Controls on recent Alaskan lake changes identified from water isotopes and remote sensing


  • The copyright line for this article was changed on 23 MAR 2015 after online publication.


[1] High-latitude lakes are important for terrestrial carbon dynamics and waterfowl habitat driving a need to better understand controls on lake area changes. To identify the existence and cause of recent lake area changes in the Yukon Flats, a region of discontinuous permafrost in north central Alaska, we evaluate remotely sensed imagery with lake water isotope compositions and hydroclimatic parameters. Isotope compositions indicate that mixtures of precipitation, river water, and groundwater source ~95% of the studied lakes. The remaining minority are more dominantly sourced by snowmelt and/or permafrost thaw. Isotope-based water balance estimates indicate 58% of lakes lose more than half of inflow by evaporation. For 26% of the lakes studied, evaporative losses exceeded supply. Surface area trend analysis indicates that most lakes were near their maximum extent in the early 1980s during a relatively cool and wet period. Subsequent reductions can be explained by moisture deficits and greater evaporation.

1 Introduction

[2] The likelihood of future warming trends in Alaska has motivated efforts to observe and evaluate recent hydrologic change [Hinzman et al., 2005; Walvoord et al., 2012]. Theory and climate models suggest greater Arctic precipitation with a warming atmosphere that is supported to some extent by Eurasian river discharge since the 1940s [Peterson et al., 2002]. However, a significant degree of interannual discharge variability is also linked with regional atmospheric dynamics such as those characterized by the Northern Annual Mode and Pacific Decadal Oscillation and/or El Niño–Southern Oscillation in the North Pacific sector [Brabets and Walvoord, 2009; Cassano and Cassano, 2010]. Other surface water trends, such as lake volume and area, are currently ambiguous, with a diversity of variation observed in continuous and discontinuous permafrost regions [Riordan et al., 2006; Smith et al., 2005]. Lake surface observations are available in Alaska from early aerial photographic surveys beginning in the late 1950s and by satellite data since the late 1970s.

[3] Lake surface areas, and corresponding volume, reflect the balance between rates of surface inflow, groundwater discharge, and precipitation falling directly onto a lake, with rates of water loss from surface outflow, groundwater recharge, and evaporation/evapotranspiration from the lake surface and surrounding wetlands. At high latitudes, permafrost distribution may be the most important factor in determining lake density in addition to surface flow [Williams, 1970]. In areas of continuous permafrost, subpermafrost groundwater is often isolated from the surface, and there are unique mechanisms for thermokarst lake dynamics such as lateral expansion and breaching [Jones et al., 2011; Labrecque et al., 2009; Plug et al., 2008]. Groundwater-surface water interactions in the interior regions of Alaska are more commonly found in areas of discontinuous permafrost [e.g., Minsley et al., 2012; Walvoord et al., 2012].

[4] Here the focus is on a region of discontinuous permafrost in northeast Alaska known as the Yukon Flats (YF) that encompasses the ~118,340 km2 low-lying area surrounding the confluence of the Yukon and Porcupine rivers (Figure 1). YF lakes occupy depressions formed by fluvial, eolian, and thermokarst processes, have a wide diversity of sizes and shapes, and remain frozen for 7 to 8 months of each year. In general, shallower basins with more gradual slopes undergo larger changes in surface extent for incremental changes in volume than deeper counterparts with steep sides. This is likely one reason why YF lakes exhibit complex spatial patterns of change for different temporal scales [Roach et al., 2011; Rover et al., 2012].

Figure 1.

The surficial geology of the Yukon Flats National Wildlife Refuge from Williams [1962] is shown on a digital elevation model (30 m) with lake locations for this study, the location of evaporite deposits from Clautice and Mowatt [1981], and the 11.2 km2 areas delineated by Heglund and Jones [2003] within which 59 lakes were sampled in 2011 (individual lake locations not shown). The rectangular box encloses the area of the surface area trend analysis. Open circle symbols indicate lakes sourced by snow and/or permafrost.

[5] Lake water isotope ratios provide a useful measure for characterizing lake hydrology. Isotopes of oxygen and hydrogen in lake water are sensitive to hydrologic processes, and in particular, the preferential loss of light isotopes by evaporation. Here we also derive estimates of the ratio of water lost by evaporation to that gained by inflow (E/I) by using an isotope-based water balance model. The isotope labels are also used to identify the dominant sources for lakes such as mixtures of rainfall and snowfall, groundwater, rivers, or thawed permafrost [e.g., Wolfe et al., 2007]. These parameters are then used in conjunction with climatic data and remotely sensed imagery to identify the patterns and causes of recent lake area changes.

2 Study Area and Methods

[6] The YF is a region of extreme seasonal temperature difference and low precipitation (<250 mm/yr) [Shulski and Wendler, 2007]. Potential evapotranspiration (PET) estimates, following Hogg [1997], indicate annual moisture deficits near 15 cm/yr. Lakes are generally fresh to moderately brackish (<1600 μS) and predominantly eutrophic or hypereutrophic [Hawkins, 1995; Heglund and Jones, 2003]. Evaporite salt films occur on lake margins and dry lakebeds (Figure 1) and consist of trona, calcite, dolomite, gypsum, and halite [Clautice and Mowatt, 1981]. Vegetation is a mosaic of grassy meadows, muskeg, marsh, and forests of spruce and birch that are highly prone to fire [Drury and Grissom, 2008].

[7] Lake water isotope samples from 83 lakes were acquired in July, August, or September between 2007 and 2010 by fixed wing aircraft (Figure 1; supporting information text). An additional set of smaller lakes (n = 33) was sampled by helicopter in September 2009. In July 2011, 59 lakes were sampled on foot within five distinct 11.2 km2 areas that were previously defined and investigated for baseline limnology [Heglund and Jones, 2003]. River water data used here are from Schuster et al. [2010] collected during the months of June through October between 2006 and 2008 from the Yukon River at Circle and Fort Yukon, the Porcupine River above Fort Yukon, and the Chandalar River above Venetie. Water for isotope analysis was collected in 30 mL Nalgene high density polyethylene or glass bottles that were filled to minimize headspace, sealed to prevent evaporation and analyzed within 2 months of collection. Water samples were prepared for oxygen and hydrogen isotope ratio analyses by automated constant temperature equilibration with CO2, and automated D/H preparation by chromium reduction, coupled to an isotope ratio mass spectrometer and are reported in δ-notation relative to Vienna standard mean ocean water (VSMOW). Analytical precision is ± 0.08‰ and ± 0.9‰ for oxygen and hydrogen, respectively.

[8] Lake evaporation-to-inflow ratios (E/I) were calculated using the isotope mass balance method developed by Gibson and Edwards [2002] which assumes a well-mixed lake undergoing evaporation while maintaining a long-term constant volume and lake water residence time (see supporting information text for additional information). The method requires isotope values for inflow, lake water, and evaporated vapor and utilizes the assumption that the oxygen isotope ratios of outflow are equivalent to lake water values. The isotopic composition of evaporated vapor is difficult to measure but has been shown to be dependent on temperature, boundary layer state, relative humidity, and the isotopic composition of atmospheric moisture.

[9] Of the 175 lakes sampled for water isotopes, 54 of them were analyzed for surface area changes and trends following the approach described in Rover et al. [2012] with additional analysis for 2010 and 2011. There was classification of 22 Landsat Multispectral Scanner, Thematic Mapper, and Enhanced Thematic Mapper Plus scenes during the ice-free season since 1979. Following a decision tree approach, water and nonwater classes were identified, and a total of 16,371 water bodies in the study area had detectable water at more than one date and were not connected to perennial streams and rivers. The water images were converted to vector polygons for calculating area at each image date [Rover et al., 2011]. When clouds, cloud shadows, or snow obstructed underlying water bodies, the observation was assigned a “no data” value. Trend analyses are based on a linear regression method with date and water area as the predictor and response, respectively, and the slope representing change in water area per time. A two-tailed t test of the slope was used to evaluate results for p-values ≥0.01. Notably, some lakes with insignificant linear regressions are found to have highly variable surface area extents.

[10] Meteorological data since 1948 was obtained from the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis [Kalnay et al., 1996] for an average of grid cells that encompass the YF area (Figure 2). They are compared with the grid cell for Fairbanks, the location of the nearest first-order climate station 230 km to the south. The records show that following an arid period during the 1950s, higher relative humidity, greater snowfall, and lower potential evaporation characterize the period between ~1970 and 1995, a change that has been attributed to natural variability of the position and strength of the Aleutian Low [Hartmann and Wendler, 2005]. From the mid-1990s to the present day, temperatures have been relatively high, humidity levels low, snowfall low, and potential evaporation high.

Figure 2.

Hydroclimatic parameters for an average of YF grid cells since 1948 to 2012 derived from NCEP-NCAR reanalysis shown with the data for the Fairbanks grid cell. Temperature (°C), relative humidity (%), potential evaporation rate (W/m2), and snow water equivalent (kg/m2) are shown as anomalies from the 1948–2012 mean.

3 Results

[11] The oxygen and hydrogen isotope ratios of lakes range from −20‰ to −5‰ and −85‰ to −160‰, respectively (Figure 3 and Tables S1 and S2). Lakes are generally enriched in heavy isotopes with respect to meteoric waters as represented by the Global Meteoric Water Line (GMWL), which is an indication of kinetic fractionation effects during evaporation. Lakes are also enriched relative to the average value for Yukon River waters (−21‰ for oxygen and −167‰ for hydrogen), which as an integration of all regional waters is typically similar to the average isotopic values of precipitation and approximates the values for groundwater. The majority of lakes plot on an evaporation line with a slope of 5 that intersects the GMWL near the average Yukon River value. This intersection indicates that the source water for most lakes is a mixture of precipitation and groundwater in addition to river water for lakes with surface connections or within floodplains.

Figure 3.

Oxygen and hydrogen isotopes and calculated evaporation-to-inflow (E/I) ratios of water in the Yukon Flats: (a) Lake isotope ratios shown with the Global Meteoric Water Line (GMWL), and estimated monthly precipitation values (black) [Bowen and Revenaugh, 2003], and Yukon, Porcupine, and Chandalar River values (gray) [Schuster et al., 2010]. Dashed arcs indicate the approximate ranges of isotope values for rain, snow, rivers, groundwater, and permafrost. Blue circles indicate lakes sourced by combinations of precipitation, rivers, and groundwater, and red circles indicate those sourced by snowmelt and/or permafrost thaw. Stars indicate approximate source values for each group. (b) Colors similar to Figure 2a show the populations of lakes with E/I > 0.5 and 1, above which evaporation losses are greater than 50% and 100% of inflow, respectively.

[12] For a small group of lakes (26 of 175), plotting a line through their position to the average river water value was found to produce unreasonably low slopes (m ≈ 3) that are inconsistent with regionally established slopes of 4.5 to 5.5 [Gibson et al., 2002] and suggest alternative source waters. Therefore, we regressed an evaporation line with a slope of 5 from each of these lakes (see supporting information text for additional discussion) and found that they intersect the GMWL near −27‰ for oxygen and −200‰ for hydrogen, values that are within the range of observed values for snowmelt and permafrost thaw [Lachniet et al., 2012; Meyer et al., 2010]. This suggests that this group of lakes is sourced by snowmelt and permafrost thaw, most likely as talik growth. Reliance on snowmelt and/or permafrost thaw for annual recharge would be more likely for basins that are isolated from deep, subpermafrost groundwater and surface flow. E/I estimates range from 0.08 to 4.93 for the data set (Figure 3b) and are affected by the isotope composition of source waters.

[13] The oxygen and hydrogen isotope compositions, and E/I ratios of the 54 lakes analyzed for area changes span a similar range as the larger data set. Fifteen are dominantly sourced by snowmelt and/or thawed permafrost. Time series of surface areas indicate that more lakes were nearer their maximum extent and generally varied coherently between 1980 and 1990 (Figure 4a). Subsequently, particularly after the mid-1990s, more lakes decreased or varied more significantly. Lakes that underwent less evaporation (E/I < 0.5) tended to have varied less in area, with no overall tendency to increase or decrease (Figure 4b). In contrast, lakes that underwent more evaporation (E/I > 1) tended to have varied more and often in opposing directions; some lakes completely disappeared while others reached their maximum areas (Figure 4c). The lakes identified as sourced by snowmelt/permafrost tended to have varied more, occupied smaller depressions, and had higher evaporation losses (Figure S1). A decreasing trend is evident for several of them since the early 2000s (Figure 4d). A summary of area trends for the different ranges of E/I is shown in Table 1.

Figure 4.

Change in lake extent as a % of maximum from 1979 to 2011 for (a) all 54 lakes sampled within the trend analysis area, (b) lakes with E/I < 0.5 (n = 12), (c) lakes with E/I > 1 (n = 22), and (d) lakes sourced by snowmelt and/or thawed permafrost (n = 15). Image acquisition years are indicated as black triangles on the horizontal time scale. Gray lines indicate lakes with no surface area trends, red lines indicate lakes with decreasing trends, and a blue line for the one lake with an increasing trend.

Table 1. Lake Area Linear Regression Analyses (p < 0.01) for E/I and Source
 NumberIncreasingNo ChangeDecreasing
E/I < 0.5120120
0.5 < E/I < 1.0200155
E/I >1.0221129
Snowmelt/permafrost thaw15078

4 Discussion

[14] The isotope results indicate that an integration of rainfall, snowfall, river inflow and/or flooding, and groundwater, is the water source for most YF lakes. Therefore, we propose that lake water budgets and area/volume variations are dominantly controlled by hydroclimate, including amounts of precipitation gain, evaporation loss, water table heights, and groundwater flow rates. Comparison with the climate data support this notion as the lakes that have diminished have done so concurrently with lower snowfall, relative humidity, and higher temperatures since ~1995 (Figures 2 and 4). Less moisture on multiannual to decadal time scales likely leads to lower water table heights and reduced subsurface and surface inflows. This may explain the greater diversity of responses as lake levels lower because corresponding surface area depends on basin shape, and there is a great diversity of basin shapes. Regional water table heights and groundwater flow rates are not known within the YF, but airborne electromagnetic imaging has revealed surface-groundwater connections [Minsley et al., 2012]. Taliks beneath larger water bodies were found to extend below the maximum depth of permafrost, allowing surface-groundwater connections, whereas smaller, shallower lakes with smaller taliks were typically isolated.

[15] The lake water isotope results also indicate that snowmelt and/or permafrost thaw is a dominant water source for smaller, shallower lakes that have dramatically varied in extent, by either total water loss and subsequent gain, or steady diminishment (Figure 4d). These lakes were generally more evaporated (Figure 3b), indicating limited groundwater interaction that is consistent with a location within permafrost or other aquatards. Subsurface drainage of lakes due to thawing permafrost and talik growth has been proposed as a mechanism for lake area reductions in other regions of Alaska [Yoshikawa and Hinzman, 2003]. Here, however, the isotope data suggest that lakes that could be within permafrost undergo significant water losses by evaporation. The geophysical evidence within the YF suggests complex groundwater flow paths and surface water connections such that subsurface conduits developed by permafrost thaw seem as likely to provide groundwater to a lake basin as drain it.

[16] There is no known evidence for thermokarst drainage in the YF similar to that which occurs in continuous permafrost regions [Jones et al., 2011]. However, snowmelt can recharge small, isolated basins by flowing over a frozen, and/or through a thawed, active layer (i.e., suprapermafrost flow), but isotope labeling cannot distinguish snowmelt versus thawed permafrost in this region because their ranges of values overlap. Nevertheless, the isotope data are consistent with the idea of recent reductions in snowfall as a mechanism for net water loss in some YF lakes [Jepsen et al., 2012].

[17] The E/I estimates of the entire 175 lake data set indicate that ~58% of lakes have evaporation losses that exceed 50% of inflow, which is strong evidence for the importance of evaporation in regional YF lake water budgets. We acknowledge that E/I estimates are derived from water isotope data that integrates hydrologic processes for an unknown, and possibly varying, time period that precedes the sample date. This, and the fact that the calculations utilize mean climate state variables from reanalysis data, all introduce sources of uncertainty that we take into consideration for our conclusions here. However, the assumption of constant volume and residence time relative to the time incorporated by the isotope data appears to be generally valid for isolated lakes that were highly evaporated. Indeed, our observations indicate minimal isotopic variation within seasons or from year to year (Figures S2 and S3).

[18] These results lead to the conclusion that the majority of recent lake area reductions in the Yukon Flats can be explained as a response to a multidecadal climate trend toward greater moisture deficit since the mid-1990s. This was characterized by increasing temperatures but also by reduced snowfall and summer humidity. Such a hydroclimatic mechanism may also be a dominant driver for other interior regions of Alaska and high-latitude regions. We cannot definitively rule out individual lake reductions related to permafrost thaw, but the isotope survey indicates that this mechanism is a potential influence on a relatively small number of YF lakes (26 of 175). An important implication of these results is that future surface water variations are likely to remain a function of decadal-scale regional atmosphere circulation variations that control storm tracks and the moisture balance in the Alaskan interior even as warmer global temperatures become more likely. Decadal-scale climate variation in Alaska is documented for the past 1000 years and the Holocene from paleohydroclimatic data [Anderson et al., 2007; Barber et al., 2004]. Such atmospheric dynamics are currently beyond the simulation ability of climate models, but the data in this study highlight their primary importance.

[19] The data also reveal remarkable hydrologic diversity at the watershed scale. It is notable that the range of isotope and E/I values within very small <10 km2 areas is nearly equivalent to the range of values observed for the entire ~118,340 km2 area (Figure S5). Similar hydrologic diversity at smaller spatial scales has been observed in other lake-rich landscapes [Euliss et al., 2004; Turner et al., 2010] and poses significant challenges for accurately “scaling-up” sparsely sampled areas to draw conclusions that span beyond local to regional controls. These findings indicate that attempts to project future high-latitude lake change will benefit from considering the effects of decadal-scale hydroclimatic variations. Furthermore, isotope ratios of lake water strengthen the basis upon which the vulnerability of individual water bodies and lake regions can be assessed.


[20] The USGS Climate and Land Use Change R&D Program and the USFWS Yukon Flats National Wildlife Refuge supported this research. We thank Tyler Lewis for providing 2011 water samples. Chris Eastoe, University of Arizona Environmental Isotope Laboratory, provided isotope data; Paco VanSistine provided GIS support. We thank John Gibson for early insights into the data and Ted Hogg for CMI and P-PET estimates. We appreciate manuscript reviews by Tyler Lewis, Brent Wolfe, and anonymous reviewers that helped to improve the manuscript. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

[21] The Editor thanks Brent Wolfe and an anonymous reviewer for their assistance in evaluating this paper.