Skip main navigation

NOAA logo


  GLERL logo
Skip About subnavigation bar


Project Overview


Hydrological Models





Great Lakes Sensitivity to Climatic Forcing

Hydrological Models

Introduction | Study Area | Methodology | Changed Climate | Sensitivities | Future | References

This section of the website is dedicated to understanding GLERL's role in this project. More information about the development of this hypothesis and the design of the overall project can be found in the Project Overview and Background sections of this website.


About 9,500 radiocarbon (14C) years before present (BP) the large upstream glacial Lake Agassiz in northwestern Ontario and Manitoba supplied melt water from the last glaciation to the upper Great Lakes through outlets to the Lake Nipigon basin and thence to the Superior and Huron-Michigan basins (Teller 1985, 1987). About 8,000 14C years BP, Lake Agassiz merged with glacial Lake Ojibway in northern Ontario and northeastern Quebec, and drained into the Ottawa River valley and thence to the North Atlantic Ocean via the St. Lawrence River valley (Teller and Leverington 2004), thereby bypassing the Great Lakes basin (Fig. 1).

paleogeography upper great lakes

Figure 1: Paleogeography of the region north of the upper Great Lakes at 7,900 14C BP when the combined outflow of Lakes Agassiz and Ojibway was routed to the Ottawa and St. Lawrence river valleys (solid arrow). At this stage the water supply of the Great Lakes was no longer supplemented by inflow from upstream sources, but was supplied by precipitation alone, as at present. Prior to 8,000 14C BP, overflow from Lake Agassiz passed into the Lake Nipigon and Lake Superior basins (open arrow). The upper Great Lakes then overflowed the North Bay outlet (open arrow) into the Mattawa River (M) and thence to the Ottawa and St Lawrence river valleys. Adapted from Figure 4p in Teller and Leverington (2004).

Recent construction of an empirical exponential model of isostatic uplift for the Great Lakes region following the last glaciation allowed comparison of the elevations of rebounding lake outlets with reconstructed lake levels based on 14C-dated water level indicators such as abandoned shorelines, isolation basins, submerged tree stumps, and unconformities (Lewis et al. in press a, b). The early Holocene results for the Huron, Georgian Bay, and Michigan basins reveal several periods of low lake levels (lower than present) due to overflow drainage through the isostatically depressed outlet area near North Bay, Ontario, to the Ottawa and St. Lawrence valleys; (Fig. 1). These results were anticipated on the basis of previous syntheses (Eschman and Karrow 1985; Hansel et al. 1985; Lewis and Anderson 1989; Barnett 1992; Clark et al. 1994; Colman et al. 1994a, b; Lewis et al. 1994; Rea et al. 1994; Moore et al. 1994, 2000; and Larson and Schaetzl 2001). Surprisingly, however, the results showed that lake levels fell below the Huron-Michigan basin outlet after 8,000 14C years BP, possibly for a few centuries during which lowest water levels were up to 30 m below the overflow sill at North Bay (Fig 2).

uncalibrated radioactivity

Figure 2: Huron basin lake level (black line) between 10,000 and 7,000 cal BP based on the original elevations of lake-level indicators computed by an empirical exponential model of glacial rebound for the Great Lakes basin that removes the effects of glacial-isostatic uplift. For clarity, data are removed from the plot except for the Stanley unconformity constraint (S). At about 7,900 14C BP (about 8800 cal BP) the lake level, indicated by the Stanley unconformity (S–line indicates original elevation and circles bracket its age of 7900±300 14C BP), descended tens of meters below the North Bay outlet (grey band), the lowest possible point of overflow for the Huron basin at the time. The thickness of the grey band indicates the depth of water over the outlet sill at full discharge. Adapted from Lewis et al. (in press a, b).

The inferred 7,900 14C years BP low stand of the Michigan and Huron basins occurred after melt water drainage from upstream glacial Lake Agassiz began to bypass the Great Lakes basin, leaving it susceptible to the early dry Holocene climate (Edwards et al. 1996). As the Great Lakes basin had then entered its present hydrological regime of water supply by precipitation only, and as differential glacial-isostatic crustal uplift was accounted for, the only known process that could explain the sub-outlet low levels is climatic reduction in water supply, either by enhanced evaporation or reduced precipitation or both. This episode of lowest levels appears as an extraordinarily severe impact of a dry climate, possibly of short duration, on the upper Great Lakes hydrological system and may have extended to the lower lakes.

Preliminary study of the camoebians and pollen in the sediment sequence of Georgian Bay suggests increased lake water salinity and reduced precipitation at the time of the closed low stands (Sarvis 2000; Blasco 2001), conditions that are consistent with reduced water supply. In Hamilton Harbor, western Lake Ontario, studies of ostracodes (Delorme 1996), diatoms, pollen and isotopes (Duthie et al. 1996) all reveal a low water phase and suggest a drier climate about 7,500 14C years BP. Similarly, in Mud Lake on the Keweenaw Peninsula beside Lake Superior, pollen and plant macrofossil analyses indicate an onset of a drying episode at 7,900 14C years BP with notably reduced water levels extending to after 7,000 14C years BP (Booth et al. 2002). As these widely-separated sites both indicate a lowering of lake levels and onset of a drier climatic episode, it is reasonable to envision that the inferred dry climate conditions that impacted the Huron and Michigan basins also affected other basins of the Great Lakes system. The low stand period correlates also to a relatively rapid transition in vegetative cover in the Great Lakes region from boreal to the mixed Great Lakes-St. Lawrence Forest as recorded in pollen diagrams (McAndrews 1994; Dyke et al. 2004). Complementary review and assessment of the available paleoclimate information for the Great Lakes watershed, coupled with new studies of proxy climatic and limnological indicators (pollen, isotopes, diatoms, ostracodes and thecamoebians) focused on the period spanning the closed low stand are in progress, and will be reported in future publications.

For this re-assessment, it would be helpful to obtain approximate information about the amplitude of climatic change that might be expected to have caused the upper Great Lakes low stands. Accordingly, we have used a hydrological model to explore the excursion from the present climate that would force the Great Lakes into hydrologic closure in terms of increased temperature and reduction in precipitation. In this study, the hydrology of the Erie and Ontario basins is considered separately from that of the upper Great Lakes, as the isostatically-depressed North Bay outlet for the upper lakes remained at a lower elevation than the St. Clair River connection to Lake St. Clair and Lake Erie until much later, about 5,500 14C years BP (Eschman and Karrow 1985). Other attributes of the region such as land cover, geography, and bathymetry were modeled as they are at present.

This low stand episode offers an opportunity, once paleoclimate is better quantified, to acquire information about the hydrological sensitivity of the Great Lakes system to high-amplitude climate change. Such information would be beneficial for model studies and projections of future levels of the Great Lakes under global warming in which some climate modeling scenarios project levels below instrumentally-observed ‘natural variability’ (Mortsch and Quinn 1996; Mortsch et al. 2000; Lofgren et al. 2002). It should be noted that the 7,900 14C years BP low stand episode occurred while the residual Laurentide ice sheet, then in the latitudes of Hudson Bay, was rapidly retreating and wasting away (Dyke et al. 2003). This was a period of rapid change in the proportions of land, ice, and water areas, with parallel changes in albedo and reorganization of atmospheric circulation (Dean et al. 2002). As a result, the inferred occurrence of closed low stand conditions in the Great Lakes basin is seen as a product of extremely unusual conditions. It is not regarded as an analog for future conditions, but rather, as a natural experiment from which important information about lake-climate sensitivity might be derived.

The purpose of this paper is to demonstrate that if a climate is extreme enough, levels on some Great Lakes would drop sufficiently to cut off outflow, thereby making those lakes terminal. We look at excursions in temperature and precipitation from the present climate to disclose those values that would drive the Great Lakes hydrology to produce terminal lakes. This is not an attempt to simulate past hydrology exactly, but to explore the possible magnitude of changed climates that might have produced terminal lakes about 7,900 14C years ago in accordance with recently acquired glacial-isostatic rebound evidence that Huron and Michigan basin lake levels had descended below their overflow outlets.

Study Area

The Great Lakes basin area is 770,000 km 2 (300,000 mi 2), about one-third of which is water surface
(Fig. 3) The basin extends 3,200 km (2,000 mi) from the western edge of Lake Superior to the St. Lawrence Power Project, Cornwall, Ontario on the St. Lawrence River. The water surface drops in a cascade over this distance some 180 m (600 ft). Lake Superior is largest and deepest and has two diversions into it: the Long Lac and Ogoki. Lake Superior flows through the lock and compensating works at Sault Ste. Marie and down the St. Mary’s River into Lake Huron where it is joined by water flowing from Lake Michigan. Lake Superior outflows and levels are regulated to balance Lakes Superior, Michigan, and Huron water levels, according to Regulation Plan 1977, under the auspices of the International Joint Commission.

location map

Figure 3: Laurentian Great Lakes location map.

Lakes Michigan and Huron are considered to be one lake hydraulically because of their connection through the deep Straits of Mackinac. A relatively small flow of Lake Michigan water is diverted into the Mississippi River basin at Chicago. The water flows from Lake Huron through the St. Clair River, Lake St. Clair, and Detroit River system into Lake Erie. The drop in water surface between Lakes Michigan-Huron and Lake Erie is only about 2 m (8 ft). This results in a large backwater effect between Lakes Erie, St. Clair, and Michigan-Huron; changes in Lakes St. Clair and Erie levels are transmitted upstream to Lake Michigan-Huron.

From Lake Erie, the flow is through the Niagara River and Welland Diversion (used for navigation and hydropower) into Lake Ontario. There is also a small diversion into the New York State Barge Canal System which is ultimately discharged into Lake Ontario. Lake Ontario outflows and levels are regulated in accordance with Regulation Plan 1958D to balance interests upstream on Lake Ontario with those downstream on the St. Lawrence Seaway. The outflows are controlled by the Moses-Saunders Power Dam between Massena, New York and Cornwall, Ontario. From Lake Ontario, the water flows through the St. Lawrence River to the Gulf of St. Lawrence and to the Atlantic Ocean. Lakes Superior, Michigan, Huron, and Ontario are very deep (229—405 m) while Lakes Erie and St. Clair are very shallow (6—64 m).

GLERL is involved in modeling Great Lakes hydrology under extreme climates (such as those that may have been available in the paleo-climate) using the Great Lakes Advanced Hydrological Prediction System (GLAHPS), which is a system of hydrologic process models used to estimate water and energy balances, whole-lake heat storage, and lake levels. These include models for rainfall-runoff, evapotranspiration, and basin moisture storage , overlake precipitation, one-dimensional (depth) lake thermodynamics for lake surface flux, thermal structure, evaporation, and heat storage, net lake supplies, and channel routing .These models have traditionally been used to study changed climates resulting from increased CO2 emissions by adapting the historical meteorological record consistent with general circulation model (GCM) simulations. Figure 4 depicts the arrangement of processes in the water balance that these models are applied to.

great lakes channels schematic

Figure 4: Arrangement schematic of Great Lakes Connecting channels, and all water flows


For the purpose of this study, we adjusted the present GLAHPS model to simulate the Great Lakes in their pre-European-settlement natural state by removing the influences of channel control works and regulation plans. Also, we kept watersheds of the upper and lower lakes separate, as they were during the early Holocene, i.e. no outflow from the Huron basin to the St. Clair-Erie basin. Accordingly, we use the models here with water balances on all lakes, and with lake outflow rating curves, selected to represent “natural” or “pre-development” conditions. We account for lake area variations with changes in water level, but do not remove present-day diversions and consumptions as they are relatively insignificant for our purpose.

First we consider all lakes as interdependent (as they are now, but with “natural” outlet and connecting channel flows) to see if simulations with historical meteorology (1948-1999) produce hydrology and lake levels comparable with the historical records. This allows us to assess the reasonableness of the modified models. Then, we model two systems of Great Lakes independently:

1) Lakes Superior, Michigan, and Huron (the upper Great Lakes), and

2) Lakes St. Clair, Erie, and Ontario (the lower Great Lakes) with no inflow from the upper Great Lakes since they drained via the North Bay outlet to the Mattawa and Ottawa rivers when overflow occurred prior to and after the low stand 7,900 14C years ago.

Next, we consider steady state hydrology by modeling over an extended period constructed by repeating the adjusted meteorological record until consecutive 52-year segments are identical. We finally consider each lake as part of its parent system (upper or lower system) with a water balance on all lakes.

Changed-Climate Methodology

The hydrology models here use daily meteorological data from 1948—1999, compiled from about 1,800 stations for over-land meteorology ( precipitation and air temperature) and about 40 stations for over-lake meteorology ( air temperature, humidity, wind speed, and cloud cover, see Fig 5)


Meteorological station network

Figure 5: Great Lakes Meteorological Station Networks.

These data, compiled for previous studies (Croley 1990; Hartmann 1990; Croley 1992b; Croley et al. 1998; Lofgren et al. 2002; Croley 2003 ), provide daily meteorological time series over each of the 121 riverine watersheds that drain into the Great Lakes and the 7 Great Lake water surfaces. Annual average precipitation and air temperature are, respectively, 80.9 cm and 2.93 oC (Superior basin) , 84.5 cm and 6.49 oC (Michigan-Huron) , 91.9 cm and 9.19 oC ( Erie), and 92.3 cm and 7.41oC ( Ontario).

We used these historical meteorological data with our hydrology models to compute the “present” or “base case” scenario (Fig 6). We then apply selected precipitation ratios and air temperature differences to the historical meteorological data and use these modified meteorological time series with our hydrology models to construct changed climate scenarios.

net basin supply

Figure 6: Net Basin Supply Comparisons

All precipitation is adjusted by multiplying the actual precipitation by a single precipitation ratio and all air temperatures are adjusted by adding a single temperature difference to the actual temperatures. In addition, humidity is adjusted; for precipitation ratios below unity, which are all that are considered here, the absolute humidity is multiplied by the ratio. Thus, if precipitation is halved, then so is humidity.

For more detailed information about the individual hydrology models used, aswell as information regarding the validation methodology used to check the accuracy of these modelsl, please download a copy of the Hydrology Models report here.

Changed Climates

Steady-State | Upper Great Lakes | Lower Great Lakes

Before applying the simulations to changed climates (i.e. changed temperature and precipitation), we ascertained that the present-day diversions in the hydrology models were on the order of a few centimeters; see Table 1 below. [Note that these diversions affect lakes upstream as well as downstream. The Chicago diversion affects Superior because resultant lower Michigan-Huron levels are used in regulation of Superior. The Welland diversion lowers Lake Erie and, because of connecting channel hydraulics (upstream and downstream lake levels determine channel flow), lower Erie levels lower Michigan-Huron and lower Michigan-Huron levels lower Superior as just discussed.] Thus, they are negligible compared to the changes in net basin supplies or drops in water levels to be simulated with changed climates. Therefore, we ignore them; no effort was made to remove these diversions from the existing models.

table 1

Table 1. Summary of Average Great Lakes Diversion Impacts (IJC 1985).

The changed climate scenarios used in this study were simple: spatially and temporally constant adjustments were applied to historical meteorology for each watershed and lake surface to estimate changed-climate meteorology for each watershed and lake surface. More complex climate change considerations in the study of terminal Great Lakes wait on improved paleoclimatic considerations. Our results are biased by the length of the historical meteorology record we used. Errors of approximation include linear adjustment of supplies for lake area, power equation hypsometric relations, and approximation of natural flow conditions and sill elevations for each Great Lake.

Steady-State Simulation

We use both the historical and modified meteorological time series with our models to simulate base case and climate change hydrology scenarios, respectively. We estimate steady state hydrology by modeling with arbitrary initial conditions (snow pack, water storages in the basins, thermal structure of the lakes, lake levels, and so forth) over an extended period constructed by repeating the adjusted meteorological record until consecutive 52-year segments are identical. (The models always converge no matter where started). The number of iterations required to reach this state depends largely on the arbitrary lake level assumed at the beginning and the final lake levels; it sometimes represents a longer time than might be expected for the climate change itself. (The effect of the initial conditions other than lake levels is much shorter, usually on the order of a couple of years.) However, since lake levels are unknown prior to the changing climate and since we want to avoid representing climate change as abrupt, we use this “steady-state” behavior to assess the effects of climate change.

Upper Great Lakes

Separating upper lakes (Superior, Michigan, Huron) from lower lakes (St. Clair, Erie, Ontario), for purposes of simulating the system about 7,900 14C years BP, is accomplished by changing Lake Michigan-Huron outflow to a function only of the water level in Lake Michigan-Huron (and not of St. Clair) and then by modeling only these upper lakes.

When lake levels are always below the sill elevation, then the lake is terminal. We looked at 36 climate scenarios, each defined in terms of the precipitation drop from the base case (0—50% in steps of 10%) and the temperature rise above the base case (0—5 oC in steps of 1 oC). We calculated the steady-state average water level resulting from each and plotted it with precipitation drop and temperature rise as shown in Figure 7 for Lakes Superior and Michigan-Huron. Note three regions in each graph of Figure 7: the region where all water levels are above the sill elevation in the lower left of the graphs, the region where all water levels are below the sill elevation in the upper right of the graphs, and the intermediate region where water levels are both above and below the sill elevation. We determined the boundaries of these regions by looking at maximum and minimum water levels in each simulation and comparing them to the sill elevations. Since behavior of steady-state average water levels is fundamentally different in each of these regions, we restricted linear interpolation in each region to only values therein.

By using linear approximations, note that the climate isolines for a terminal Lake Superior in Figure 7 drop about 1 oC for every 4.7% change in precipitation. Figure 7 suggests that Lake Superior should be a terminal lake for climates with a temperature rise T ( oC) and a precipitation drop P (%) such that 4.7T + P > 60.

Likewise, the isolines in Figure 7 for a terminal Lake Michigan-Huron drop about 1 oC for every 4.5% change in precipitation; Lake Michigan-Huron should be terminal for climates with a temperature rise T ( oC) and a precipitation drop P (%) such that 4.5T + P > 63.

upper great lakes

Figure 7: Steady-state upper Great Lakes average water levels as a function of temperature rise and precipitation drop relative to the present base climate.

Lower Great Lakes

For the lower Great Lakes, we looked at Lakes St. Clair, Erie, and Ontario, with no inflow to Lake St. Clair as was the case prior to and after the low stand 7,900 years ago when Michigan-Huron flowed into the Mattawa and Ottawa watersheds; see . Since the St. Clair lake bottom is above its sill elevation, there can be flow into Erie even when Lake St. Clair is empty. Thus, St. Clair can never be terminal (with water still in it); it can only dry up. We have to consider Lakes St. Clair and Erie as one water body to investigate Lake Erie becoming terminal. We looked again at the 36 climate scenarios, previously defined, and calculated the steady-state water levels resulting from each. We found that Lake Erie became terminal in this range of climate variations but Lake Ontario did not. Therefore, we considered a larger range of climate variations by taking nine precipitation ratios (0—80% in steps of 10%) and eleven temperature rises (0—10 oC in steps of 1 oC) and plotted the average steady-state water level resulting from each in Figure 5 for Lakes Erie and Ontario.

Note that we again define three regions in each graph of Figure 8 for water levels above the sill, below the sill, and both above and below the sill, by looking at maximum and minimum water levels and sill elevations. We again restrict linear interpolation in each region to only values in that region. Note that the isolines for a terminal Lake Erie in Figure 8 drop about 1 oC for every 4.7% change in precipitation. Figure 8 suggests that Lake Erie should be a terminal lake for climates with a temperature rise T ( oC) and a precipitation drop P (%) such that 4.7T + P > 51. Likewise, the isolines in Figure 8 for a terminal Lake Ontario drop about 1 oC for every 3.5% change in precipitation; Lake Ontario should be terminal for climates with a temperature rise T ( oC) and a precipitation drop P (%) such that 3.5T + P > 71.

lower great lakes

Figure 8: Steady state lower Great Lakes average water levels as a function of temperature rise and precipitation drop relative to the present base climate.


Each climate considered herein is specified over the entire upper Great Lakes basin or over the entire lower Great Lakes basin in their respective analyses. That is, the same changes made to historical data, to construct a hypothetical climate, were used across all water bodies and their basins in each analysis. For example, the 1 oC increase applied to Lake Ontario meteorological data was applied at the same time to the Lake Erie meteorological data in the analyses. Thus, no consideration is made of more complex changed climates (such as a 1 oC change in Lake Erie air temperatures with a 2 oC change in Lake Ontario air temperatures). Given this limitation, the order of the lakes going terminal as climate gets warmer and drier is approximately: Erie, Superior, Michigan-Huron, and Ontario. (The order varies a little depending on the path of the changes from the present climate taken in Figures 7 and 8). For both Great Lake subsystems (upper and lower), the uppermost lake goes terminal before the lowermost lake; this is not strictly necessary. There may be climates (where meteorological conditions are different over the uppermost lake and the lowermost lake) that would yield the lowermost lake terminal while the uppermost lake was not terminal. However, those changed climates were not investigated herein. As more is learned about past climates from paleoclimatic considerations, we can fine tune the observations made herein.

Likewise, the climate changes considered herein were simplified. We multiplied all historical daily precipitation amounts, without regard to season of the year, by a constant ratio and we added to all historical daily air temperatures, again without regard to season of the year, a constant value. Undoubtedly, we could consider more reasonable changes by considering the season of the year, and even location. Again, as more is learned about past climates from paleoclimatic considerations, we can make these additional considerations. However, we think these results are generally indicative of how climate effects would influence Great Lakes terminal lake status Indeed, Lofgren et al. (2002) summarized many of the past Great Lake climate studies that used general circulation model experiments for 2 CO2 studies; those climates that were warmer and drier showed good agreement with Figure 7.

Since we used only the available 52 years of daily meteorological data, continuously repeated, to represent steady-state, we biased our results somewhat; only the storm events on record are represented. The “transitional zone” in both Figures 7 and 8 might be wider if a longer period were used since more marginal storm events would be included that allowed some small outflow at water levels close to sill elevations.

There are also many errors of approximation in this study; our calculations used over-lake precipitation, over-lake evaporation, and runoff to the lake from models that assumed fixed values (coordinated between the U.S. and Canada) for lake areas and volumes, and then adjusted them for the actual lake and basin areas obtained in a comprehensive water balance. Better consideration would modify the runoff and lake evaporation models directly to consider the actual lake areas and volumes in an integrated water balance that employs these models directly. Likewise, the hypsometric relations and outflow relations (both rating coefficients and sill elevations) could be improved. Different sill elevations would shift the “terminal” lines in Figures 7 and 8.

Finally, the results do not exactly represent past hydrology (for example, paleo-lake areas have not been incorporated) so that results should be interpreted as exploration on the question of what could be reasonably envisioned as the effect of various climate scenarios on the hydrology of the pre-development Great Lakes. This is an attempt to study the question of “What magnitude of drying and warming of the present climate might produce terminal lakes as a guide to possible climate that apparently produced hydrologic closure of at least some of the Great Lakes about 7,900 C years BP?”

Future Endeavors

It is possible that other aspects of climate may have been a factor in lowering lake levels about 7,900 14C years BP. In a study of the oxygen isotopic composition of inorganic carbonate, cellulose from fossil wood, and lake sediments in southern Ontario, Edwards et al. (1996) show significant increases of mean annual temperature and summer relative humidity from 8,000 to 7,000 14C years BP. The early part of this interval is characterized as cold and dry; only the latter part is interpreted as warm and dry relative to the present climate. West of Lake Superior, a study of Lake Ann sediments adjacent to a Holocene sand dune field showed that dune (wind) activity started about 8,000 14C years BP and was associated with relatively severe drought conditions (Keen and Shane 1990).

Thus future paleohydrological modeling focused on the 7,900 14C years BP Great Lakes low stand may need to explore potential impacts under cold/dry and windy conditions. Likewise, insolation would have been different at the time of the low stands than at present.

References Cited

Last updated: 2006-08-08 ks