Linkage between Three Gorges Dam impacts and the dramatic recessions in China’s largest freshwater lake, Poyang Lake

Despite comprising a small portion of the earth’s surface, lakes are vitally important for global ecosystem cycling. However, lake systems worldwide are extremely fragile, and many are shrinking due to changing climate and anthropogenic activities. Here, we show that Poyang Lake, the largest freshwater lake in China, has experienced a dramatic and prolonged recession, which began in late September of 2003. We further demonstrate that abnormally low levels appear during October, 28 days ahead of the normal initiation of the dry season, which greatly imperiled the lake’s wetland areas and function as an ecosystem for wintering waterbirds. An increase in the river-lake water level gradient induced by the Three Gorges Dam (TGD) altered the lake balance by inducing greater discharge into the Changjiang River, which is probably responsible for the current lake shrinkage. Occasional episodes of arid climate, as well as local sand mining, will aggravate the lake recession crisis. Although impacts of TGD on the Poyang Lake recession can be overruled by episodic extreme droughts, we argue that the average contributions of precipitation variation, human activities in the Poyang Lake catchment and TGD regulation to the Poyang Lake recession can be quantified as 39.1%, 4.6% and 56.3%, respectively.

and sediment load from the Ganjiang, Fuhe and Xinjiang Rivers are measured at the main stream, while measurements for the Xiushui and Raohe Rivers are recorded at upstream tributaries (Table S1).

Grouped frequency distribution
Grouped frequency distribution is used to detect the statistical characteristics of a large number of continuous variables. The technique, groups values into intervals according to their amplitude and assigns each interval to a frequency. The grouped frequency can be expressed as either relative frequency or relative cumulative frequency.
The relative frequency is defined as: where ni is the number of observations that occur in a certain class, and n is the total number of observations.
The relative cumulative frequency is set as the quotient between the sum of all of the classes that are less than or equal to the one under consideration and the total number of observations. In this study, grouped frequency distribution is used to generate the relative frequencies and relative cumulative frequencies of daily water levels at Duchang and Hukou ( Fig. 2; Fig. 3; Fig. S3). Water level at Duchang is categorized into 8 groups with an interval of 2 m while water levels at Hukou are classified into 9 groups with the same interval.

Sen's slope estimator
Sen's slope estimator accounts for the magnitude of trend in a sample of N pairs of data through a non-parametric procedure 1 : Where xj and xk are measurements at times j and k (j>k), respectively. The median of Sen's slope estimator is computed by: The sign of Qmed reflects the direction of the trend while the value of Qmed shows the steepness of the trend. To test whether the median slope is significantly different than 0, Qmed is tested with a two-sided test at the 100(1-α) % confidence interval 2 . In this paper, Sen's slope test is adopted to detect whether the time series of water level differences between pre-and post-TGD periods at Duchang from 1 st September to 31 st November has a significant tendency, as well as the true slope of the trend (Fig.S2 a).

Standard normal homogeneity test
The standard normal homogeneity test (SNHT) identifies the occurrence of abrupt change by comparing the mean of the first a years of the record with that of the last na years based on the statistic T0 3 : a a n a n Y is the mean value of the data set, s is the standard deviation of the data set, and n is the length of the data set.
When T0 approaches the maximum value at year a=A, a significant shift occurs if T0 is larger than the critical value at a given significance level. Here, SNHT is applied to the time series of water level for the same day of every year at Duchang from 1960-2012 to identify the occurrence of abrupt changes (Fig.S2b).

Water level estimation
When calculating the variable of inflow-induced water level variation, we assumed the Poyang Lake as a frustum (Fig. S15). The formula for a frustum volume is: where S ' is the area of the upper base, S is the area of the lower base, and H is the height.
Because the volume and the upper base area of the frustum are known, the area of the bottom base can be estimated as a function of H, and H can then be determined by solving Equation S6.

Definition of extreme drought events
The low-frequency of rare events is often defined by a particular high or low quantile 4 .
In this study, monthly October precipitation over the Poyang Lake during 1960-2012 is categorized on the basis of the 5% quantile. Extreme drought conditions correspond to the precipitation below the 5% quantile. Accordingly, 2004 is determined as an extreme drought year over the Poyang Lake basin.

Poyang Lake basin and TGD regulation on Poyang Lake recession
The variation of Poyang Lake's water level between pre-and post-TGD stages can be expressed via the linear equation: where hp, he, ha, hT, and hg represent the variation of lake water level in response to precipitation variation, evapotranspiration variation, anthropogenic activities in the Poyang Lake basin ,TGD regulation and groundwater variation, respectively. The effect of evapotranspiration variation, groundwater variation and regional water consumption variation on lake level decline are ignored firstly due to their limited effect, and then are further discussed as follows to avoid possible uncertainties.
Anthropogenic activities. Human activities in the Poyang Lake basin are mainly sand mining and dam operation.
Sand from Poyang Lake was exported at a rate of 236 million m 3 pear year from 2001-2008 5 . Averaging this amount of sand over the total lake area of 3000 km 2 , the entire lake level will be almost 0.1 m lower in average with respect to pre-TGD period, which explains 4.4% of lake level decline. We use average rather than accumulative variable to represent the effect of sand mining on lake level variation in this study for the following reasons: 1) Sand mining over the Poyang Lake primarily created local deep pools and pits, which usually are lower than the lake bed datum (Fig. S9). The stored water in these scars thereof are likely mainly been recharged by groundwater and partly been filled by lake surface water, which would not generate entire lake decline; 2) The trapped water in the previous old sand scars are dead water, which rarely been measured or participate in the water cycle of the Poyang Lake even in low lake level scenario (Fig.   S9). Therefore, the lake surface water only partly been affected by the new added scores that are created within the current year. It's probably that we overestimate the contribution of sand mining to water level reduction of Poyang Lake in this study.
Further investigation related to the number of new scars in each year and the incised volume of each scar is needed to accurately analyze the relationship between stored water and lake level.
The Poyang lake catchment is controlled by 25 large and 211 medium sized reservoirs (Table S2), which significantly affect sediment input to the lake region.
Annual gross sediment loss due to upstream reservoir operation is 104310 4 t during 2003-2012, which corresponds to a 0.004 m lowering every year and accounts for 0.2% of lake level decline.
Regional water consumption is another possible factor affecting the lake level variation. Poyang Lake, covering 97% of the area of Jiangxi Province, is the main water resource supplier of Jiangxi Province 6 . Over 90% of water resources of Jiangxi Province comes from the Poyang Lake (Poyang Lake region, 20.4%; Poyang Lake basin, 70.6%) 7 . Therefore, the water consumption of Jiangxi Province can represent the situation of water use over the Poyang Lake catchment. The yearly water consumption, including industry, agriculture and urban use of Jiangxi Province during 2000-2012, is shown in Fig. S8. It is found that water use of Jiangxi Province increases by 15.8×10 8 m 3 from 211.1×10 8 m 3 in pre-TGD period to 226.9×10 8 m 3 in post-TGD period along with the population growth and economic development. Among them, 45% are unreturned water [8][9] . Since the highest water use occurs during July to September in Jiangxi Province, this study estimates the water use variation in October by the average variable: divide the yearly increased unreturned water by 12. Accordingly, increased regional water consumption leads to a 0.032 m water level reduction in October, among which 0.007 m is caused by lake region water use, 0.025 m is caused by lake basin water use.
In short, the contribution of human activities to the water level reduction of the Poyang Lake is 4.6%, which increases to 6% when regional water consumption variation is considered.

Climate change.
Changes in precipitation is the most critical factor determining the impact of climate change. Precipitation affects the Poyang Lake catchment through two ways. Precipitation over the lake region affects lake level directly. Precipitation over the lake basin affects the lake via received inflow. Relative to the pre-TGD stage, precipitation over the lake region decreased by 0.015 m, which equals 0.7% of total lake level decrease. The contribution of lake catchment inflow decreased by 14. Lake, its influence is ignored in this study.
Therefore, the contribution of precipitation on Poyang Lake shrinkage is 39.1%, which decreased to 37.8% when regional water use and evaporation are taken into account.
Three Gorges Dam regulation. TGD-induced downstream riverbed erosion increases the topographic gradient between the Poyang Lake and the Changjiang River, which increases the lake's outflow ability, discharging a huge amount of water into the river.
Moreover, the filling of reservoirs from September to early November further heightens the lake-river gradient and lake outflow. Because climate change, human activities in the lake basin and TGD operation are the three major effects on the water level variation of the Poyang Lake, the influence of the TGD on lake recession can be quantified as the difference between Δh and the sum of hp and ha. Therefore, the TGD is responsible for 1.26 m and 56.3% of lake recession in October. We note, though, that during extreme drought conditions, the impact of TGD regulation on Poyang Lake recession can be smaller than the effects of climate forcing. Moreover, when the water level decreases dramatically, groundwater may participate in the water interaction between the Poyang Lake and the Changjiang River to keep the lake water balance 12 .Baseflow, as the net flow from groundwater storage to a stream, is a common way to estimate the groundwater characteristic 13  All in all, TGD induced downstream riverbed erosion and water storage explains 56.3% of the Poyang Lake recession. However, this contribution decreases to 54.9% in view of the effect of groundwater, water use and evapotranspiration.   Note: Lake level is measured at Duchang station. Δh is water level variation; 1 is precipitation variation; 2 is anthropogenic activities; and 3 is TGD regulation. Note: 1 is water use variation; 2 is precipitation variation; 3 is anthropogenic activities; 4 is groundwater variation and 5 is TGD regulation.          Although there is a slight decrease trend in water level reduction from lake outlet to the upper reaches, water level reduction in the three station are still comparable. Moreover, Duchang station is locate in the central area of the Poyang Lake. Therefore, water level reduction of 2.24 m at Duchang station is selected to represent the average water level reduction in the whole lake region. The figure was created with Matlab R2009b.