Michigan-Huron water-surface temperatures, ice cover data were obtained from the NOAA Great Lakes Ice Atlas [Assel, 2003, 2005]. They are available as daily spatial percentages of ice cover between December 1st and May 31st of each year, and are averaged further to obtain a single annual percentage for each winter between 1973 and 2005. This index represents the average percentage of lake ice for the entire Michigan-Huron system. There exists a strong negative correlation between these indices and succeeding summertime water-surface temperatures of r2 = 0.56 (p < 0.001), which is illustrated in Figure 2b; the time evolutions of these variables are also illustrated in Figure 2e. The recent periods of record-low wintertime ice cover are clearly concurrent with periods of record-high summertime water-surface temperatures, and increased summertime evaporative losses (Figure 2d).
6. Summary and Discussion
 The water levels of Lakes Michigan and Huron have varied since record keeping began in 1865, and their unpredictable extremes have proven problematic. While the annual water level cycle has been fairly well understood, the longer-term and most extreme fluctuations exhibited a high degree of randomness, and clear connections of these fluctuations to climate variability have been lacking.
 In this study, the Michigan-Huron levels are analyzed after damping effects from the system's outflow are removed, and the negative error trend in the lake level time series, is accounted for. This analysis reveals a significant negative correlation between the filtered levels and the AMO index. The apparent connection of the lake levels to the AMO is further supported by previous studies that have found the AMO to have had a strong inverse effect on long-term precipitation patterns over much of the United States, including over the Michigan-Huron basin [Enfield et al., 2001; McCabe et al., 2004].
 In addition to the multidecadal modes, Fye et al.  connected an 8-yr cycle of the North Atlantic Oscillation (NAO) index to moisture anomalies during the late winter to summer months, Small and Islam  identified a 12-yr fall precipitation periodicity (both were found to be strongest over the central U.S), and Jutla et al.  identified a precipitation dipole pattern over eastern North America and southern Canada which was found to have a significant negative correlation with North Atlantic sea-surface temperatures. These periodicities are consistent with lake level cycles identified by Hanrahan et al. , who speculated that their interaction might result in a ∼30-yr lake level periodicity [Thompson and Baedke, 1997]; we have now established the existence of this lake level cycle.
 It was recently determined that long-term lake level fluctuations were primarily resulting from changes in precipitation [Brinkmann, 2000; Hanrahan et al., 2009], and since the pre-1980 interannual variation of evaporation was relatively small, it had little effect on long-term lake level variations. Over the last few decades however, evaporation rates have begun to increase, resulting in decreasing lake levels despite continued above-average precipitation totals. The Michigan-Huron levels have been near record lows since about 2000 (see Figure 1a, thin black line), and the three-year 1997–2000 lake level drop was second only to that which occurred during the 1930s drought [Assel et al., 2004]. Many are still pointing toward channel dredging as the primary reason that the Michigan-Huron lake levels are decreasing [Egan, 2007, 2009], but the present analysis excludes outflow effects and corrects for the questionable negative error trend.
 Most of the increasing evaporative losses are occurring in the summer months due to rapidly increasing summertime water temperatures relative to the overlying air. This differential heating has also been observed in Lake Superior, which is likely a consequence of decreasing annual ice cover [Austin and Colman, 2007]. Reduced spring ice cover results in the redirecting of energy, which would have otherwise been used during melting, increasing the amount of total radiation that is absorbed by the water during the summer months. There exist significant correlations between average wintertime ice cover and summer water-surface temperatures, for both Lake Michigan-Huron and Lake Superior.
 This period of increased evaporation which began around 1980, is consistent with the recent onset of rapid global warming. While the global temperature increase has not been spatially uniform, observations over the U.S., including the Great Lakes' region, have indeed revealed significant increases in annual mean temperature over the last few decades [Easterling et al., 2007; Hansen et al., 2006; Meehl et al., 2007]. Furthermore, earlier seasonal breakups of lake and river ice have previously been observed in the Northern Hemisphere, which are correlated with recently increasing surface temperatures [Magnuson et al., 2000].
 If the current temperature trend continues, it can be assumed that the positive evaporation trend will also continue, but a current study suggests that a recent dynamically-driven climate shift may result in the masking of global warming for the upcoming decades [Swanson and Tsonis, 2009]. If the Great Lakes' region were to experience a leveling off of air temperatures, it is likely that evaporation will no longer exhibit a steady increase, but will instead level off about a higher mean. We conjecture that at that time, the annual evaporation would resume the small interannual variance behavior, and the lake levels would return to their precipitation-flux driven variations about a new, lower level steady state, at least until global temperatures resume their increasing trend.
 In terms of both natural climate variability and anthropogenic forcing, more work needs to be done to gain a better understanding of how the lakes respond to atmospheric changes. The drastic differences between long-term heating rates of the lakes and their overlying air need to be examined further, and the long-term effects of decreasing wintertime ice cover should be estimated in order to improve our understanding of the lake level decreases. Also, because it is possible that the lake levels will at some point, again closely follow changes in precipitation, a clearer understanding of underlying North Atlantic forcing is needed to help further detangle the identified overlapping periodicities.
 We are grateful to K. Kompoltowicz at USACE for providing the necessary lake level data, and to three anonymous reviewers who provided valuable comments and suggestions. The spectral analyses were performed using MTM-SSA toolkit developed by the Theoretical Climate Dynamics group at the University of California-Los Angeles: http://www.atmos.ucla.edu/tcd/ssa/. This research was supported by the Office of Science (BER), U. S. Department of Energy (DOE) grant DE-FG02-07ER64428 (S. K.) and the University of Wisconsin-Milwaukee Research Growth Initiative (RGI) 01-01 (P. R.).