Long-Term Caspian Sea Level Change

Abstract

Caspian Sea level (CSL) has undergone substantial fluctuations during the past several hundred years. The causes over the entire historical period are uncertain, but we investigate here large changes seen in the past several decades. We use climate model-predicted precipitation (P), evaporation (E), and observed river runoff (R) to reconstruct long-term CSL changes for 1979–2015 and show that PER (P-E + R) flux predictions agree very well with observed CSL changes. The observed rapid CSL increase (about 12.74 cm/yr) and significant drop (~−6.72 cm/yr) during the periods 1979–1995 and 1996–2015 are well accounted for by integrated PER flux predictions of ~+12.38 and ~−6.79 cm/yr, respectively. We show that increased evaporation rates over the Caspian Sea play a dominant role in reversing the increasing trend in CSL during the past 37 years. The current long-term decline in CSL is expected to continue into the foreseeable future, under global warming scenarios.

Full text

Long-term Caspian Sea level change
J. L. Chen
1
, T. Pekker
1
, C. R. Wilson
1,2
, B. D. Tapley
1
, A. G. Kostianoy
3,4
, J.-F. Cretaux
5
,
and E. S. Safarov
6
1
Center for Space Research, University of Texas at Austin, Austin, Texas, USA,
2
Department of Geological Sciences, Jackson
School of Geosciences, University of Texas at Austin, Austin, Texas, USA,
3
P.P. Shirshov Institute of Oceanology, Moscow,
Russia,
4
Laboratory of Integrated Studies of Water Resources, S.Yu. Witte Moscow University, Moscow, Russia,
5
Legos/CNES,
Toulouse, France,
6
Department of the Caspian Sea Level Problem, Institute of Geography, Baku, Azerbaijan
Abstract Caspian Sea level (CSL) has undergone substantial fluctuations during the past several hundred
years. The causes over the entire historical period are uncertain, but we investigate here large changes seen
in the past several decades. We use climate model-predicted precipitation (P), evaporation (E), and observed
river runoff (R) to reconstruct long-term CSL changes for 1979–2015 and show that PER (P-E+R)flux
predictions agree very well with observed CSL changes. The observed rapid CSL increase (about 12.74 cm/yr)
and significant drop (~6.72 cm/yr) during the periods 1979–1995 and 1996–2015 are well accounted for by
integrated PER flux predictions of ~+12.38 and ~6.79 cm/yr, respectively. We show that increased
evaporation rates over the Caspian Sea play a dominant role in reversing the increasing trend in CSL during
the past 37 years. The current long-term decline in CSL is expected to continue into the foreseeable future,
under global warming scenarios.
1. Introduction
The Caspian Sea is the world’s largest inland water body, covering an area of ~371,000 km
2
(excluding the
Kara-Bogaz-Gol Bay) and extending 1200 km from north to south (from Wikipedia online information at
https://en.wikipedia.org/wiki/Caspian_Sea). Located within an endorheic (no outflow) basin between
Europe and Asia, the Caspian Sea is surrounded by five countries: Russia, Kazakhstan, Turkmenistan, Iran,
and Azerbaijan (see Figure 1). Without a connection to the global oceans, average Caspian Sea level (CSL)
is currently approximately 27.5 m below mean sea level. Over the past several hundred years, CSL has been
characterized by substantial fluctuations, including changes of several meters within the past few decades
[e.g., Kosarev and Yablonskaya, 1994; Cazanave et al., 1997; Panin, 2007; Panin et al., 2014]. As an enclosed
basin, CSL variation is controlled mainly by water inflow from rivers and precipitation, and loss from evapora-
tion and discharge to the Kara-Bogaz-Gol (KBG) Bay [Kosarev et al., 2009]. The entire Caspian Sea catchment
basin has an area of approximately 3.5 × 10
6
km
2
, almost 10 times that of the Caspian Sea (Figure 1) and
accounting for approximately 10% of the global area of closed basins. The length of the Caspian Sea
watershed from north to south is about 2500 km and from west to east is about 1000 km [Zonn and
Kostianoy, 2016]. This makes CSL particularly sensitive to climatic condition in the catchment. The Volga river
basin to the north comprises more than half the catchment, with an area of ~1.38 × 10
6
km
2
. With a relatively
wet climate, it contributes over 80% of total influx to the Caspian Sea [Arpe et al., 2000; Renssen et al., 2007;
Ozyavas et al., 2010; Roshan et al., 2012].
Long-term CSL fluctuations of several meters have significantly altered coastal ecosystems, especially in the
northern parts of the Caspian Sea (United Nations Environment Programme Report, 2011; available at http://
www.grida.no/publications/caspian-sea/), where average water depth (north of 45°N) is near 5 m (Figure 1).
For example, there were widespread coastal changes resulting from a CSL drop of about 3 m between the
early 1930s and 1970s. CSL was far more stable (typically half-meter fluctuations) in the preceding century,
which was largely preindustrial in this region (1840–1940). The rapid pace of change since the 1930s has been
attributed to Volga River discharge fluctuations following upstream rainfall reduction over catchment area
and dam constructions. However, it is unlikely that this is a full explanation for continued decreases from
1940s to late 1970s, nor of a rapid 2.5 m increase over two decades after 1977, which was absolutely
unexpected for scientific community.
While CSL change largely reflects a balance between inflow from rivers and precipitation and outflow from
surface evaporation, a detailed understanding of large multidecade CSL fluctuations is lacking. With both
CHEN ET AL. LONG-TERM CASPIAN SEA LEVEL CHANGE 6993
PUBLICATION
S
Geophysical Research Letters
RESEARCH LETTER
10.1002/2017GL073958
Key Points:
•PER flux-reconstructed Caspian Sea
level change agrees remarkably well
with tide gauge and satellite
measurements
•Increased evaporation rates over the
Caspian Sea play a dominant role in
reversing the Caspian Sea level trends
during the past 37 years
•The current Caspian Sea level decline
is expected to continue into the
foreseeable future, under global
warming scenarios
Supporting Information:
•Supporting Information S1
Correspondence to:
J. L. Chen,
chen@csr.utexas.edu
Citation:
Chen, J. L., T. Pekker, C. R. Wilson,
B. D. Tapley, A. G. Kostianoy,
J.-F. Cretaux, and E. S. Safarov (2017),
Long-term Caspian Sea level change,
Geophys. Res. Lett.,44, 6993–7001,
doi:10.1002/2017GL073958.
Received 20 APR 2017
Accepted 20 JUN 2017
Accepted article online 21 JUN 2017
Published online 12 JUL 2017
©2017. American Geophysical Union.
All Rights Reserved.

large-magnitude increase and decrease evident during the past several decades, prediction of CSL is
challenging. Long-term CSL trends predicted by different climate models show substantially large
discrepancy and are often contradictory with each other [e.g., Elguindi and Giorgi, 2006; Roshan et al.,
2012]. The main motivation of the present study is to develop an understanding of these past changes
through a comprehensive analysis of historical tide gauge measurements and more recent satellite
altimeter observations [Lebedev and Kostianoy, 2008; Cretaux et al., 2016; Chen et al., 2017], observed river
discharge (R), and climate model estimates of precipitation (P) and evaporation (E)fluxes. The goal is to
reconstruct past long-term CSL variations using integrated P,E, and Rfluxes and to lay the groundwork for
Figure 1. Map of the Caspian Sea and Caspian drainage (enclosed by the red contour line). The Caspian Sea is surrounded
by five countries: Russia, Kazakhstan, Turkmenistan, Iran, and Azerbaijan. Four tide gauge stations (1 = Makhachkala,
2 = Fort Shevchenko, 3 = Baku, and 4 = Turkmenbashi), from which the historical Caspian Sea level observation time series
is derived, are marked by magenta dots.
Geophysical Research Letters 10.1002/2017GL073958
CHEN ET AL. LONG-TERM CASPIAN SEA LEVEL CHANGE 6994

predictions of future CSL trends. We will focus on years 1979 through 2015, which span the common period
of available CSL observations and the needed P,E, and Rfluxes.
2. Data Processing
2.1. Caspian Sea Level Change Measurements
A historical record of CSL change from tide gauge measurements over the period 1840–2000 (with seasonal
and shorter time scale variations removed) is from a previous analysis [Kostianoy et al., 2014], and derived as
the average of observations at four tide gauge stations marked by magenta dots in Figure 1. To extend the
CSL series to the more recent period, we concatenate the tide gauge and satellite altimeter CSL change time
series [Lebedev and Kostianoy, 2008; Cretaux et al., 2016] to generate a CSL change time series covering the
period from 1840 to 2015 (Figure 2a). More information on the comparison between tide gauge and altimeter
CSL changes and concatenation of the two time series is provided in Figure S1 in the supporting information.
2.2. River Discharge Measurements
In situ monthly river discharge (runoff (R)) fluxes for the Volga River cover the period 1881 and 2015 and
represent the total runoff in the delta apex of the Volga river mouth [Kostianoy et al., 2014]. The original
Volga river discharge fluxes (in units m
3
/s) are converted into equivalent monthly CSL height change rates,
by evenly allocating the monthly integrated Volga discharge water volumes over the Caspian Sea (using
an area of ~371,000 km
2
). The Volga river is believed to account for ~80% of total discharge (or runoff) into
the Caspian Sea [Arpe et al., 2000; Ozyavas et al., 2010; Roshan et al., 2012]. However, considering year-to-year
fluctuations of climatic conditions in different river basins, the actual percentage of the Volga’s share would
be expected to fluctuate. We do not have the access to river discharge observations for other major rivers for
this period (1979–2015). In the present analysis, we use limited historical records for other major rivers from
the Global River Discharge (RivDIS) Project (http://daac.ornl.gov/RIVDIS/rivdis.shtml) [Vorosmarty et al., 1996]
to estimate potential river discharge rates for other major rivers including the Ural, Kura, Samur, Terek, and
Shafa Rivers. The temporal coverage of the RivDIS database ends in 1984, with various starting times for
different rivers.
Based on the available RivDIS river discharge data, over the period 1965–1984, Volga discharge accounts for
~90% of the total and is slightly higher (91.5%) for 1979–1984. In this study we use this percentage (91.5%) to
estimate total river discharge using only Volga observations for the period 1979–2003. After 2003 period we
adjust the percentage to 80% based on estimates from previous studies [Arpe et al., 2000; Ozyavas et al., 2010;
Roshan et al., 2012]. Details of computations and assumptions about other rivers’discharge contribution are
provided in Figure S2.
The use of different ratios of Volga river discharge over total discharge into the Caspian Sea over the two
periods (1979–2003 and 2004–2015) is somewhat arbitrary. We provide detailed analysis on how this will
affect the reconstructed CSL change over the studied period (1979–2015) in supporting information, using
four case studies: case 1—assuming Volga discharge accounts for 91.5% for period 1979–2003 and 80%
for period 2004–2015, case 2—assuming Volga discharge accounts for 91.5% for the entire period
1979–2015, case 3—assuming Volga discharge accounts for 80% for the entire period 1979–2015,and case
4—assuming Volga discharge accounts for 80% for the entire period 1979–2015, but with the PER (P-E+R)
mean removed. Case 1 (the case adopted in the present analysis) provides the best agreement between
constructed CSL change and CSL observation (see Figure S3 for details).
2.3. Precipitation and Evaporation Fluxes
Monthly mean Pand Efluxes over the Caspian Sea during the studied period (1979–2015) are computed
using model predictions from the National Centers for Environmental Prediction Climate Forecast System
(CFS) [Saha et al., 2014]. CFS is a fully coupled climate model, including components of the atmosphere,
ocean, sea ice, and land, and covers the period from 1979 to the present. CFS global Pand Eflux rates are
computed from two CFS runs. Over the period 1979–2010, monthly Pand Eflux rates are directly from the
CFS Reanalysis outputs (covering up to the end of 2010) [Saha et al., 2010], and over the period
2011–2015, monthly Pand Eflux rates are computed from integrating the 6-hourly Pand Eproducts from
the CFS version 2 (CFSv2) operational model [Saha et al., 2011] (hereafter CFS/CFSv2 is denoted as CFS).
We exclude the Kara-Bogaz-Gol Bay, when computing monthly Pand Efluxes over the Caspian Sea (more
Geophysical Research Letters 10.1002/2017GL073958
CHEN ET AL. LONG-TERM CASPIAN SEA LEVEL CHANGE 6995

Figure 2. (a) Monthly mean Caspian Sea level (CSL) changes observed by tide gauges (1940–1997) and satellite altimetry (1997 to 2015, provided by Legos/CNES
http://hydroweb.theia-land.fr/). A systematic bias between tide gauge and altimeter series is removed (using a 4 year overlapping period 1993–1996). (b) Observed
monthly mean CSL changes (blue curve), and predicted Caspian Sea level changes (red curve) from integrated monthly precipitation (P) and evaporation (E) from the
CFS climate model, and (R) from Volga River and estimated contributions from other rivers. An offset of ~2769 cm is added to the observed CSL time series for
plotting purposes (setting sea level to be zero at 1940.0). Seasonal and shorter time scale variations are removed from all time series.
Geophysical Research Letters 10.1002/2017GL073958
CHEN ET AL. LONG-TERM CASPIAN SEA LEVEL CHANGE 6996

information on CFS Pand Eflux and potential bias between the two CFS models is provided in section S3 in
the supporting information).
2.4. Reconstruction of Caspian Sea Level Change
CSL changes can be estimated by the following water mass balance equation:
dH
dt ¼PEþR(1)
where His the mean CSL and tis the time. P,E, and Rare the precipitation, evaporation, and river discharge
rates expressed in equivalent water height change over the Caspian Sea in a unit time. River discharge rates R,
measured in rates of volume change of water, are converted into equivalent CSL height changes by evenly
distributing discharge water over the Caspian Sea. Accordingly, His the integral of combined P,E, and Rfluxes
over time:
HtðÞ¼∫
t
t0
PEþRðÞdt (2)
where t
0
is the starting time, 1979.0 in the present study. When P,E, and Rfluxes are available (as introduced
in previous sections), the above equation (2) can help reconstruct a continuous record of CSL change.
Figure 3. (a) Comparison of accumulated yearly precipitation (P) and evaporation (E) from the CFS model and observed (R). The means over two different periods
(1979–1995 and 1996–2015) for each time series are marked by dashed lines, with mean values labeled in corresponding colors. (b) Yearly (P-E+R) budget for the
Caspian Sea, with means over two different periods (1979–1995 and 1996–2015) marked by dashed lines (and labeled).
Geophysical Research Letters 10.1002/2017GL073958
CHEN ET AL. LONG-TERM CASPIAN SEA LEVEL CHANGE 6997

Another way to estimate PER (P-E+R) contributions to CSL rates is to examine the accumulated yearly budget
(YB) of P,E, and Rfluxes. From equation (1), the yearly mean of (P-E+R) is an estimate of average CSL rate
(dH/dt) for that year. A PER YB analysis provides both an estimate of CSL rate, and a useful way to understand
individual contributions of P,E, and R.
3. Results
Using monthly Pand Efluxes (in cm/month) computed from the CFS model, and monthly Rfluxes (in
cm/month) based on observed Volga River discharge, with considerations of possible contributions from
other rivers in the Caspian drainage basin, we reconstruct a monthly CSL change time series the 37 year
period (1979–2015). We show in Figure 2b the comparison between tide gauge and satellite altimeter-
observed CSL changes with the reconstructed CSL series from PER flux integration (FI) over the studied
period. This is a period of rapid increase (~13.09 cm/yr over 1978–1995), followed by rapid decrease
(~6.72 cm/yr over 1996–2015) (see Figure 2a).
Figure 2b shows excellent agreement between the PER FI estimates and observed CSL, especially from 1979
to about 2010. Seasonal and shorter time scale variations have been removed from both time series. Over
1979–1995, rates of observed CSL and FI are ~12.74 cm/yr and 12.38 cm/yr, respectively. Despite some
differences, average CSL rates during the decreasing period (1996–2015) also agree remarkably well
(6.72 cm/yr versus 6.79 cm/yr). Similarly good agreements are also found for the period 2005–2015
(9.13 cm/yr versus 8.48 cm/yr). These rates are estimated via direct average (details of rate estimation
are provided in section S5).
The left plot of Figure 3 shows accumulated separate yearly rates of P,E, and Rover the period 1979–2015,
and mean values for two different periods (1979–1995 and 1996–2015). The right plot shows similar analysis
and comparisons for the PER YB. There are two distinctively different means (+12.25 and 6.90 cm/yr, respec-
tively) of the PER YB rates over the two different periods (1979–1995 and 1996–2015). These values agree very
well with observed CSL rates (+12.74 and 6.72 cm/yr, respectively) for the two periods. During the period
from 1994 to 1996, yearly PER flux over the Caspian Sea has dropped ~50 cm (in equivalent CSL), which is
dominantly attributed to the drop of about the same magnitude in R. Even though during the few years
Figure 4. Relative long-term Caspian Sea level change rates between the two periods 1996–2015 and 1979–1995 from tide
gauge and altimeter observations (the cyan bar), and contributions from accumulated yearly mean PER (P-E+R) budget
over the Caspian Sea (the stacked bar), with individual P,E, and Rcontributions represented by blue, red, and green bars,
respectively.
Geophysical Research Letters 10.1002/2017GL073958
CHEN ET AL. LONG-TERM CASPIAN SEA LEVEL CHANGE 6998

after 1996, the significant drop in yearly Rflux is mostly recovered, average combined PER flux remains
substantially lower.
Between the two periods (1979–1995 and 1996–2015), the relative mean CSL change rate is ~19.15 cm/yr
(from +12.25 to 6.90 cm/yr), with the largest contribution from an average increase in Eof about
8.82 cm/yr, accompanied by an average decrease in Pof about 5.13 cm/yr. In the same period, Rdropped
by about 5.20 cm/yr. Figure 4 summarizes the different contributions from P,E, and Ryearly fluxes to the
relative CSL rate using stacked bars. We list in Table 1 the details of the comparison, including estimated
CSL rates during the three studied periods (1979–1995, 1996–2015, and 2005–2015) from the PER YB
method (based on equation (1)).
4. Conclusions and Discussion
Using in situ river discharge (R) data and model-predicted precipitation and evaporation fluxes (Pand E), we
have reconstructed long-term CSL changes during the 37 year period 1979–2015. Both PER FI and PER
YB-estimated CSL rates agree remarkably well with tide gauge and altimeter observations (see Table 1).
Please note that the PER FI and YB methods should provide identical rate estimates when the time series
are long enough and the method for computing the FI rate is consistent with that for the YB (i.e., using direct
average; see section S5). While increased precipitation in the Volga drainage basin is believed to be the main
driving force of the significant CSL increase during the period 1979–1995, our analysis clearly shows that
increased evaporation rate over the Caspian Sea during the 37 year period has played the dominant role
in reversing the CSL trends and driving the current CSL decline, apparently exceeding effects of changes
in Volga River inflow (see Figure 4 and Table 1).
An increase in Eover the Caspian Sea is closely related to surface temperature increases (see Figure S6).
With continued warming in the northern hemisphere, one can expect yearly accumulated evaporation
rates over the Caspian Sea to continue increasing for the foreseeable future. Even without a long-term
increase in surface temperature, the current situation in which Eexceeds P+Rwould indicate likely
continued long-term CSL decline, although with superimposed interannual fluctuations of either sign.
Without an accompanying future increase in Pover either the Caspian Sea or the surrounding catchment,
CSL decline would be expected to continue. The history of the nearby Aral Sea over the past several
decades shows how long-term water flux imbalance altering the level of an enclosed lake can lead to
dramatic ecosystem consequences. While similar consequences in the Southern Caspian Sea are unlikely,
the shallow (~5 m depth) northern part of the sea and KBG Bay are much more vulnerable [Kosarev and
Kostianoy, 2005; Kosarev et al., 2009].
Uncertainty in the present analysis includes a lack of adequate in situ measurements of discharge from
major rivers, other than the Volga, and of outflow to the KBG Bay over this period (1979–2015). Our estimate
of contributions from other rivers using limited historical data is approximate, considering the different time
spans of the RivDIS data sets and locations of the stations. While accurate quantification of potential
Table 1. Average Caspian Sea Level Change Rates for Three Periods 1979–1995, 1996–2015, and 2005–2015 From
Observations, and Contributions From Mean-Accumulated Yearly Precipitation (P), Evaporation (E), and River Runoff (R),
and Total PER (P-E+R) Mean Yearly Budget (YB) Over the Caspian Sea
a
Caspian Sea Level Rates 1979.01–1995.12 (cm/yr) 1996.01–2015.12 (cm/yr) 2005.01–2015.12 (cm/yr)
Observations 12.74 6.72 9.13
Precipitation (P) 39.66 34.53 34.55
Evaporation (E) 107.73 116.55 117.55
P-E68.07 82.02 83.01
Volga Runoff (R) 74.02 65.23 62.21
Total Runoff (R
*
) 80.32 75.12 74.65
P-E+R
*
(YB) 12.25 6.90 8.36
P-E+R
*
(FI) 12.38 6.79 8.48
a
Separate estimates of Caspian Sea level rates based on monthly flux integration (FI) of PER are also included for
comparison. The total runoff (R
*
) is computed from the Volga River runoff data, based on the assumption that the
Volga River accounts for 91.5% of the total runoff over the period 1979–2003 and 80% for 2004–2015.
Geophysical Research Letters 10.1002/2017GL073958
CHEN ET AL. LONG-TERM CASPIAN SEA LEVEL CHANGE 6999

contribution of the outflow to the KBG Bay on long-term CSL change is difficult, limited data from previous
publications [Zonn and Kostianoy, 2016] appear to suggest that the contribution could be important.
Including this additional outflow in the analysis would further increase the predicted CSL decreasing rates
(since 1996). Varied assumptions concerning contributions from other rivers would change PER estimates
somewhat, but not the main conclusion concerning the dominant effect of increased evaporation rates.
Another uncertainty comes from using CFS model estimates of Pand E, instead of measurements.
However, CFS is an advanced model fully coupled with other components of the climate system (including
the oceans, land, and sea ice), and the excellent agreement between CSL observations and integrated PER
predictions appears to have validated CFS estimates of Pand Efluxes.
We carried out similar calculations using Pand Eestimates from two other climate models, the European
Center for Medium-Range Weather Forecasting ERA-Interim atmospheric reanalysis (ECM) [Dee et al., 2011]
and the Japanese 55-year reanalysis (JRA) [Kobayashi et al., 2015]. Both models appear to significantly under-
estimate P-Eflux over the Caspian Sea, leading to significantly larger and unrealistic CSL increasing trends. It is
impossible to close the long-term CSL budget using ECM or JRA P-Efluxes, even assuming that other rivers
make no contribution to R. Further analysis shows that mean atmospheric surface (air2m) temperatures over
the Caspian Sea from the ECM model is systematically lower than those from the CFS, except for over the last
5 years (2011–2015) when the CFSv2 operational model is used (which may be due to potential bias between
CFS reanalysis and operational models). The consistently lower mean surface (air2m) temperature from ECM
is likely the leading cause to the underestimation of Eand P-Efluxes. Detailed analysis and comparison of the
CFS and ECM P,E, and P-Efluxes is provided in section S3.
We note that a previous study [Arpe et al., 2014] found good agreement between observed CSL change and
ECM P-Eflux estimates, plus Volga River R. In that study average annual PER flux was removed, including its
mean value. The effect of this after integration over time would be to exclude linear trends, which are a domi-
nant feature at longer time scales. Removing PER flux means (or using PER flux anomalies) appears to be a
valid approach when focusing in nonlinear CSL changes [Arpe et al., 2014]. If the PER flux mean was not
removed, the PER-predicted, long-term CSL changes (based on ECM P-E, plus Volga River R) would be signif-
icantly larger than those observed. Additional analyses and discussions of PER-reconstructed, long-term CSL
changes using three different climate models (CFS, ECM, and JRA) and how systematic biases in climate
model-predicted P-Efluxes and different treatments of the seasonal means of the PER fluxes would affect
the reconstructed CSL change are provided in Figure S7.
Although the agreement between CSL observations and PER predictions is remarkable in general over the
studied period, the discrepancy is notably larger since 2010, especially during around 2011 and 2012.
While the exact cause for this large discrepancy is unknown, it can be related to potential large bias in the
Pand Eflux estimates from the CFS operational model (see Figure S5), and river discharge from other rivers.
The scaled-up estimation of total river discharge using Volga’s data cannot accurately quantify other rivers’
contribution, considering regional climate conditions can be very different between the Volga and other
drainage basins.
Due to the enclosed nature of the Caspian Sea, CSL change is particularly sensitive to any imbalance between
the inflow (Pand R) and outflow (E). Long-term CSL change is mainly controlled by extended period of
imbalanced PER flux. Over the past decade, increased evaporation rates over the Caspian Sea associated with
increased surface air temperature and other changing climate factors (such surface humidity and wind)
cannot be balanced by the precipitation and river discharge, leading to significant CSL drop. Based on the
same CFS model estimates, during the period 1979–2015, yearly accumulated Pover the Volga drainage
basin also shows a clear decreasing trend (detailed analysis not shown here), consistent with the decrease
in observed Rflux (see Figure 3a).
We neglected groundwater flux in the water balance analysis, but this contribution is estimated to be
small [Zekster, 1996], and is probably negligible, given other uncertainties. Groundwater contribution is
expected to partly compensate the impact from the outflow to the KBG. While satellite altimeter CSL mea-
surements are available with a measure of uncertainty [Cretaux et al., 2016], none is available for other data
(tide gauge CSL, CFS Pand Eestimates, and Volga River observations of R). Further analysis of uncertainty
would be difficult and unlikely to affect the main conclusion concerning the importance of increased
evaporation rates.
Geophysical Research Letters 10.1002/2017GL073958
CHEN ET AL. LONG-TERM CASPIAN SEA LEVEL CHANGE 7000

References
Arpe, K., L. Bengtsson, G. S. Golitsyn, I. I. Mokhov, V. A. Semenov, and P. V. Sporyshev (200 0), Connection between Caspian Sea level variability
and ENSO, Geophys. Res. Lett.,27, 2693–2696, doi:10.1029/1999GL002374.
Arpe, K., S. A. G. Leroy, F. Wetterhall, V. Khan, S. Hagemann, and H. Lahijani (2014), Prediction of the Caspian Sea level using ECMWF seasonal
forecasts and reanalysis, Theor. Appl. Climatol.,117,41–60, doi:10.1007/s00704-013-0937-6.
Cazanave, A, P. Bonnefond, K. Dominh, and P. Shaeffer (1997), Caspian sea level from TOPEX-Poseidon altimetry: Level now falling, Geophys.
Res. Lett.,48, 881–884, doi:10.1029/97GL00809.
Chen, J. L., C. R. Wilson, B. D. Tapley, H. Save, and J.-F. Cretaux (2017), Long-term and seasonal Caspian Sea level change from satellite gravity
and altimeter measurements, J. Geophys. Res. Solid Earth,122, 2274–2290, doi:10.1002/2016JB013595.
Cretaux, J.-F., R. Abarca-del-Río, M. Bergé-Nguyen, A. Arsen, V. Drolon, G. Clos, and P. Maisongrande (2016), Lake volume monitoring from
space, Surv. Geophys.,37, 269–305, doi:10.1007/s10712-016-9362-6.
Dee, D. P., et al. (2011), The ERA-Interim reanalysis: Configuration and performance of the data assimilation system, Q. J. R. Meteo rol. Soc.,137,
553–597, doi:10.1002/qj.828.
Elguindi, N., and F. Giorgi (2006), Projected changes in the Caspian Sea level for the 21st century based on the latest AOGCM simulations,
Geophys. Res. Lett.,33, L08706, doi:10.1029/2006GL025943.
Kobayashi, S., et al. (2015), The JRA-55 reanalysis: General specifications and basic characteristics, J. Meteorol. Soc. Jpn.,93(1), 5–48,
doi:10.2151/jmsj.2015-001.
Kosarev, A. N., and A. G. Kostianoy (2005), Kara-Bogaz-Gol Bay, in The Caspian Sea Environment, edited by A. G. Kostianoy and A. N. Kosarev,
pp. 211–221, Springer, Berlin.
Kosarev, A. N., and E. A. Yablonskaya (1994), The Caspian Sea, p. 259, Academic, The Hague.
Kosarev, A. N., A. G. Kostianoy, and I. S. Zonn (2009), Kara-Bogaz-Gol Bay: Physical and chemical evolution, in Aquatic Geochemistry,Special
Issue: Saline Lakes and Global Change, vol. 15, N 1-2, pp. 223–236, Springer Science+Business Media, Berlin, Germany, doi:10.1007/s10498-
008-9054-z.
Kostianoy, A. G., A. I. Ginzburg, S. A. Lebedev, and N. A. Sheremet (2014), The Southern seas of Russia, in Second Assessment Report of
Roshydromet on Climate Change and Its Consequences on Territory of Russian Federation [in Russian], edited by V. M. Kattsov and
S. M. Semenov, pp. 644–683, Inst. of Global Clim. and Ecol., Moscow.
Lebedev, S. A., and A. G. Kostianoy (2008), Integrated using of satellite altimetry in investigation of meteorological, hydrological and
hydrodynamic regime of the Caspian Sea, J. Terr. Atmos. Ocean. Sci.,19(1–2), 71–82.
Ozyavas, A., S. D. Khan, and J. F. Casey (2010), A possible connection of Caspian Sea level fluctuations with meteorological factors and
seismicity, Earth Planet. Sci. Lett.,299, 150–158, doi:10.1016/j.epsl.2010.08.030.
Panin, G. N. (2007), Caspian Sea level fluctuations as a consequence of regional climatic change, in Global Change: Enough Water for All?,
edited by J. L. Lozán, et al., pp. 216–219, Wissenschaftliche Auswertungen, Hamburg.
Panin, G. N., I. V. Solomonova, and T. Yu. Vyruchalkina (2014), Regime of water balance components of the Caspian Sea, Water Resour.,41(5),
505–511, doi:10.1134/S0097807814050078.
Roshan, G., M. Moghbel, and S. Grab (2012), Modeling Caspian Sea water level oscillations under different scenarios of increasing
atmospheric carbon dioxide concentrations, Iran. J. Environ. Health Sci. Eng.,9, 24. [Available at http://www.ijehse.com/content/9/1/24.]
Renssen, H. B. C. Lougheed, J. C. J. H. Aerts, H. De Moel, P. J. Ward, and J. C. J. Kwadijk (2007), Simul ating long-term Caspian Sea level changes:
The impact of Holocene and future climate conditions, Earth Planet. Sci. Lett.,261(3), 685–693, doi:10.1016/j.epsl.2007.07.037.
Saha, S., et al. (2010), NCEP Climate Forecast System Reanalysis (CFSR) monthly products, January 1979 to December 2010. Research Data
Archive at the National Center for Atmospheric Research, Computational and Information Systems Laboratory, doi:10.5065/D6DN438J.
Saha, S., et al. (2011), NCEP Climate Forecast System version 2 (CFSv2) 6-hourly products. Research Data Archive at the National Center for
Atmospheric Research, Computational and Information Systems Laboratory, doi:10.5065/D61C1TXF.
Saha, S., et al. (2014), The NCEP Climate Forecast System version 2, J. Clim.,27, 2185–2208, doi:10.1175/JCLI-D-12-00823.1.
Vorosmarty, C. J., B. M. Fekete, and B. A. Tucker (1996), River Discharge Database , Version 1.0 (RivDIS v1.0), Volumes 0 through 6. A contribution
to IHP-V Theme 1. Technical Documents in Hydrology Series, U.N. Educ., Sci. and Cult. Organ., Paris, France.
Zekster, I. S. (1996), Groundwater discharge into lakes: A review of recent studies with particular regard to large saline lakes in central Asia,
Int. J. Salt Lake Res. 4, 233–249.
Zonn, I. S., and A. G. Kostianoy (2016), Chapter 124: The Caspian Sea basin, in Handbook of Applied Hydrology, edited by Vijay P. Singh,
2nd ed., p. 124-1–124-7, McGraw-Hill Education, Columbus, Ohio.
Geophysical Research Letters 10.1002/2017GL073958
CHEN ET AL. LONG-TERM CASPIAN SEA LEVEL CHANGE 7001
Acknowledgments
The authors are grateful to the two
anonymous reviewer and Anny
Cazenave for their comprehensive and
insightful comments, which have led to
improved presentation of the results.
J.L.C., P.T., C.R.W., and B.D.T. were
supported by the NASA GRACE and
GRACE Follow-On Projects (under
contract NNL14AA00C and JPL
subcontract 1478584), NASA ESI
Program (NNX12AM86G and
NNX17AG96G), and NASA GRACE
Science Team Program (NNX12AJ97G),
and A.G.K. was supported by the
Russian Science Foundation grant
14-50-00095. Related climate model
estimates of precipitation and
evaporation flux data are available
online from UCAR/NCAR’s Research
Data Archive (https://doi.org/10.5065/
D6DN438J and https://doi.org/10.5065/
D61C1TXF). Satellite altimeter
observations of Caspian Sea level
changes are provided by Legos/CNES
and available online at http://hydroweb.
theia-land.fr. The RivDIS river discharge
data are available online at http://daac.
ornl.gov/RIVDIS/rivdis.shtml. Historical
tide gauge measurements of Caspian
Sea level change and Volga river
discharge data (after 1984) are not
available to the public and provided to
this study through collaborations.