Understanding the Causes of Hydrologic Changes in the Eurasian Arctic UBC/UW Hydrology Conference September 29, 2006 Jennifer Adam Dennis Lettenmaier Department of Civil and Environmental Engineering University of Washington We have been investigating the causes of increased stream flow for northern Eurasian rivers and this study involves two parallel efforts. The first uses mainly observational data and the second is a modeling study. In this talk, I will be presenting some of the results from the observational data analysis.

Study Domain: Eurasian Arctic Study Period 1936 - 2000 Study Basins 3 primary 9 secondary Indigirka Lena Yenisey The colored areas in this figure represent mean annual temperature for all the land areas that discharge river flow to the Arctic Ocean and Hudson Bay. In this study, we are focused on Eurasia and have selected three primary study basins and nine secondary basins and sub-basins. The primary basins are the Lena, Yenisey, and Ob’ which are all among the top ten longest river systems in the world. The secondary basins include two independent basins (the Indigirka in the far east and the Severnaya Dvina which is our only study basin in european russia), and seven sub-basins to the Lena, Yenisey, and Ob. The eurasian part of our study domain has a strong east/west temperature gradient, such that eastern siberia is extremely cold and western siberia is comparatively warm. Therefore, there is much diversity in the hydrologic processes between these basins. Why eurasia?? Ob’ Severnaya Dvina -18 -12 -6 0 6 Mean Annual Air Temperature, C

Observed Streamflow Trends Discharge, km3/yr Annual trend for the 6 largest rivers Peterson et al. 2002 Discharge to Arctic Ocean from six largest Eurasian rivers is increasing, 1936 to 1998: +128 km3/yr (~7% increase) Most significant trends during the winter (low-flow) season The primary reason we are interested in the Eurasian Arctic has to do with a study that was published in Science in 2002. In this paper, Peterson et al. show that the combined riverine input of the six largest northern eurasian basins has increased by approximately 7% since the 1930’s. Furthermore, of all seasons, the winter flows have the most significant trends, which is shown here for the Ob’ river. The bottom figure shows the hydrograph, or mean monthly flows, at the outlet of the Ob’ basin. The shape of this hydrograph is typical for northern rivers in which there are strong contrasts between seasons – with very high flows during the late spring and summer and very low flows during the cold season. Therefore, the most significant changes are occurring during the season with the least flow. The purpose of this study is to investigate the causes of these stream flow changes. J F M A M J J A S O N D 10 20 30 40 Discharge, m3/s GRDC Monthly Means Ob’ 1950 1960 1970 1980 Discharge, km3 Winter Trend, Ob’ Bowling et al. 2002

Importance of Arctic Climate Studies 1 Largest warming observed in Arctic regions Arctic Regions, >62 °N Northern Hemisphere Observed SAT Trend, °C/yr There are a number of reasons why it is important to study the impacts and consequences of climate change in the Arctic. First, the Arctic regions are warming faster than the rest of the globe, on average. This is a plot of surface air temperature trend slope for various periods, all ending in 2001. This plot, which was made using observed historical data, shows that warming in the arctic has been greater in the longterm and also accelerating faster than warming over the entire hemisphere. Stress that this is historical and calculated from observed data. Jones et al. 1999

Importance of Arctic Climate Studies 2 Arctic regions sensitive to change; many changes observed already Hydrological sensitivity to small changes in temperature, especially when and where temperatures are near threshold 0 °C Currently experiencing system-wide change: rivers, permafrost, snow, glaciers, wetlands, vegetation zonation, fire frequency… Second, the Arctic is especially sensitive to change. One of the reasons why the arctic is so sensitive to warming is because temperatures are within a few degrees of freezing over much of the region much of the time – therefore a slight change in temperature can invoke a phase change in water which affects land surface hydrology in various ways. Some of the changes already observed include changes to streamflow, snow, and permafrost.

Importance of Arctic Climate Studies 3 Potential for feedback response to global climate Global Thermohaline Circulation Third, climate-induced changes to the Arctic have the potential to affect global climate and several feedback mechanisms have been identified. One of these feedbacks is an ocean circulation feedback, involves what is known as the global conveyor belt, or the thermohaline circulation. This circulation caries heat via the gulf stream from the lower latitudes to the higher latitudes in the north atlantic and is the reason why europe is considerably warmer than other regions at the same latitude. The major drivers of this circulation involve freshwater export to the north atlantic, so any changes in the freshwater budget in the Arctic and North Atlantic are a concern.

Importance of Arctic Climate Studies 4 Low level of understanding How Arctic land surface hydrology will respond to predicted climate changes How elements of the cryosphere (e.g. permafrost) modulate these interactions How Arctic hydrological changes may provide a feedback response to global climate Fourth, …

Research Question To what extent can long-term trends in Eurasian Arctic streamflow be explained by changes in precipitation and what are the most likely explanations for discrepancies between streamflow and precipitation trends, e.g. reservoir construction, changes in evapotranspiration (ET), and/or changes in soil moisture dynamics (such as permafrost degradation)?

Streamflow Trend Attribution Water Balance: Hypothesized contributors: Acceleration of the hydrologic cycle: P , ET? Permafrost Degradation: dS/dt , ET? Reservoir Operation: dS/dt?, ET? Other: fires, land use, wetlands, clouds, … Storage,S: ground water/ice, lakes, surface ice… ? We used the basic water balance equation to study this problem. This equation is valid for all spatial and time scales. Given that streamflow, which is symbolized by Q, is increasing, it follows that one or more of the following must be true: precipitation is increasing, evapotranspiration is decreasing, or there must be a release of water from storage. This storage can be either ground water or ice (such as permafrost) or surface water or ice (such as lakes and glaciers). Numerous potential controls have been hypothesized, but researchers continue to focus on three of them as the most likely primary contributors to increased streamflow. First, an acceleration of the world’s hydrologic cycle could result in more precipitation over the northern basins. Second, warming of the permafrost, or frozen ground, could release stored water, that may contribute to streamflow. Third, various reservoirs were constructed in the primary basins during the study period. Reservoirs certainly affect streamflow seasonality, but it is also possible that annual streamflow volumes may be affected through changes in storage release and/or evaporation. There are other potential contributors that play smaller roles, but we focus on the first three agents because they are most likely to play a primary role.

Status of Current Research McClelland et al. (JGR, 2004) Berezovskaya et al. (GRL, 2004) Pavelsky and Smith (JGR, in review) New contributions from our work: Seasonality taken into account Multiple periods Defined a set of most likely primary controls (most to least important) for a range of basins across the domain Where did these papers occur? In the last two years, there have been three publications dealing explicitly with this question. Mclell considered the 6 largest basins as a whole and considered separately permafrost degradation, increased fires, and reservoirs as the primary agent of change and found that not a single one of these agents can solely be responsible for observed streamflow increases, therefore suggesting that the precipitation changes are likely the controlling agent. Bere compared precipitation trends to streamflow trends for the three primary basins and found them inconsistent, especially for the basins with the largest increases. Pavel and smith performed a similar analysis to that of berezovskaya but at a finer spatial resolution for 66 basins in the eurasian arctic – they found that, cumulatively, precip changes can explain 62% of the discharge changes, and therefore plays an important role but not the only important role and often not even the primary role for some basins. The work that I have done for this paper builds on the work of these other studies with three new contributions: We took seasonality into account in our trend testing which is important to do because the most significant changes are occuring during the winter low-flow season we did not limit ourselves to looking at trends for just one or two periods but for a large number of periods of varying duration during the study period – this is because the controls on streamflow variability operate at varying time-scales and in different periods a set of most likely primary controls listed for each basin in order from most important to least important.

Major Reservoirs (Capacity>1 km3) Lena: 1 Yenisey: 8 Ob’: 3 With the potential agents of change in mind (precipitation changes, reservoir construction, permafrost degradation), I will offer a little more background information. This figure shows the locations of the major reservoirs in these river basins. Give with respect to grand coulee – e.g. grand coulee (about 10 MAF is equal to about 12 km3) – our reservirs range in size from about 2 km3 to about 170 km3 McClelland et al. 2004

Frozen Ground and Hydrology Key features of frozen ground permafrost vs. seasonally frozen ground – permafrost is perennially frozen; hydrologic activity is constrained to the warm season active (thawed) layer in permafrost impermeable boundary – frozen groundwater is hydrologically inactive and limits infiltration, although conduits can exist in non-continuous permafrost types; baseflow generation is affected excess ground ice – beyond what unfrozen soil can hold; mainly in top 10 m of permafrost; inherent uncertainty Permafrost degradation can affect the seasonality of stream flow (possibly by increasing winter discharge) as well as annual stream flow volumes. Annual streamflow volumes will increase if there is a net melting of excess ground ice. Excess ground ice can occur as massive ice, or as thin layers or flakes, or in expanded soil pores due to cryosuction. Describe the photos briefly.

Permafrost Distribution (Brown et al. 1998) Lena: 100% permafrost (all types) Yenisey: 89% permafrost (all types) Ob’: 26% permafrost (all types) This image shows the permafrost distribution for the study domain, and this distribution looks very similar to the temperature distribution shown earlier, in which there is continuous permafrost (defined as 90-100% permafrost by area) in the coldest regions of siberia, discontinuous, sporadic, and isolated permfrost exist in the central areas of siberia, and seasonally frozen soil in the west. Continuous , 90-100% Discontinuous, 50-90% Sporadic, 10-50% Seasonally Frozen Ground Isolated, <10%

Hypothesis Formulation Q, mm/yr 320 280 240 200 -2 0 T, C Correlation Analysis: Annual Basin Mean Air Temperature vs. Annual Streamflow e.g. Ob’: ρ=-0.4 COLD basins: T < -15 °C T control on Q through permafrost melt (contributes to Q in winter) THRESHOLD basins: -15 °C < T < -5 °C T control on Q through ET changes and permafrost melt WARM basins: T > -5 °C T control through summer ET changes Resulting Hypothesis: We investigated the relationship between air temperature and stream flow variability by calculating the correlation coefficient between the two variables for each of the twelve study basins. Each plot shows the correlation coefficient along the y-axis plotted against (first) basin mean annual air temperature and (second) percent continuous permafrost ob the x-axis. There is a plot for annual, summer, and winter correlations. Each circle represents one of the study basins. Look first at the annual correlations. From this plot, it becomes apparent that the basins can be sorted into one of three regimes. First, in dark blue is the very cold basin, followed by the “threshold” basins in green that are underlain by permafrost but are not 100% continuous, the Lena and Yenisey basins and sub-basins. The cold and threshold basins all have positive correlation coefficients, indicating that as temperature increases, stream flow also increases, most likely due to the melting of ground ice. Third, in red, are the warm basins that are primarily in the region of seasonally frozen soil, the Ob’ and Severnaya Dvina basins and sub-basins. These basins all have negative correlations, indicating that as temperature increases, stream flow decreases, most likely due to increased evaporation over the basin. Furthermore, if the correlations are examined separately for the warm season and the cold season, it becomes apparent that summer effects lead to the negative correlations for the warm basins whereas winter effects lead to the positive correlation coefficient for the cold basin. The threshold basins appear to be affected by both summer and winter effects. In summary, … -15 -10 -5 0 T, C 0 20 40 60 80 100 Cont. Permafrost, % ρ 0.6 0.3 0.0 -0.3 -0.6 ρ=

Trend Analysis Selection of trend test: * Sensitive to seasonal differences in trend Varying periods between 1936 and 2000 (~400) Test for 99% significance Calculate trends for precipitation, temperature, and streamflow (gauged and reconstructed) Linear Regression (LR) Mann-Kendall/ Sen Slope (MK) Seasonal* Mann-Kendall/Sen Slope (SMK) Annual Data Monthly Data Normally distributed Non-parametric We used trend analysis to evaluate the potential role that precipitation may play in controlling stream flow trends. We considered three trend tests: linear regression, the non-parametric mann-kendall test, and a seasonal variation of the mann-kendall test. Although the first two tests produced similar results, they were poor at discerning trends when they occurred only during the winter low-flow season. This is because these changes are obscured by the high variability of the summer peak flows. Instead, we opted to use the seasonal mann-kendall test which is sensitive to seasonal differences. This is because the seasonal mann-kandell statistic is calculated by summing over the seasonal statistics rather than the seasonal flows. This is important because, as I mentioned earlier, the most significant changes have occurred during the winter. Because the controls on streamflow variability operate at varying time-scales and in different periods, we computed the trend for a large number of periods between 1936 and 2000. We then tested for 99% significance using a two-tailed procedure. This was done for stream flow, precipitation, and temperature data. I should also mention that we are using two streamflow products: we use gauged data for all study basins, and for the primary study basins we are also using a reconstructed dataset in which the effects of dams have been removed.

Temperature & Precipitation Trends, 99% Significant Lena Yeni. Temperature & Precipitation Trends, 99% Significant Ob’ Secondary basins are in order from coldest to warmest Secondary Basins These plots show the observed trends for the precipitation and temperature data. The figure in the upper left shows, for the Lena basin, a line for each of the trends that were found to be significant at 99%. The color of each line gives the trend magnitude, blue for cooling, or red for warming. The line begins and ends at the start and end of the period, so that the length of the line portrays the period length. Plots of this kind can be thought of as an alternative way to view a time series. The first three plots are for the primary basins, the Lena, Yenisey, and Ob’. Whereas the following nine are for the secondary basins in order from east to west. There is a general long-term warming occurring over northern eurasia, although not all the basins are showing long periods of warming. The ob’ shows significant warming only since the mid 60’s where as the very cold basin (the Indigirka) as well as the warm european basin (the severnaya dvina) are not showing much warming at all. The right plots indicates that precipitation is not increasing in the long term. Each line segment is an analysis for a particular period, e.g. . The color represents the temperature trend for that particular period. Also, although I performed trend analysis on nearly 400 periods between 1936 and 2000, I only plotted the results for each period if the trend for that period was significant. This is the reason that some basins have more lines than other basins. Boxes with no lines indicate that there are no significant changes occuring for that basin. Temp. C/year Prcp. mm/year2

Streamflow Trends, 99% Ob’ Lena Yenisey Indigirka Aldan (Lena) Lena (head) Ob’ (head) These plots show the observed trends for the streamflow data. For each of the primary study basins, there is a plot both for gauged streamflow data and for the reconstructed data. There are also separate plots for each of the secondary basins, in which the basins are ordered from east to west (in other words from cold to warm). I will point out a few key features in these figures. First, we can get an indication of how the reservoirs have affected streamflow trends. For the Lena, there are a large number of periods with significant trends for the gauged data, where as there are much fewer periods of long-term increasing flows for the reconstructed data. This indicates that the Lena’s one large reservoir has played an important role in the observed streamflow changes for that basin, probably mostly through a chance in seasonality. For the Yenisey, there are fewer periods with significant trends with the reconstructed data, but there are still enough periods with positive trends to indicate that the reservoirs are not the only important control for this basin. For the Ob’, the reconstructed dataset actually has more significant periods, but the general trends are similar which indicates that the reservoirs have some effect but are probably not the primary player. The secondary basins demonstrate more diverse behavior with some basins not showing much change, others showing mainly positive trends, or mainly negative trends, and others showing a mix between the two. Recall that there were very few significant longterm trends for precipitation. The lack of positive trends is one of the reasons why research to understand controls on long-term stream flow trends has been confounded. Although there is agreement that precipitation must be a major (if not the major) player, observations do not show significant precipitation increases. The explanation is that there is considerably more variability in precipitation during the season when it is changing than there is in streamflow during the season when it is changing. Actuality, there are increases occurring in the precipitation that can more than account for some of the changes in streamflow but because of this large variance, they are not statistically significant. Yenisey S. Dvina Q, mm/year2

Precipitation Trends (streamflow periods 99%) Ob’ Lena Indigirka Aldan (Lena) Lena (head) Ob’ (head) Therefore, rather than examining precipitation trends for periods in which they are statistically significant, we examined the precipitation trends for the periods in which the streamflow observations are statistically significant, and plotted the results in the same way that we plotted the streamflow trends. So by comparing these trends to the observed streamflow trends, we can get an indication of the importance of precipitation to streamflow change for each basin. Yenisey S. Dvina Prcp. mm/year2

Stream Flow/Precipitation Trends Lena Gauged Recon. Stream Flow Trend, mm/yr2 This comparison was done using scatter plots in which I plotted the stream flow trend along the y-axis and the precipitation trend along the x-axis such that the one-to-one line always goes through the diagonal of the plot. The primary basins are at the top for both gauged and reconstructed trends, followed by the nine secondary basins. I’ll discuss the primary basins only. Looking at the Lena, there are large differences in the scatter between the gauged and reconstructed trends which, as I’ve stated already, indicates that the reservoir is a major player. Looking at the reconstructed data, there are a few points on the one to one line – a few of these are long-term precipitation increases large enough to explain the streamflow increase for that period, and some larger decreasing trends that match for the periods during the middle of the century. Points in the upper left corner indicate that streamflow is increasing regardless of precipitation decreases which suggests permafrost degradation during the later half of the century. Points in the upper right quadrant below the one to one line indicate that evaporation may be increasing because the precipitation trends are larger than the streamflow trends. For the Yenisey, there are smaller changes between the gauged and reconstructed suggesting a smaller role for reservoirs (despite the fact that there are 8 of them there). Streamflow trends are almost all positive even though precipitation trends are mainly negative and this points to permafrost degradation as the primary control, although a few points lie on the one to one lie so precipitation may play a minor role. There are some differences between the Ob’ gauged and reconstructed plots, but for the most part the points line along the one to one line, indicating precipitation has the primary role. Gauged Precipitation Trend, mm/yr2

Example: Lena Gauged Reconstructed Is Gauged different from Reconstructed? YES reservoirs important Stream Flow Trend, mm/yr Precipitation Trend, mm/yr -2 0 2 0 2 2 -2 Gauged (with reservoir) 2. Do points ~form a straight line? YES precipitation important 3. Is this line tilted away from the one-to-one line? YES ET changes occurring This comparison was done using scatter plots in which I plotted the stream flow trend along the y-axis and the precipitation trend along the x-axis such that the one-to-one line always goes through the diagonal of the plot. The primary basins are at the top for both gauged and reconstructed trends, followed by the nine secondary basins. I’ll discuss the primary basins only. Looking at the Lena, there are large differences in the scatter between the gauged and reconstructed trends which, as I’ve stated already, indicates that the reservoir is a major player. Looking at the reconstructed data, there are a few points on the one to one line – a few of these are long-term precipitation increases large enough to explain the streamflow increase for that period, and some larger decreasing trends that match for the periods during the middle of the century. Points in the upper left corner indicate that streamflow is increasing regardless of precipitation decreases which suggests permafrost degradation during the later half of the century. Points in the upper right quadrant below the one to one line indicate that evaporation may be increasing because the precipitation trends are larger than the streamflow trends. For the Yenisey, there are smaller changes between the gauged and reconstructed suggesting a smaller role for reservoirs (despite the fact that there are 8 of them there). Streamflow trends are almost all positive even though precipitation trends are mainly negative and this points to permafrost degradation as the primary control, although a few points lie on the one to one lie so precipitation may play a minor role. There are some differences between the Ob’ gauged and reconstructed plots, but for the most part the points line along the one to one line, indicating precipitation has the primary role. Reconstructed (without reservoir) 4. Are there scattered points above the one-to-one line? ~YES permafrost important?

Example: Yenisey Gauged Reconstructed Is Gauged different from Reconstructed? ~YES reservoirs important? 4 2 Gauged (with reservoir) 2. Do points ~form a straight line? NO precip. not important? Stream Flow Trend, mm/yr 0 2 4 2 3. Is this line tilted away from the one-to-one line? NA 4. Are there scattered points above the one-to-one line? YES permafrost important This comparison was done using scatter plots in which I plotted the stream flow trend along the y-axis and the precipitation trend along the x-axis such that the one-to-one line always goes through the diagonal of the plot. The primary basins are at the top for both gauged and reconstructed trends, followed by the nine secondary basins. I’ll discuss the primary basins only. Looking at the Lena, there are large differences in the scatter between the gauged and reconstructed trends which, as I’ve stated already, indicates that the reservoir is a major player. Looking at the reconstructed data, there are a few points on the one to one line – a few of these are long-term precipitation increases large enough to explain the streamflow increase for that period, and some larger decreasing trends that match for the periods during the middle of the century. Points in the upper left corner indicate that streamflow is increasing regardless of precipitation decreases which suggests permafrost degradation during the later half of the century. Points in the upper right quadrant below the one to one line indicate that evaporation may be increasing because the precipitation trends are larger than the streamflow trends. For the Yenisey, there are smaller changes between the gauged and reconstructed suggesting a smaller role for reservoirs (despite the fact that there are 8 of them there). Streamflow trends are almost all positive even though precipitation trends are mainly negative and this points to permafrost degradation as the primary control, although a few points lie on the one to one lie so precipitation may play a minor role. There are some differences between the Ob’ gauged and reconstructed plots, but for the most part the points line along the one to one line, indicating precipitation has the primary role. Reconstructed (without reservoir) 0 2 Precipitation Trend, mm/yr

Summary of Controls Basin Most Likely Controls Lena Reservoir, Precipitation, Permafrost?, ET?, other? Yenisey Permafrost, Reservoirs, Precipitation? Ob’ Precipitation, Reservoirs?, ET? Indigirka other and Precipitation (little change) Aldan (Lena) Permafrost, Precipitation? Lena (head) Precipitation, Permafrost?, other? Nizhn. (Yenisey) ? (change is small) Pod. (Yenisey) Precipitation and ET Ob’ (head) Precipitation and other (ET?, Reservoir?) Irtish (Ob’) Precipitation and ET (Reservoirs?) Tobol (Ob’) Precipitation?, ET?, other? S. Dvina Precipitation and ET (other?) This table summarizes what the results are indicating as the most likely primary controls for each of the basins (in order of most to least important). As suspected, not one of the hypothesized controls is uniformly dominant, and in fact there are usually two and sometimes three or more important contributing factors. Furthermore, there are probably other controls that are as yet unknown. The most obvious case is for the coldest basin, the Indigirka. We’ve shown that precipitation may be playing a role, but it’s not clear what is playing the primary role in that basin. In general precipitation is playing the primary role for the warmer basins, whereas permafrost degradation is playing a primary role for some of the threshold basins. We have also shown that reservoirs are having an effect.

Final Remarks Warm basins: Precipitation coupled with ET effects are major contributors; reservoirs play a role for regulated basins. Permafrost basins: complex interactions among contributors Threshold basins: difficult to predict which basins most influenced by permafrost Ground ice is the main complicating factor in understanding long-term streamflow changes Modeling study concurrently being applied (e.g., to explore the role of ground ice as a linkage between climate and streamflow changes)

Questions?

Stream Flow/Precipitation Trends Yenisey Gauged Recon. Stream Flow Trend, mm/yr2 This comparison was done using scatter plots in which I plotted the stream flow trend along the y-axis and the precipitation trend along the x-axis such that the one-to-one line always goes through the diagonal of the plot. The primary basins are at the top for both gauged and reconstructed trends, followed by the nine secondary basins. I’ll discuss the primary basins only. Looking at the Lena, there are large differences in the scatter between the gauged and reconstructed trends which, as I’ve stated already, indicates that the reservoir is a major player. Looking at the reconstructed data, there are a few points on the one to one line – a few of these are long-term precipitation increases large enough to explain the streamflow increase for that period, and some larger decreasing trends that match for the periods during the middle of the century. Points in the upper left corner indicate that streamflow is increasing regardless of precipitation decreases which suggests permafrost degradation during the later half of the century. Points in the upper right quadrant below the one to one line indicate that evaporation may be increasing because the precipitation trends are larger than the streamflow trends. For the Yenisey, there are smaller changes between the gauged and reconstructed suggesting a smaller role for reservoirs (despite the fact that there are 8 of them there). Streamflow trends are almost all positive even though precipitation trends are mainly negative and this points to permafrost degradation as the primary control, although a few points lie on the one to one lie so precipitation may play a minor role. There are some differences between the Ob’ gauged and reconstructed plots, but for the most part the points line along the one to one line, indicating precipitation has the primary role. Gauged Precipitation Trend, mm/yr2

Reservoirs