Shrinking lakes of the Arctic: Spatial relationships and trajectory of change



[1] Over the past 3 decades the Arctic has seen substantial warming. Previous local to regional scale studies have shown a considerable reduction in the size of lakes in this region. The subsequent exposure of carbon- and methane-rich sediments and the direct changes in surface albedo feed back into the drivers of regional and global climate change. Understanding and quantifying changes in the Arctic is a critical component of climate modeling due to the cooling effect of the Arctic on the global climate. The current work utilizes global satellite data from the Moderate Resolution Imaging Spectro-radiometer (MODIS) instrument to investigate changes in lakes across Canada between 2000 and 2009. The results show a net reduction of more than 6,700 km2 in the surface area of water in lakes across Canada. Modest gains in the southern regions are offset by larger losses in surface area farther north. Additionally, spatial analysis shows that the lakes showing change are clustered in groups. This suggests that local variability may play a role in the observed changes. Further work is needed to extend the analysis to the circumpolar Arctic.

1. Introduction

[2] The high northern latitudes are known to contain substantial stores of carbon [Post et al., 1982; Billings, 1987; Gorham, 1991] with much of this stored in sediments in lakes and bogs [Hinkel et al., 2001; Camill, 2005]. Exposing highly organic sediments to the atmosphere can result in a switch from anaerobic activity to aerobic activity changing the net fluxes of carbon from negative (storage) to positive (release). However, the impact of changes in lakes on carbon is complex. In well-drained areas this can lead to the release of more CO2 simultaneously with a decrease in methane production. In areas of poor drainage standing water sustains oxygen-deprived conditions leading to conversion of plant detritus to methane [Stokstad, 2004]. As water tables fall CO2 emissions may be expected to increase substantially while those of methane decline sharply [Gorham, 1991]. Additional carbon transfer from rich organic lake sediments occurs through aquatic food chains in freshwater habitats [Schell, 1983]. For example, in Finland lake sediments contain 0.6 Pg C, comparable to the amount of forest biomass [Kortelainen et al., 2004]. Molot and Dillon [1996] using Canadian data, estimate much larger storage in lakes. Cole et al. [2007] show that a major unknown is the potential significance of small lakes and streams in degassing CO2.

[3] A growing number of studies present evidence of substantial changes in fresh-water ponds and lakes in high northern latitudes globally. Hinkel et al. [2007] used Landsat data to show that 50 lakes drained completely or partially over a 25 year period in the western Arctic coastal plain of Northern Alaska. Riordan et al. [2006] observed substantial changes in lake area using air photographs and satellite data for nine small sample areas throughout Alaska. Since the 1950s surface water area has declined by 31% to 4% of the total non-oceanic area in the study region. Smith et al. [2005] reported large reductions in lake abundance and area in a survey of 10,000 lakes in Central Siberia, where the total number of lakes greater than 40 ha declined by 11%. Others have also reported declines in both the total number of lakes and in the surface area of water in the remaining lakes in small case studies in the North American Arctic including Stow et al. [2004], Yoshikawa and Hinzman [2003] and Klein et al. [2005]. The changes are contrary to some earlier expectations that lakes would increase associated with warming and permafrost melting [e.g., Gorham, 1991]. Relatively small local increases in abundance and area of lakes have been observed, including locations close to where lakes have declined in area [Riordan et al., 2006; Hinkel et al., 2007]. These changes reported in Alaska and Siberia are found in areas largely underlain by permafrost soils. However, Smol and Douglas [2007] have found that lakes and ponds are also drying in areas that are underlain by permafrost bedrock on Ellesmere Island of Canada.

[4] The location, extent, and dynamics of Arctic lakes and ponds are not well-documented. Many of these water bodies are small - 1 to 10 ha in size - and historically, they have not been well mapped due to limited accessibility and incomplete or unavailable remotely sensed data [Downing et al., 2006; Grosse et al., 2008]. Many popular databases of lakes and inland water bodies are incomplete, outdated, or limited to coarse spatial resolution. Downing et al. [2006] expose the deficiencies in the representation of smaller water bodies in the Global Lakes and Wetlands Database (GLWD) of Lehner and Döll [2004]. For a complete description of available inland water datasets see Carroll et al. [2009].

[5] Recent advances in multi-sensor data fusion methodologies led to development of a comprehensive global map of surface water from the Shuttle RADAR Topography Mission (SRTM) combined with the daily 250 m MODIS observations [Carroll et al., 2009]. Maps produced from finer resolution satellite data such as Landsat (30 m spatial resolution) can produce more accurate local scale information, however Landsat does not provide sufficient temporal resolution to capture the dynamics of the lakes in the Arctic [Grosse et al., 2008]. In addition, at present there is no global or pan-Arctic dataset that depicts water bodies at 30 m. The MODIS methodology detects water bodies that are as small as 2–3 hectares in size and offers unprecedented temporal frequency of observations allowing for dynamic observations of inter-annual and seasonal variability of surface water extent. When compared to Landsat data the new global 250 m water map correctly maps 80% of the total lakes and over 90% of the surface area of water [Carroll et al., 2009]. The lakes that were missed were either smaller than 3 hectares or were particularly long and narrow; in either case the lakes were undetectable at 250 m spatial resolution.

[6] The primary objective of this research is to demonstrate the capabilities of the new MODIS-based daily water detection data product at 250 m spatial resolution to quantify the change in Arctic lakes. This research utilizes a remotely sensed dataset to analyze the spatial and temporal dynamics of lakes and ponds in the Canadian Arctic. In this project we: 1) quantify the extent of surface water contained in lakes, ponds, reservoirs and open water wetlands, 2) provide a spatially explicit quantitative assessment of changes in their surface area between 2000 and 2009.

2. Study Area

[7] The North American Arctic is characterized by a mosaic of lakes and ponds. In the region between 50°N and 70°N there are over 1.2 million water bodies greater than 3 hectares in size as shown in the global 250 m land water map [Carroll et al., 2009]. The study area was restricted to the extent of Canada as it is representative of the conditions we are investigating. This restriction resulted in a total lake count of ∼0.8 million lakes. The final study area is shown in Figure 1.

Figure 1.

Study area is Canada from 50 to 70 degrees north latitude. Large lakes can be seen in shades of blue with small lakes shown in lighter blue colors.

3. Methods

[8] Daily observations from the MODIS instrument have been classified, using a decision tree classifier, as water or land to produce a static global 250 m resolution water map [Carroll et al., 2009]. While the major advantage of using MODIS data lies in the ability to observe changes across very large regions, these data can also be used to resolve changes in the extent of specific lakes over varying time steps (Figure 2). For this study the MODIS water data have been assembled into annual representations of extent of water bodies that were covered with water for a minimum of 6 days during the study period. The minimum of 6 days was selected to reduce commission errors from cloud shadows and short duration flooding. All data were acquired during the summer months (June, July, August) for each year corresponding to snow and ice-free period in the study area. The data have been analyzed to produce an annual map of change in surface water extent for each year from 2000 to 2009 and used as an input to the current analysis of the spatial dimensions of change in Arctic lakes. The annual maps have been compared to the validated static MODIS water map [Carroll et al., 2009] and investigated for major discrepancies. Additional accuracy assessment of dynamic changes in lake extent will further increase the confidence in the reported observations. The study period of 2000–2009 was selected as this was the maximum extent of data at the time of the analysis.

Figure 2.

Time series of the surface water extent (black) of Hay Lake in Alberta, Canada from 2000–2009. Time series was assembled from daily observations of water during the summer months to provide a maximum extent of water area.

[9] Annual and decadal summary maps of changes in lake surface extent were derived from the MODIS surface water data set. The input raster files were imported to a Geographic Information System (GIS) and converted to vector format to identify contiguous individual water bodies as discrete objects, i.e., lakes. We calculated surface area of each lake and assessed the net change in area per lake as (Net Change = Gains(by area)-Losses(by area)). Subsequently, the lakes with changes were analyzed to look for clusters of lakes showing change. These annual net change maps represent the base product from which the spatial analysis was performed. For visualization purposes the results were averaged from native MODIS resolution (231.65 m) to 23 km and represented as a “change per 23 km grid cell”. This allows the user to see the changes on the regional scale that would otherwise be difficult to see.

[10] The analysis focused on two general assessment schemes. First, we evaluated the net change in lake-surface area over the entire study area irrespective of regional specifics. Second, we used a set of clustering statistical tests to determine what, if any, spatial relationship exists between individual lakes in Canada. The Getis and Ord Gi statistics (global and local) [Getis and Ord, 1992] were calculated to investigate if a set of points or polygons are clustered in a way that precludes random chance. The Global Gi statistic determines whether there are hot spots (clusters of high values) or cool spots (clusters of low values) in the data set. The Global Gi statistic was calculated for the dataset using Area as the parameter from a random sample of 10% of the total database selected through Hawth's Tools (H. L. Beyer, Hawth's Analysis Tools for ArcGIS, 2004, accessed May 2010, available at This resulted in a dataset of ∼76,000 lakes, and this dataset was then carried through to all future analyses. This implies that all statistical tests were applied to a random selection of 76,000 lakes irrespective of the category analyzed. The Global Gi statistic results in a Z-score which shows the confidence level with which the clusters have been determined. Features with a Z-score greater than the 95% confidence level are considered significant with a larger Z-score indicating stronger significance. The local Gi statistic shows the magnitude of the clustering and should only be run if the Global Gi showed significant clustering.

4. Results and Discussion

[11] There were approximately 1,300,000 individual polygons identified as distinguishable water bodies. The histogram of the data using the area of the lake, shown in Figure 3, shows that the data follow a Pareto distribution where small lakes account for the majority of the lakes in the landscape. This distribution was predicted by Downing et al. [2006] based on field observations and use of local data sets compared to global coarse resolution data sets. To focus our analysis on changes, we eliminated lakes which showed no “net” change (i.e., either there was no change at all or the gains offset the losses). A scatter plot of the 760,000 remaining lakes showing area of lake on the x axis and area of change on the y axis (Figure 4) shows the greatest variability in net change occurs with the smaller lakes less than 300 km2 in size and that the majority of lakes showed a net decrease in area between 2000 and 2009. The overall amount of change between 2000 and 2009 is 22,010 km2 surface area gain and 28,697 km2 loss for an overall net loss in the region of nearly 6,700 km2. The areas of loss are found primarily in the north and east while the gains are concentrated predominantly in the south and west (Figure 5). There are 3 areas labeled in Figure 5 as areas of interest. One of the two largest areas of gain (label 1 in Figure 5) is located in Alberta around Hay Lake which experienced a massive expansion in 2007 which has yet to subside (Figure 2). The second area of large gain (label 2 in Figure 5) is in Quebec and represents the La Grande reservoir complex which began filling in 2003. These two features represent over 10% of the total amount of gains shown during the study period. The third area is in the province of Nunavut (label 3 in Figure 5) and shows the substantial area of net loss. These decreases could be an indication that higher temperatures over the last several decades have resulted in a greater number of ice free days and increased evaporation resulting in the decrease in lake size over time [Klein et al., 2005; McGuire et al., 2007].

Figure 3.

Histogram showing the size distribution of the lakes in Canada by area in km2 as detected by the global 250 m land water map. Both the count of the number of lakes and the area have been put on a log scale to maximize the visibility of the bins in the graph.

Figure 4.

Scatterplot of the area of lake (x axis) against the area of net change (y axis) showing the dynamics of change in lakes with areas smaller than 300 km2.

Figure 5.

Net change in surface water per 23 km grid cell for Canada (50N to 70N) from 2000 to 2009. Area labeled 1 shows net gain for Hay Lake (as shown in Figure 2), Area 2 shows net gain for man-made reservoir (LaGrande reservoir), Area 3 shows large areas of net loss in Nunavut province in Canada.

[12] The Global Gi test determined that lakes in general are shown to cluster only slightly with a Z-score 1.71. This indicates that there is a slight amount of clustering among all lakes in Canada; however, since the generic lake clustering is low, if established, the clustering of lakes showing change will not be simply related to the distribution of the lakes in the region. The Global Gi test performed on the random sample of “net-loss” lakes established a strong likelihood that these lakes were clustered (Z-score = 36.83). Similarly, the Global Gi test for “net-gain” lakes showed a strong likelihood for clustering as well (Z-score = 34.01). In each case there was less than a 1% chance that the clustering was due to random chance. The results of the local Gi analysis confirmed that lakes with large areas of net gains and net losses cluster at the 95% confidence interval.

5. Conclusions

[13] Changes in lakes in the Arctic have been observed in a number of local and small regional studies over the past decade. However, only recent advances in the development of remotely-sensed data products have allowed for a large-scale wall-to-wall regional assessment of surface water dynamics. This study presents the first-level quantification of multi-year change in lake extent in high northern latitudes across Canada. The results of this project bring us to four main conclusions. First there was a significant decrease in the total area of open water across Canada ∼6,700 km2 between 2000 and 2009. Second, the net gain and loss of lake area are regional phenomena clustered within specific geographic sub-regions and most likely represent different landscape and climatic drivers at work. Third, clustering of changes in lakes is significant and unrelated to the natural pattern of distribution of lakes in Canada. Fourth, small lakes and ponds are dominant in the Arctic and are also most prone to change due to shallow depth. Previous studies have focused on boreal areas or the Alaskan North slope, however the current results suggest there may be far more change occurring in the tundra farther north that is not currently accounted for. It is likely that this decrease in surface area of inland water is already impacting ecosystem functioning in high northern latitudes of North America and creates multiple feedbacks into the climatic system through modification of surface albedo, evapotranspiration, and carbon cycling. Validation of these changes using field work and fine resolution data is the next challenge.


[14] This work was funded by grants from NASA Earth Sciences Division Terrestrial Ecology program and through the MODIS science team.

[15] The Editor thanks Vanesa Bohn and an anonymous reviewer for their assistance in evaluating this paper.