Potential impact of water transfer policy implementation on lake eutrophication on the Shandong Peninsula: a difference-in-differences approach

Traditional research on lake eutrophication has failed to consider the effect of the South-to-North Water Transfer Project (SNWTP) policy; thus, the difference-in-differences (DID) model, which is usually applied to economic factors, was innovatively introduced to evaluate the effect of such policies on lake eutrophication. Nansi Lake and Dongping Lake in the Shandong Peninsula were selected as the experimental group, and Daming Lake and Mata Lake were selected as the control group. The eutrophication indices of the experimental group and the control group were calculated by the measured chlorophyll-a, total phosphorus, total nitrogen, water transparency and chemical oxygen demand data and used as the explanatory variables of the DID model. Nine environmental and socio-economic factors, such as dissolved oxygen and rural population, were selected as the control variables of the DID model to analyze the impact of the SNWTP policy on lake eutrophication. A joint consideration of environmental and socio-economic factors showed that the eutrophication degree of the experimental lakes deteriorated by 7.10% compared with the control under the influence of the implemented policy. Dissolved oxygen is the main factor affecting the eutrophication of the Shandong Peninsula. This study verifies that the DID model has the potential for use in quantitative analyses of the effect of the SNWTP policy on lake eutrophication.


INTRODUCTION
China's South-to-North Water Transfer Project (SNWTP) is of great strategic significance for alleviating water shortages, improving the ecological environment of the water demand area and promoting sustainable economic and social development. China has invested approximately $20 billion and resettled >300,000 people to construct the pipeline, and the SNWTP has become the largest and most expensive interbasin water transfer mega-project in the world (Resources ; Li et al. ). The final water volume transferred is expected to reach 44.8 billion m 3 /year by 2050 when the eastern, central and western routes are fully implemented (Zhuang et al. ). The first phase of the eastern route of the SNWTP (SNWTP-ER) was successfully completed and has been operational since late 2015 (Li et al. ). A total of more than 3 × 10 10 m 3 of water has been transferred from the lower Yangtze River in Yangzhou city, Jiangsu Province, and crossed the Huai River to the water shortage areas in the Yellow River basin (Guo et al. a, b). When completed, the SNWTP-ER will consist of 1,156 km of canals and 54 pumping stations designed to lift water up to 65 m over the Yellow River (Wang et al. ). The pumped water will be diverted from the south to the north, primarily through the existing Grand Canal, and impound in a chain of natural lakes, as regulating reservoirs, namely Gao-Bao-Shaobo Lake (GBSL), Hongze Lake (HZL), Luoma Lake (LML), Nansi Lake (NSL) and Dongping Lake (DPL) (Wang et al. ; Zhang ; Wu et al. ; Guo et al. , a).
Any interbasin water transfer project causes complex physical, chemical, hydrological and biological changes to the receiving system (Zeng et al. ; Yang et al. ; Yao et al. a, b, ; Yinglan et al. a). The water and sediment quality of the SNWTP-ER is of particular concern due to its large contribution to the total volume of transferred water, and it is also one of the most potentially polluted routes given its proximity to urban and industrial activities (Wu et al. ). With the vigorous development of the manufacturing industry in the Yangtze River Delta and the Bohai Rim, large amounts of untreated industrial wastewater are directly discharged to the lakes and rivers along the eastern route, and agricultural production has led to augmented fertilizer application, which has substantially increased the loadings of nutrients (including nitrogen and phosphorus) and organic matter into river streams, thereby deteriorating water quality Eutrophication has been recognized as the primary water quality issue for most of the lake ecosystems in the world (Smith & Schindler ; Wang et al. ), and nitrogen and phosphorus are the primary reasons for algal blooms caused by excess nutrients (Diersing ; Fang et al. ). Transferred water with a high content of nutrients has detrimental effects in the receiving water system, such as algal blooms and decreased dissolved oxygen (DO) (Barrow ; Zeng et al. ). Although a water transfer project can improve the water quality, it might also increase algal growth, and nutrient-rich water pumped from upstream causes cyanobacteria blooms in the receiving reservoir (Davies et al. ; Zhang et al. ). Previous studies on the SNWTP-ER have not determined whether it could control algal blooms or contribute to eutrophication in the water shortage period (Wu et al. ). This research generally examines the influence of the SNWTP on environmental factors (total nitrogen and total phosphorus) and hydrological factors (water quantity flow rate) directly based on water quality status and monitoring data (Guo et al. b; Rogers et al. ; Xu et al. ).
However, to the best of our knowledge, few studies have considered the impact of SNWTP policy implementation hidden behind the monitoring data. In other words, the possibility of eutrophication in the receiving reservoirs without policy implementation should be comparatively analyzed when considering the impact of the SNWTP policy. If only the effects of potential changes of the receiving reservoirs on eutrophication are compared in parallel before and after SNWTP policy implementation, then the impact of policy implementation itself could be largely ignored and the socio-economic and environmental effects of SNWTP operation policy can be investigated (Rogers et al. ). This study aims to establish an approach for determining the influence of water transfer policy on receiving reservoir eutrophication. Nansi Lake and Dongping Lake were selected as the objects of the transferred water policy, and Daming Lake and Mata Lake (without water transferred) were treated as controls. Nansi Lake and Dongping Lake are the largest freshwater shallow lakes in the Shandong Peninsula and play a vital role in absorbing pollutants and storing water (Zhuang et al. ). The difference-in-differences (DID) model, a common policy impact model used in economics, was introduced to establish a comprehensive modeling approach for evaluating the impacts of SNWTP policy implementation on eutrophication in Nansi and Dongping Lakes and analyzing the key factors and processes driving the underlying mechanisms.
The objectives of the study are to: (1) unravel the spatial and temporal distribution of lake eutrophication in Nansi Lake, Dongping Lake, Daming Lake and Mata Lake; (2) use the SNWTP policy as a single factor or combine environmental and socio-economic factors as a composite factor to establish the DID model for evaluating variation trends in eutrophication on water transfer lakes; (3) apply parallel trend and robustness analyses to validate the feasibility of the DID model; and (4) identify the primary driving factors that affect eutrophication and supply policy recommendations to develop management strategies for water quality safety and pollution control.

Study area
Nansi Lake (34 27 0 -35 20 0 N, 116 34 0 -117 21 0 E) is located in Weishan County, southwest Shandong Peninsula, China ( Figure 1). The lake, which is approximately 126 km long from south to north and 5-25 km wide from east to west, has an area of 1,266 km 2 and a total storage volume of 6.37 × 10 10 m 3 and is the first largest freshwater lake in the Shandong Peninsula along the eastern route. Nansi Lake consists of Nanyang Lake, Dushan Lake, Zhaoyang Lake and Weishan Lake without physical boundaries defining each lake, and it was divided into the upper and lower sections by the Erji Dam Pumping Station Hinge Project in 1960. The upper lake lies on the northern side of the Erji Dam, and the lower lake is located on the southern side.
Five state-controlled monitoring sections were located in Nansi Lake, namely the Qianbaikou (S1) and Nanyang   Since the implementation of the SNWDP-ER policy in 2015, Nansi Lake and Dongping Lake have served as the water-supplying lakes and impounded reservoirs; however, maintaining good water quality while meeting water demands remains a great challenge (Grant et al. ). In recent years, Nansi Lake is surrounded by the dense industrial and population zones, which has resulted in a large amount of industrial and domestic sewage discharged into Nansi Lake each year, thus exacerbating water pollution and eutrophication (Li ; Yao et al. a, b). Daming Lake is located in Jinan City and Mata Lake is located in Zibo City, and both lakes were identified as key nature reserves of Shandong Peninsula, China ( Figure 1).
The Lixiating (S9) and Mata Lake (S10) state-controlled monitoring sections are distributed in Daming Lake and Mata Lake, respectively. The SNWTP-ER policy has not been implemented on Daming Lake and Mata Lake, and the spatial location and characteristics are adjacent to Nansi Lake and Dongping Lake. Reducing the difference between the experimental groups and the control groups is expected to improve the accuracy of the DID model. In this study, considering the comprehensiveness, integrity and availability of the data, among the 13 lakes and reservoirs in the Shandong Peninsula, Daming Lake and Mata Lake were treated as the control group.
Basic principle of the DID model DID estimation has become an increasingly popular method of estimating the econometric evaluation of the implementation effect of projects or public policies. The notable superiority of DID estimation is derived from its simplicity as well as its potential to circumvent many of the endogenous problems that typically arise when making comparisons between heterogeneous individuals (Meyer ; Bertrand et al. ). In the assessment, the policy experimental group and the control group generally do not have complete randomness in sample allocation. The experiment involving a nonrandom allocation policy experimental group and control group is known as a natural trial, and its important feature is that systematic differences might occur between the experimental group and the control group prior to the implementation of the experiment. If the initial difference is ignored and only a horizontal comparison between the experimental group and the control group is performed after implementing the experiment, the estimated experimental effect is likely to be biased due to the mixed effect of the initial difference. The DID model was first introduced in 1985 to solve this problem (Ashenfelter & Card ), and since then, increasing attention has been focused on applying the model.

Setting and verification of the DID model in lake eutrophication research
Since the route of the SNWTP-ER was opened to water supply, its impact on lake eutrophication in Shandong Province might be due to the policy effect of the project and the time effect of the time trend changes in lake water eutrophication. The question of how to analyze the policy effect for correctly evaluating the impact of the route of the SNWTP-ER on lake eutrophication in Shandong Province is highly important. The DID model can effectively analyze and objectively evaluate the policy effect. In summary, we apply the DID model to research the change of lake eutrophication in Shandong Province before and after the route of the SNWTP-ER.
Using trophic level indices (TLIs) as the explained variable, 'treated' ¼ 1 indicates the water inflow lake of the section in the line of the SNWTP-ER and 'treated' ¼ 0 indicates that the lake where the section is located has no water diversion of the SNWTP-ER line. Time is used to express the 'time'. The value of the year when the SNWTP-ER is open to water and the following years is 1; otherwise, the value is 0. We use 'did' to represent the implementation effect of the SNWTP-ER, i.e., the intersection of 'treated' and 'time'. Additionally, x it is the control variable, which includes the time fixed effect and regional fixed effect, and θ it represents the constant term and disturbance term. The basic measurement model is shown as follows: where β 0 represents the common initial eutrophication mean value of all sections before the SNWTP-ER, β 1 represents the policy effect of the SNWTP-ER after controlling the initial eutrophication difference and common trend, .
where W j is the correlative weight for the TLI of j, TLI (j) is the TLI of j, TLI Environmental factors and socio-economic factors were selected as the control variables, and nine control variable  Quantitative and qualitative analysis of the impact of

SNWTP-ER policy on lake eutrophication
An analysis of the temporal and spatial distribution shows that the eutrophication degree could be affected by the SNWTP-ER policy. However, it is impossible to quantitatively and qualitatively analyze the impact of the policy.
Therefore, the DID model was introduced to estimate the potential influence of the SNWTP-ER policy on the lake eutrophication degree in the Shandong Peninsula.
The results are shown in Table 2  does not include fixed time and regional effects, and the results showed that the relationship is rather poor, with R 2 ¼ 0.30 and a P-value at only 10% significance. Model 2 includes fixed time and regional effects, and the SNWTP-ER policy effect of lake eutrophication is significant, with the P-value passing the 5% level test, and the fitting degree of the relationship has slightly improved. The β 1 value of Model 2 indicates that when considering only the influence of the SNWTP-ER policy, the eutrophication degree of the experimental lakes deteriorates by 6.20% compared with that of the control lakes.
Models 3 and 4 added the environmental control variables shown in Table 2, such as T w , DO, HS and N:P ratio. There are no fixed time and regional effects in Model 3, and the significance of the P-value is only 10%.
After fixing the time and regional effects in Model 4, the relationship between the TLIs and policy effects and environmental factors is significant, with R 2 ¼ 0.70, and the policy effect is significantly improved at the 1% level.
The quantitative and qualitative analysis results of Model 4 show that the β 1 value is positive at 5.89, which indicates that the eutrophication degree of the experimental lakes is significantly increased by 5.89% compared with the control  Models 5-7 add GDP, GOVA, GOVAH, RP and NIT as socio-economic control variables. Without the fixed time and regional effects, the significance of Model 5 is at the 1% level and the relationship coefficient R 2 is 0.60. However, Model 7 includes fixed time and regional effects, and the significance of the policy effect decreased to the 5% level, with R 2 below 0.5. This result could occur because the selected control variables are not comprehensive without the fixed time and regional effects. The value of β 1 in Model 7 is similar to that of Model 4, which indicates that the degree to which eutrophication is affected by socio-economic factors and environmental factors does not significantly differ. In contrast, as shown in Table 2 As shown in Table 2 year after the policy is implemented. This result indicates that the parallel trends assumption outlined above can be evaluated using a regression model in this study and that the policy effect appears after the SNWTP policy is implemented, with apparent changes over time.
To verify the robustness of the DID model, the interpreted variables could be redefined for the DID regression.
TP was replaced in the interpreted variables, and the policy cut-off point was still 2015. The robustness analysis results listed in Table 3 suggest that the core variables passed the significance test with a P-value of 5 or 1%, which is consistent with the regression results in Table 2.  Table 4 show that the TLIs are significantly correlated with DO, HS, GOVA, GOVAH and NIT, with the strongest correlation between GOVAH and NIT, which is not consistent with the conclusion of the DID model. Moreover, HS, GOVAH, GOVA and NIT were negatively correlated with the TLIs, indicating that when the Pearson's analysis did The results showed that the relationship between the TLIs and five control variables was significant, with P ¼ 0.00 < 0.05; however, the correlation coefficient R 2 ¼ 0.51 was not better than that of the DID Model 11, with R 2 ¼ 0.80. In addition, the N:P ratio and RP did not pass the t-test of significance at P-values greater than 0.05, although DO is still treated as a dominant factor affecting water quality, which is consistent with the DID model and the Pearson correlation analysis. Moreover, the DID model could be used as an alternative method to evaluate and quantify the impact of SNWTP policy on lake water quality in Shandong Province. The advantage of the DID model is that it could reveal the extent of policy impacts hidden in the experimental monitoring data. In addition, the DID model offers a diversified perspective that includes policy, environment and socio-economic diversification to analyze the impact of the SNWTP on lake water quality, thus making the entire assessment and analysis more comprehensive than conventional models.

Policy recommendations
Based on the above quantitative and qualitative analysis results of the impact of the SNWTP-ER policy on lake eutrophication, the main influencing factors are selected from Model 11, and we propose the following three suggestions.  Note: *, ** and *** indicate significant at the level of 10, 5 and 1%, respectively.