SEARCH

SEARCH BY CITATION

Keywords:

  • climate change;
  • river runoff;
  • Arctic Ocean

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Hydrological Model and Its Application
  5. 3. Climate Change Scenarios
  6. 4. Change in Runoff to the Arctic Ocean
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] Observational evidence suggests that river inflows to the Arctic Ocean have increased over the last 30 years. Continued increases have the potential to alter the freshwater balance in the Arctic and North Atlantic Oceans and hence the thermohaline circulation. Simulations with a macroscale hydrological model and climate change scenarios derived from six climate models and two emissions scenarios suggest increases of up to 31% in river inflows to the Arctic by the 2080s under high emissions and up to 24% under lower emissions, although there are large differences between climate models. Uncertainty analysis suggests low sensitivity to model form and parameterization but higher sensitivity to the input data used to drive the model. The addition of up to 0.048 sverdrup (Sv, 106 m3 s−1) is a large proportion of the 0.06–0.15 Sv of additional freshwater that may trigger thermohaline collapse. Changes in the spatial distribution of inflows to the Arctic Ocean may influence circulation patterns within the ocean.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Hydrological Model and Its Application
  5. 3. Climate Change Scenarios
  6. 4. Change in Runoff to the Arctic Ocean
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] Rivers draining North America and Eurasia are by far the largest single source of freshwater for the Arctic Ocean: the total catchment area of 22 million km2 is approximately 1.5 times the area of the Ocean itself. Most of these inflows come through a small number of major rivers, including the Yenisey, Lena, Ob', Pechora, Kolyma and Severnaya Dvina in Eurasia and the Mackenzie in North America. Freshwater inflows to the Arctic Ocean influence sea ice formation, density-driven dynamics and stratification, ocean circulation and the biogeochemistry of Arctic aquatic ecosystems. For example, the formation of a cold halocline shielding the surface from heat stored at greater depths depends on river runoff [Steele and Boyd, 1998; Anderson et al., 2004], and a modeling study showed how increases in runoff during the mid-Holocene led to increased freshwater discharge out of the Fram Strait east of Greenland [Prange and Lohmann, 2003]. Macdonald et al. [1999] partially attribute freshening of the upper layers of the Beaufort Sea in the 1990s to a redistribution of river inflows from western to eastern Siberia.

[3] Perhaps the greatest concern, however, is over the extent to which increasing freshwater inflows from the Arctic Ocean to the North Atlantic would lead to a significant weakening of the Atlantic thermohaline circulation and consequent major reductions in temperature across much of the Northern Hemisphere. Vellinga and Wood [2002] showed that the shut down of the thermohaline circulation would within a decade reduce average temperatures across Europe and parts of North America by between 3 and 5°C. Simulations with ocean models [Rahmstorf, 2002] suggest that an increase of between 0.06 and 0.15 sverdrup (Sv, 106 m3 s−1: 1892 to 4730 km3 yr−1) in the amount of freshwater entering the North Atlantic may be sufficient to shut down the thermohaline circulation. Table 1 summarizes the current sources of freshwater into the Arctic Ocean and North Atlantic: while collapse of the Greenland ice sheet would generate very large volumes of water over the long timescale [Gregory et al., 2004], the largest potential source of increased freshwater inputs over the 21st century is increased discharges from rivers draining into the Arctic Ocean.

Table 1. Freshwater Inputs to the Arctic Ocean and North Atlantic
 Input, km3 yr−1Input, SvSource
Precipitation8000.026Gleick [1993]
Meltwater from Greenland2370.0075Lammers et al. [2001]
Icebergs from Greenland3160.01Lammers et al. [2001]
Sea ice melt25000.08Zhang et al. [2003]
River runoff48000.152Lammers et al. [2001]

[4] Analysis of observed runoff shows that discharges into the Arctic have increased over the last 30 years, particularly during the winter [Yang et al., 2002, 2004; Serreze et al., 2002; Lammers et al., 2001]. Peterson et al. [2002] estimated that between 1936 and 1999 aggregated discharge from the six largest Eurasian rivers increased by around 2 km3 yr−1. This increase has been attributed partly to increased precipitation over the areas draining to the Arctic, but partly to changes in permafrost conditions and the gradual release of stored frozen water.

[5] Most climate change scenarios project increases in precipitation at high latitude, and a small number of simulation studies have found increases in runoff in high-latitude basins [e.g., Nijssen et al., 2001a, 2001b; Arora and Boer, 2001; Van der Linden et al., 2003]. Peterson et al. [2002] developed an empirical relationship between air temperature and the volume of inflow from the six largest Arctic rivers, and projected an increase in total runoff by 2100 of between 18 and 70% above the 1961–1990 mean. Increases at the top of this range may provide enough freshwater to influence significantly the thermohaline circulation in the North Atlantic.

[6] This study describes the implications of climate change for the volume and geographical distribution of inflows to the Arctic Ocean, using a macroscale hydrological model to simulate runoff and scenarios derived from several climate models representing two different emissions scenarios. It differs from earlier studies by presenting information on the spatial distribution of change as well as changes in the total volume of inflows to the Arctic Ocean, and by using a range of state-of-the-art climate scenarios representing different emissions. The next section describes the hydrological model and its data requirements, and is followed by a description of the climate scenarios used. Changes in inflows to the Arctic Ocean are then presented, and the final section draws some conclusions.

2. Hydrological Model and Its Application

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Hydrological Model and Its Application
  5. 3. Climate Change Scenarios
  6. 4. Change in Runoff to the Arctic Ocean
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

2.1. Model

[7] River runoff is simulated at a spatial resolution of 0.5° × 0.5° using a macroscale hydrological model which simulates the evolution of the components of the water balance on a daily time step [Arnell, 1999, 2003]. Precipitation falls as snow if temperature is below 0°C, and snowmelts at 4 mm d−1 °C−1 once temperature exceeds 0°C. Precipitation is intercepted by vegetation, and only that in excess of interception capacity (a function of vegetation type) falls to the ground. The rest is evaporated. Potential evaporation is calculated using the Penman-Monteith formula, with stomatal and aerodynamic resistances and leaf area dependent on vegetation type. Water that reaches the ground becomes “quick flow” if the soil is saturated (this is not necessarily overland flow) and infiltrates if the soil is unsaturated. Soil moisture is depleted by evaporation and drainage to groundwater and the stream (“slow flow”). Actual evaporation is a function of potential evaporation and soil moisture content. Soil moisture storage capacity varies across the grid cell following a power distribution, and this means that the proportion of the cell that is saturated varies over time. The proportion of the cell that generates “quick flow” therefore varies with degree of saturation, and in practice streamflow can be generated from at least part of the catchment at almost any time. This is similar in principle to the approach used in the VIC model [Liang et al., 1994], although the precise treatment of water in the soil differs. The average cell soil moisture storage capacity is dependent on soil texture and root depth (based on vegetation class), and the parameter (b) describing the variability in storage capacity is fixed. Models which do not include variations in soil moisture storage capacity (equivalent to b = 0 in the current model) tend to generate no quick flow during summer and the simulated total volume of runoff is very dependent on the rate at which soils become saturated in autumn.

[8] The model is not calibrated to observed runoff data, and physically based model parameters describing vegetation cover and soil properties are determined from spatial databases, as outlined below. The 0.5° × 0.5° grid cells are grouped into watersheds (using Döll and Lehner's [2002] drainage direction map) and these are further grouped to estimate runoff to the different parts of the Arctic Ocean shown in Figure 1.

image

Figure 1. The Pan-Arctic river system organized by sea boundaries.

Download figure to PowerPoint

[9] The model does not explicitly represent freeze-thaw processes characteristic of high-latitude rivers. When the soil is frozen water cannot infiltrate, and both the hydrologically active depth and properties of the soil vary through the year. Rawlins et al. [2003] showed that inclusion of soil ice effects in a hydrological model led to increases in simulated runoff, although Pitman et al. [1999] concluded that the effects were small over large areas. The omission of freeze-thaw processes also means that the model does not incorporate the effects of permafrost degradation. Warmer temperatures would result in a deeper active layer (as simulated by Stendel and Christensen [2002]), which would in turn lead to increased drainage to groundwater in early winter and subsequent increases in winter runoff (and, indeed, this is an explanation for part of the observed increase in winter flows in the major Siberian rivers [Serreze et al., 2002; Frauenfeld et al., 2004]).

2.2. Input Data

[10] The model uses monthly climate data spanning the period 1961 to 1990 at a resolution of 0.5 × 0.5° [New et al., 1999]. The New et al. [1999] data set contains monthly time series for precipitation and temperature, together with mean monthly vapor pressure, wind speed, sunshine hours and number of rain days.

[11] The precipitation data do not include corrections for the undercatch of solid precipitation, which may be very substantial. Adam and Lettenmaier [2003] developed a bias correction procedure which results in an increase in annual precipitation north of 60°N of 27.8% in North America/Greenland and 18.3% in Eurasia. Adam and Lettenmaier's gridded monthly correction factors were applied to the New et al. mean monthly precipitation, producing a 17% increase in estimated total annual precipitation across the land area draining to the Arctic Ocean.

[12] Daily precipitation is generated from monthly total precipitation and the mean monthly number of rain days assuming a two-state Markov model with daily precipitation magnitudes following a gamma distribution with globally fixed shape parameter. The generated daily values are adjusted to reproduce the original monthly total precipitation, and in practice the precise form and parameterization of the disaggregation model does not therefore affect largely simulated average runoff.

[13] Daily temperature is generated stochastically from the monthly temperature data by first fitting a smooth curve to the monthly data and then generating random deviations from the smooth daily value. This enables the more accurate simulation of the fluctuations between rain and snow when temperatures are close to zero, but has no effect when temperatures are well below zero.

[14] Land cover is taken from the global land cover data set produced by de Fries et al. [1998]. This data set was derived from AVHRR satellite data from 1984 classified following procedures developed with the aid of Landsat training data, and has an original spatial resolution of 8 km with land cover divided into 13 classes. For the model application, the dominant land cover class in each 0.5° × 0.5° cell was determined by overlaying a grid onto the 8 km resolution data. Leaf area index, stomatal conductance, root depth, interception storage capacity and percent cover were specified for each land cover class (Table 2): they do not vary through the year.

Table 2. Model Parameters by Land Cover Typea
 Root Depth, mStomatal Conductance rsInterception Capacity, mmLeaf Area IndexVegetation Height, mPercent Cover
  • a

    Root depth, stomatal conductance, leaf area index, canopy height, and canopy capacity data are taken from the nearest land cover class in the Wilson and Henderson-Sellers [1985] data set.

Evergreen needleleaf0.9851.2619.180
Evergreen broadleaf1.51300.7929.490
Deciduous needleleaf0.985141080
Deciduous broadleaf1.21000.6514.980
Mixed forests1.11000.861880
Woodlands1.11000.861850
Wooded grasslands1.11000.861825
Closed bushlands0.980131.740
Open shrublands0.680121.425
Grass0.6700.130.60
Croplands1.21000.6514.910
Bare0.11000000
Mosses/lichens0.11000000

[15] Field capacity and saturation storage capacity are estimated from soil texture using empirical equations developed by Saxton et al. [1986] and used by Vorosmarty et al. [1989]. The geographic distribution of soil texture classes was taken from the FAO Digital Soil Map of the World. The dominant soil class within each 0.5° × 0.5° cell was determined by overlaying a grid onto the FAO polygon coverage.

2.3. Validation

[16] Average annual total runoff into the Arctic Ocean was estimated by Lammers et al. [2001] from observed data to be 4804 km3 yr−1. The simulated 1961–1990 mean total runoff is equal to 4924 km3 yr−1, an overestimation of just 2%. However, estimates for individual river basins may be considerably less precise, as summarized in Table 3. Runoff from the Ob is overestimated (particularly when runoff is expressed in millimeters depth), for example, while runoff from the Lena and Yenisei is underestimated. Similarly, flows in the Nelson are overestimated and those for the Yukon underestimated, and a visual comparison of the spatial pattern of simulated runoff (Figure 2) with the map of observed runoff presented by Lammers et al. [2001] suggests overestimation of runoff in areas draining to the Hudson Bay. Therefore, although the total inflows to the Arctic are approximately correct (because errors balance each other out) the precise geographic distribution of these inflows is incorrect.

image

Figure 2. Simulated average annual runoff across area draining to the Arctic Ocean.

Download figure to PowerPoint

Table 3. Observed and Simulated Baseline Average Annual Runoff for Major River Basinsa
RiverGaugeGauged Area, km2Observed Runoff, mmSimulated Runoff, mmPercent Difference in Runoff DepthObserved Runoff, km3Simulated Runoff, km3Percent Difference in Runoff Volume
  • a

    Observed data are taken from the Global Runoff Data Centre. Differences in percentage error between runoff expressed in km3 yr−1 and mm yr−1 arise due to differences between the observed and simulated basin areas.

ObSalekhard2950000136166224014072
YeniseiIgarka2440000227191−16554467−16
LenaKyusyur2430000222151−32540365−32
KolymaKolymskoye526000186177−59895−3
PechoraUst Tsilma248000433365−16107103−4
Severnaya DvinaUst Pinega348000302286−510582−22
IndigirkVorontso305000169140−175243−17
MackenzieNorman Wells1570000172154−11270243−10
YukonKaltag76700029391−6922568−70
SlaveFitzgerald6060001771791107107−1
NelsonBladder Rapids10000007616111276159110

[17] These errors represent the effects of a combination of factors, including inaccuracies in the delineation of drainage basins at the 0.5° × 0.5° resolution (this effect is seen in the Ob), inaccuracies in the observed data used for validation (including the effects of human interventions on observed runoff), inaccuracies in input data (especially precipitation), and incorrect model formulation or parameterization.

[18] The effect of input data and model characteristics on the simulated runoff to the Arctic Ocean are shown in Table 4. Using precipitation data uncorrected for gauge bias results in simulated total baseline runoff 26% less than that simulated using corrected precipitation. The effect is greatest in the Chuckhi Sea, Arctic Archipelago and Bering Strait basins where the precipitation corrections are greatest.

Table 4. Effect of Input Data and Model Uncertainty on Simulated Baseline Average Annual Runoffa
 Simulated Baseline Runoff, km3 yr−1Ratio of Annual Runoff to Annual PrecipitationPercent Difference From Simulated Baseline Runoff
Uncorrected Precipitationb = 0b = 1b = 2Soil Capacity −25%Soil Capacity +25%
  • a

    See Figure 1 for locations of basins.

Norwegian Sea2350.76−20.9−0.20.10.20.6−0.4
Barents Sea4650.52−28.4−2.31.12.23.1−2.4
Kara Sea13600.40−26.8−7.83.77.35.7−4.0
Laptev Sea5430.39−20.6−6.83.67.26.9−4.7
East Siberian1940.48−23.3−3.82.14.45.5−3.9
Chuckhi780.71−41.9−0.60.30.71.0−0.7
Bering Strait2250.49−33.6−4.32.14.23.1−2.2
Beaufort Sea2830.34−22.8−9.14.99.77.0−5.0
Arctic Arch.1650.59−37.5−0.90.51.12.3−1.7
Hudson Bay9300.47−23.5−4.62.44.93.8−2.7
Hudson Strait2260.76−25.7−0.30.20.40.9−0.7
Foxe Basin1190.72−27.3−0.30.10.31.0−0.7
Baffin Bay830.72−38.4−0.10.10.10.2−0.2
Total49240.46−26.1−4.92.55.04.3−3.0

[19] The broad effects of model uncertainty were assessed by repeating simulations with altered parameter values. Two key model parameters were affected. The parameter b describes the variation in soil moisture storage capacity across the catchment/grid cell. A value of 0 means that there is no variability, and effectively converts the model to a conventional lumped water balance model with no subgrid variability. A value of 1 means that soil moisture storage capacity varies linearly across the cell (for example, 50% of the cell has soil moisture storage less than 50% of the maximum), and as b increases beyond 1 the proportion of the cell with a relatively low storage capacity increases. The parameter b is given a value of 0.5 in the “core” simulations. The second parameter to be varied, cell average soil moisture storage capacity, is based on soil texture and root depth. In the uncertainty analysis it is varied by plus or minus 25% in each grid cell.

[20] Table 4 shows that varying the b parameter and average soil moisture storage capacity has relatively little effect on the absolute value of simulated baseline runoff. The sensitivity of estimated runoff to model characteristics is greatest in the cells and regions with the lowest ratio of average annual runoff to average annual precipitation (as also shown by Arnell [1999] in Europe).

[21] The performance of the model in simulating runoff into the Arctic Ocean is therefore most determined by the quality of the input precipitation data: Rawlins et al. [2003] came to a similar conclusion with their model which, incidentally, performs similarly to the model used in the current study.

3. Climate Change Scenarios

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Hydrological Model and Its Application
  5. 3. Climate Change Scenarios
  6. 4. Change in Runoff to the Arctic Ocean
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[22] The Intergovernmental Panel on Climate Change (IPCC) [2000] defined four “families” of scenarios, each with different patterns of future greenhouse gas emissions, and these have been used to drive a number of global climate models [IPCC, 2001]. Most simulations have been conducted with the high emissions A2 scenario and low emissions B2 scenario. Results from six climate models (HadCM3, ECHAM4, CSIRO MkII, CGCM2, GFDL-R30 and CCSR/NIES2) are available from the IPCC's Data Distribution Centre (http://www.ipcc-ddc.cru.uea.ac.uk), and these were used in the current study. Three repetitions (“ensemble members”) of the A2 scenario and two repetitions of B2 are available for the HadCM3 model.

[23] Table 5 shows the average annual temperature change across the area draining to the Arctic Ocean, relative to the 1961–1990 mean, together with global average temperature change, as simulated by the six models. Increases in the Arctic are substantially greater than the global average increase. The CCSR/NIES2 model produces the largest temperature increases globally and across the Arctic. ECHAM4 simulates the second largest temperature increases in the Arctic, but only the third or fourth highest global average increase.

Table 5. Increase in Global Average Temperature Relative to 1961–1990a
 HadCM3ECHAM4CGCM2CSIRO2GFDL-R30CCSR
  • a

    The values for the HadCM3 simulations for the A2 and B2 emissions scenarios represent the range between the ensemble members. Values are in °C.

A2
Global
   20200.9–1.11.21.21.11.11.1
   20502.1–2.12.12.52.22.12.7
   20803.5–3.73.64.13.83.44.9
Arctic
   20201.6–1.73.01.82.12.02.3
   20503.3–3.55.03.84.03.76.1
   20805.6–6.07.75.96.55.711.7
 
B2
Global
   20201.0–1.01.31.21.31.11.3
   20501.7–1.82.02.002.21.82.7
   20802.6–2.62.82.93.02.53.9
Arctic
   20201.7–1.83.11.92.61.73.3
   20502.8–3.15.02.93.93.06.5
   20804.1–4.36.34.25.24.29.4

[24] Changes in monthly mean temperature, precipitation, net radiation and wind speed (except for GFDL-R30) relative to the 1961–1990 mean were applied to the observed gridded data, to produce 30-year time series characteristic of the 2020s, 2050s and 2080s. Data on vapor pressure changes were available through the IPCC-DDC only for the HadCM3 simulations, so for all the other climate models relative humidity was assumed to remain constant and future vapor pressure was estimated from future temperature and baseline relative humidity. No changes were made to the relative variability in monthly climate from year to year, because information on potential changes is not available through the IPCC-DDC.

[25] Climate change is likely to alter catchment vegetation cover, with replacement of tundra by boreal forest across parts of the study region [White et al., 2000]. This would potentially affect interception and the rate of evaporation, together with feedbacks between the climate and land surface. Although the hydrological model includes land cover, changes in land cover at the 0.5° × 0.5° resolution under each climate scenario are not available so have not been considered in this study.

4. Change in Runoff to the Arctic Ocean

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Hydrological Model and Its Application
  5. 3. Climate Change Scenarios
  6. 4. Change in Runoff to the Arctic Ocean
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

4.1. Total Runoff to the Arctic Ocean

[26] Table 6 shows the percentage change in total precipitation, potential evaporation, actual evaporation, and runoff in the land areas draining to the Arctic Ocean, and Figure 3 shows the change in total runoff in graphical form.

image

Figure 3. Percentage change in average annual river runoff to the Arctic Sea, using scenarios from six climate models and A2 and B2 emissions scenarios.

Download figure to PowerPoint

Table 6. Percentage Change in Water Balance Components Across Land Areas Draining Into the Arctic Ocean, Relative to the 1961–1990 Meana
 HadCM3ECHAM4CGCM2CSIRO2GFDL-R30CCSR
  • a

    The values for the HadCM3 simulations for the A2 and B2 emissions scenarios represent the range between the ensemble members.

A2
Precipitation
   20206–7125978
   205013–15188161120
   208022–242914271838
Potential evaporation
   20209–101512141214
   205020–212523252135
   208037–384033403270
Actual evaporation
   2020812911910
   205015–171816191527
   208027–292825312352
Runoff
   20204–7131656
   205010–1419−112712
   208016–19311221222
 
B2
Precipitation
   20208–91359612
   205013–13207161022
   208018–192310211330
Potential evaporation
   202010–111714171321
   205018–182618251938
   208024–253224332554
Actual evaporation
   20208–913912916
   205014–151913191430
   208020–212318261841
Runoff
   20207–9140637
   205011–1120112512
   208015–1724015716

[27] All the climate models simulate widespread increases in precipitation, with increases over the area draining to the Arctic Ocean ranging from 14 to 38% by the 2080s under A2 emissions and 10–30 under B2 emissions. The CGCM2 model produces the smallest percentage increases in precipitation. By the 2080s, potential evaporation is projected to increase by 32–70% under A2 and 24–54% under B2, with the largest increases with the CCSR climate model. These increases are largely due to increases in temperature, although net radiation increases by large percentages (typically at least 20%, and by more than 50% with the CCSR model) across the study area. Wind speed also increases under the HadCM3, ECHAM4, CGCM2, and CSIRO2 models, although there is little change with CCSR (data are not available for GFDL-R30). Actual evaporation increases by a smaller proportion than potential evaporation, reflecting the constraints imposed by the availability of water.

[28] Both the amount and spatial pattern (Figure 4) of change in runoff are closely related to the change in precipitation, although the effects of the particularly large increase in precipitation with the CCSR model are tempered by the large increases in potential evaporation. Percentage increases in runoff are greatest in the driest catchments with the lowest ratio of runoff to precipitation. The runoff ratio across the entire drainage basin area declines slightly from its current value of 0.46 to between 0.40 and 0.44 by the 2080s under A2 emissions, although it increases slightly with the ECHAM4 model. Changes under B2 emissions are slightly smaller. There is little clear difference in change between the emissions scenarios by the 2020s or 2050s, but by the 2080s changes in precipitation, potential and actual evaporation, and runoff are clearly greater under the high emissions A2 scenario than under B2.

image

Figure 4. Percentage change in average annual runoff by 0.5° × 0.5° grid cell by the 2080s, using scenarios from six climate models and for the A2 emissions scenario.

Download figure to PowerPoint

[29] By the 2080s increases in total runoff to the Arctic Ocean are between 58 and 1500 km3 yr−1 (0.002 to 0.048 Sv) under A2 and between 10 and 1160 km3 yr−1 (0 to 0.037 Sv) under B2. The top of the range estimates are close to, but below, Rahmstorf's [2002] lowest estimate of the threshold input of freshwater needed to shut down the thermohaline circulation in the North Atlantic. The simulated increases in runoff are smaller under all except the ECHAM4 scenarios than the 18–70% increases estimated for six Eurasian rivers by Peterson et al. [2002] from their empirical relationship between annual runoff and global surface air temperature.

4.2. Spatial Distribution of Change

[30] The largest percentage increases in precipitation occur under all climate models in eastern Siberia, Alaska and northern Canada, and this is reflected in the spatial distribution of change in average annual runoff (Figure 4 and Table 7). Under the A2 emissions scenario, runoff to the East Siberian Sea increases by 6–74% by the 2080s, Chuckhi Sea inflows increase by up to 58%, and flows to the Laptev Sea rise by up to 50%. However, rivers flowing into the East Siberian and Chuckhi seas contribute less than 10% of the total inflows from Eurasia, so while large changes in these areas might produce changes in sea ice characteristics, they will have relatively little effect on the Arctic Ocean freshwater budget. Relative increases are smaller further west: runoff into the Kara Sea, which accounts for nearly half of all inflows into the Arctic from Eurasia, changes by between 0 and +38%. Changes in flows from the Ob (the westernmost of the two major rivers entering the Kara Sea) are very small, partly due to large percentage reductions in runoff in its headwaters in central Asia. Percentage increases in runoff are considerably smaller for the rivers draining North America, and for the two largest basins (draining into the Beaufort Sea and Hudson Bay) two of the seven climate model scenarios result in a decrease in runoff.

Table 7. Percentage Change in Mean Runoff by Sea Basin: 2080s Compared to 1961–1990 Meana
 Percent Change From Baseline
HadCM3ECHAM4CGCM2CSIRO2GFDL-R30CCSR
  • a

    The values for the HadCM3 simulations for the A2 and B2 emissions scenarios represent the range between the ensemble members. See Figure 1 for locations of basins and Table 3 for simulated baseline runoff by sea basin.

A2
Norwegian Sea2–103014221423
Barents Sea−1–841724332
Kara Sea4–11380181121
Laptev Sea39–41507502435
East Siberian Sea54–74416433435
Chuckhi Sea43–49581272127
Bering Strait43–51444242016
Beaufort Sea15–18−5−61628
Arctic Arch.28–3650−5482136
Hudson Bay4–6−1−84−311
Hudson Strait19–23326212121
Foxe Basin24–314315212429
Baffin Bay20–30455423424
 
B2
Norwegian Sea3–823715915
Barents Sea4–7351111119
Kara Sea7–832112812
Laptev Sea30–31384371324
East Siberian Sea49–56334302033
Chuckhi Sea38–40365181314
Bering Strait41–41305181314
Beaufort Sea13–19−3−101169
Arctic Arch.21–2328−740927
Hudson Bay8–11−2−60411
Hudson Strait18–2326215814
Foxe Basin23–2435621822
Baffin Bay18–19237321317

4.3. Effect of Data and Model Uncertainty on Change in Runoff to the Arctic Ocean

[31] Section 2.3 showed how the absolute value of simulated runoff to the Arctic Ocean was influenced to a large extent by the input precipitation data and to a lesser extent by model form and parameterization. Table 8 shows the percentage change in runoff, by sea, using different input data sets and model parameter sets, under one of the HadCM3 A2 ensemble scenarios for the 2080s (similar results are found with the other climate models).

Table 8. Effect of Data and Model Uncertainty on Estimated Change in Runoff to the Arctic Ocean
 Average Annual Runoff in the 2080s as Percent Difference From 1961–1990 Average Annual Runoff
Core SimulationUncorrected Precipitationb = 0b = 1b = 2Soil Capacity −25%Soil Capacity +25%
Norwegian Sea5.12.14.95.25.35.54.9
Barents Sea8.51.86.79.210.010.06.9
Kara Sea6.81.45.57.37.78.65.5
Laptev Sea41.438.043.440.439.541.740.9
East Siberian74.170.077.072.570.872.275.3
Chuckhi48.845.248.648.949.049.148.6
Bering Strait51.249.252.350.750.251.151.3
Beaufort Sea14.812.612.915.115.316.613.4
Arctic Arch.33.230.332.633.533.734.032.5
Hudson Bay3.6−1.21.34.55.25.62.0
Hudson Strait19.316.019.119.419.619.818.9
Foxe Basin31.029.230.831.031.131.430.6
Baffin Bay30.412.229.331.031.632.129.1
Total18.413.617.721.121.419.617.4

[32] Using precipitation data uncorrected for gauge bias results in a considerably smaller percentage change in runoff by the 2080s, and in one case (Hudson Bay) changes the direction of change. Applying the precipitation bias correction produces larger increases in runoff because the correction factors are greater in winter than summer and percentage increases in precipitation are highest in winter under all scenarios. Varying the model form and parameterization has very little effect on the simulated percentage change in runoff, and there is little variation in effect between seas.

5. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Hydrological Model and Its Application
  5. 3. Climate Change Scenarios
  6. 4. Change in Runoff to the Arctic Ocean
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[33] Sea ice formation, stratification and circulation within the Arctic Ocean are influenced by the volume and distribution of freshwater inflows from basins draining Eurasia and North America, and an increasing volume of freshwater flows from the Arctic to the North Atlantic could lead to a significant weakening of the thermohaline circulation. There is empirical evidence that flows to the Arctic have increased over the last three decades, at least from Eurasia, and most climate change scenarios project increases in precipitation in the future across most of the parts of Eurasia and North America draining to the Arctic. This paper describes the application of a macroscale hydrological model to simulate future river flows draining to the Arctic Ocean under climate change scenarios assuming different rates of future emissions and using different climate models. The model simulates well the total volume of river inflows to the Arctic Ocean, although underestimations in some areas, including the Yenisei, Lena and Yukon catchments, are offset by overestimations in others, including the Ob and Nelson catchments. The model also does not explicitly take into account the changing effects of human interventions, land cover change or the effects of thawing of permafrost, so the results are to be seen as indicative only of general patterns.

[34] Under all but one of the climate models runoff increases during the 21st century, to a maximum of 31% under high emissions (A2) and 24% under low emissions (B2): under the remaining climate model there is little change in runoff because relatively low increases in precipitation are offset by increases in potential evaporation. The simulated increases in runoff – corresponding to a maximum additional 0.048 and 0.037 Sv under A2 and B2 respectively – are smaller than those estimated by Peterson et al. [2002] on the basis of an empirical relationship between global temperature and runoff in six major Arctic river basins. They are close to, but below, Rahmstorf's [2002] lowest estimate of the threshold for the amount of additional freshwater necessary to switch off the thermohaline circulation.

[35] There is little clear difference between high and low emissions to the 2050s, but thereafter the high emissions scenario has a greater effect on river flows. There is much greater variability between climate models, with increases in runoff varying from 1% to 31% by the 2080s under the A2 emissions scenario. The different models produce broadly similar patterns of change in precipitation, and the variation in impacts on runoff is largely due to differences in the simulated magnitude of change in precipitation. Under each climate model the greatest percentage increase in runoff occurs in eastern Siberia and Alaska, resulting in a shift in the relative distribution of inflows to the Arctic Ocean away from North America toward the Laptev and East Siberian seas.

[36] A model sensitivity analysis showed that uncertainties in model form and parameterization had little effect on the simulated changes in runoff to the Arctic Ocean, but the simulated changes were more sensitive to the input baseline precipitation data used to drive the model. Explicit incorporation of the degradation of permafrost, however, would probably lead to larger increases in river runoff, although the magnitude of this effect is currently uncertain.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Hydrological Model and Its Application
  5. 3. Climate Change Scenarios
  6. 4. Change in Runoff to the Arctic Ocean
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[37] Climate baseline data and HadCM3 climate change scenarios were provided through the Climate Impacts LINK project at the University of East Anglia by David Viner. The other climate scenarios were taken from the IPCC Data Distribution Centre Web site (http://ipcc-ddc.cru.uea.ac.uk). Jennifer Adam from the University of Washington kindly provided the monthly precipitation correction factors. The helpful comments of the referees are gratefully acknowledged.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Hydrological Model and Its Application
  5. 3. Climate Change Scenarios
  6. 4. Change in Runoff to the Arctic Ocean
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Hydrological Model and Its Application
  5. 3. Climate Change Scenarios
  6. 4. Change in Runoff to the Arctic Ocean
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
jgrd11802-sup-0001-t01.txtplain text document0KTab-delimited Table 1.
jgrd11802-sup-0002-t02.txtplain text document1KTab-delimited Table 2.
jgrd11802-sup-0003-t03.txtplain text document1KTab-delimited Table 3.
jgrd11802-sup-0004-t04.txtplain text document1KTab-delimited Table 4.
jgrd11802-sup-0005-t05.txtplain text document1KTab-delimited Table 5.
jgrd11802-sup-0006-t06.txtplain text document1KTab-delimited Table 6.
jgrd11802-sup-0007-t07.txtplain text document1KTab-delimited Table 7.
jgrd11802-sup-0008-t08.txtplain text document1KTab-delimited Table 8.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.