Influence of reservoir impoundment on rainfall erosivity in the Three Gorges Reservoir region of China

Spatiotemporal variations in global and regional precipitation are among the most relevant aspects of climate change in warming conditions (Pendergrass et al. 2017; Yao et al. 2020). An upward trend of rainfall extremes is frequently responsible for natural hazards such as floods, landslides, and debris flows (Donat et al. 2017; Pendergrass et al. 2017). Apart from global warming, other anthropogenic factors such as dam construction are also important causes of extreme precipitation variation (Hossain et al. 2010). The moisture-heat exchange above the water surface area becomes more intense after reservoir impoundment, and therefore, evaporation changes coupled with significant land use and complicated surrounding topography may alter the regional precipitation pattern (Gangoiti et al. 2011; Keller et al. 2021). Hossain et al. (2010) and Yigzaw et al. (2012) demonstrated that extreme precipitation (99th percentile) showed a 4% increase per year after dam construction. Pizarro et al. (2013) compared the intensity of extreme precipitation located near water bodies to those away from water bodies, finding higher precipitation intensity at locations closer to water bodies.

The Three Gorges Reservoir (TGR) area – which contains Three Gorges Dam (TGD), the largest hydroelectric project in the world (Xu et al. 2013) – has a water surface area of 1,084 km² (Miller et al. 2005). Specific field experiments and idealized numerical simulations suggested that the climatic influence of the TGR would be primarily within tens of kilometers of the Yangtze River waterway (Miller et al. 2005; Wu et al. 2006). There are studies that reported variations in precipitation and extreme precipitation over the TGR area (Fang et al. 2010; Zhao & Shepherd 2012). Some of these indicated no significant change in precipitation and precipitation pattern after the TGD began operating (in the year 2003) (Miller et al. 2005; Song et al. 2020). However, a numerical simulation by Wu et al. (2006) demonstrated that the TGD construction reduced precipitation in the vicinity of the TGR after the water impounding (135 m) in June 2003. Extreme precipitation in the TGR area was found by Lv et al. (2018) to have become more frequent since 1985 due to climate warming.

The TGR has been impounding water since 2003 (Xu et al. 2013). The extensive water surface areas of reservoirs can influence regional convective patterns and associated thunderstorms by widening open water bodies and enhancing the moisture supply for precipitation, hence serving as a feeder for precipitation (Miller et al. 2005; Woldemichael et al. 2014). Moreover, above the water surfaces, the available convective potential energy is increased, which in turn might induce more precipitation (Degu et al. 2011; Hsu et al. 2016). However, most of the aforementioned studies of the effects of dam operation on local precipitation only considered precipitation datasets before 2010 in the TGR area (Xiao et al. 2010; Liu et al. 2020). This is far from sufficient to understand the local precipitation and extreme precipitation changes before and after the impoundment of the reservoir. A few studies used data from more recent time series, but did not reveal the rainfall change before and after impoundment due to the different research aim (e.g., Lv et al. 2018) or the temporal constraints of data series (Song et al. 2020). Therefore, references discussing the effects of reservoir impoundment on the local climate in the TGR area are still very limited.

The TGR area has one of the most serious soil erosion problems in China (Shi et al. 2021). The annual amount of eroded soil is as high as 150 million tons in this area (Xiao et al. 2020). Rainfall erosivity, considering rainfall amounts and intensities, is an important parameter in soil erosion risk assessments for future land uses and climate change (Shin et al. 2019; Xu et al. 2021). In the TGR area, the changes in the amounts, intensities, and frequencies of precipitation are influenced by TGD operation and certainly will be bound to affect the rainfall erosivity and degree of soil erosion. However, the previous studies have paid more attention to the long-term change trends of annual precipitation (Fang et al. 2010; Degu et al. 2011; Zhao & Shepherd 2012), while very few studies focused on the erosive precipitation variation influenced by the dam operation, including the composition of the erosive rainfall and its respective erosivity contribution (Liu et al. 2020).

Some studies have shown that extreme rainfall plays a key role in soil erosion, and most soil erosion usually results from a few extreme rainfall events (Shi et al. 2021; Xu et al. 2021). The effects of extreme rainfall on soil erosion have become one of the main concerns addressed in soil erosion studies (Gu et al. 2016). Studies referring to extreme rainfall in the TGR area are relatively few (Fang et al. 2010; Lv et al. 2018), indicating an increase both in frequency and intensity of the extreme precipitation. The impact of the TGR impoundment on extreme precipitation has not been clarified. Therefore, the variations of heavy precipitation, including extreme precipitation, wait to be revealed, along with the corresponding erosivity contribution pre- and post-impoundment. It will be of great practical significance to understand reservoir effects on local soil erosion, to further reduce the uncertainty of soil erosion in the future.

Consequently, in the present study, the interannual and seasonal variability of erosive precipitation, classified into different intensity categories (light, moderate, heavy, and very heavy), are analyzed based on the latest daily precipitation data series (as of 2020). Furthermore, the erosivity contribution of different categories of rainfall including extreme precipitation is examined by using a calculation model of rainfall erosivity that is based on daily rainfall. Besides, the changes in precipitation and erosivity before and after the reservoir impoundment, as well as how they influenced by impoundment are discussed. The results from the present study can help understand the influence of TGR impoundment on local precipitation characteristics, especially erosive rainfall events. This will be a guide towards more appropriate strategies to address soil and water loss to provide a scientific basis for ecological security and regional sustainable development in the TGR region.

In the present study, we used daily precipitation data from the Wanzhou station in Chongqing City, China for the period of 1980–2020 (Figure 2). The data were obtained from the China Meteorological Data Sharing Service System (http://www.cma.gov.cn/2011qxfw/2011qsjgx/) and subjected to quality control and homogeneity tests using RClimDex and RHtest software, respectively. These two pieces of software are freely available from the Expert Team on Climate Change Detection and Indices (ETCCDI) website (https://www.wcrp-climate.org/etccdi).

Percentile-based thresholds have commonly been used to define extreme events in past studies. For example, heavy precipitation events were often defined as the top 5% of events, and very heavy precipitation events as the top 1% (Martin et al. 2016; Wu et al. 2016). In the present study, we used three percentile values (85, 95, and 99%) were used as thresholds to divide all erosive precipitation events (daily precipitation ≥10 mm) into four non-overlapping categories: light, moderate, heavy, and very heavy. Days with precipitation totals ≥0.1 mm were treated as precipitation events. The variables examined in this study were defined as follows:

• Light precipitation: the precipitation amount per year below a threshold value of the 85th percentile of daily precipitation amounts distribution from 1980 to 2020 (mm, <85th percentile).

• Moderate precipitation: the total amount of precipitation from the moderate precipitation days (mm, 85th–95th percentile).

• Heavy precipitation: the total amount of precipitation from heavy precipitation days (mm, 95th–99th percentile).

• Very heavy precipitation: the total amount of precipitation from very heavy precipitation days (mm, ≥99th percentile).

In order to study the rainfall erosivity contribution of a single/short-term heavy rainfall, two extreme precipitation indices Rx1day and Rx5day were used, defined as annual maximum 1-day and maximum consecutive 5-day precipitation, respectively, and are two of the recommended indices by the ETCCDMI to investigate extreme climatic variations.

One common non-parametric method for trend detection adopted for this study is the Mann–Kendall (M-K) test (Mann 1945; Kendall 1975). The M-K test can be used to detect the existence of statistically significant trends (Liu et al. 2020). However, it does not provide magnitudes of the trend slope. Therefore, another commonly used non-parametric approach, namely the Theil–Sen estimator (Sen 1968; Theil 1992), was used to estimate the slope of the trends.

The results indicated that the most significant influence of reservoir impoundment was on the heavy precipitation category. The average annual heavy precipitation amount during the post-impoundment period decreased by 55.2 mm, which shows a reduction of 24.9% compared with that before impoundment (p < 0.05). Furthermore, the rainfall erosivity was reduced by 24.3% (350.6 MJ mm ha⁻¹ h⁻¹10 y⁻¹). In addition, the moderate precipitation and erosivity increased by 11.8 and 8.2%, respectively.

Comparing the trend changes between the pre- and post-impoundment periods for the total and erosive precipitation (Figure 3), Theil–Sen's slope indicated that all the trends became weaker after the impoundment since the absolute values of the slopes decreased. The coefficient of variation (Cv) indicates the degree of fluctuation of interannual variation. As Table 1 shows, Cv of total and erosive precipitation during the post-impoundment period (2003–2020) had a slight decrease within 0.01–0.02, while the Cv of rainfall erosivity decreased by 0.16, which indicates the fluctuation degree of annual precipitation and erosivity has been reduced after the reservoir had been impounded. It can also be observed in Figure 3 that especially from 2004 to 2016, the fluctuations of precipitation and rainfall erosivity tended to be stable.

Table 1 presents the Cv values of the light, moderate, and heavy precipitation and their corresponding erosivity differing in the range of 0.01–0.05 in the pre- and post-impoundment periods while the same parameters for the very heavy precipitation category decreased by 0.27 and 0.51, respectively, after the reservoir impoundment. The results indicated that the degree of interannual fluctuation of very heavy precipitation and erosivity is clearly reduced. It can also be seen intuitively from Figure 4(b) that from 2003 to 2020, except for 2017, the annual very heavy rainfall erosivity fluctuated between 0 and 4,560.4 MJ mm ha⁻¹ h⁻¹10 y⁻¹ while before 2003 it had fluctuated between 0 and 9,121.4 MJ mm ha⁻¹ h⁻¹10 y⁻¹.

Based on the results above, it is inferred that the reservoir impoundment not only reduced the heavy precipitation but also partially transformed it into moderate precipitation. Although the reservoir storage had little effect on the annual average amount and proportion of the very heavy precipitation and erosivity, the interannual variations had been modified in that the distributions became smoother during the post-impoundment period.

The distribution of moderate precipitation and erosivity percentage both showed a noticeable shift toward the right of the peak distribution during the post-impoundment period, indicating that the moderate precipitation percentage increased in the 2003–2020 period (Figure 7(a)). During the post-impoundment period, the proportion of moderate precipitation was mainly distributed between 22 and 26%, and the average proportion has increased by 2.9% compared with that before the impoundment. For heavy precipitation, the PDF had a significant rise and shifted toward the left of the peak location in the second period (Figure 7(a)), implying that the heavy precipitation percentage decreased (p < 0.05). After the reservoir impoundment, the proportion of heavy precipitation was concentrated around 18–20%, and the annual average proportion was 5.5% lower than that before the impoundment. Meanwhile, the rise of the PDF curve of moderate and heavy categories indicated that the proportions were more concentrically distributed and had a decrease in interannual variation. The proportions of light and very heavy precipitation did not change much before and after the impoundment, with an average annual proportion of 36.8 and 18.2%, respectively.

The erosivity proportion had the same change as that of the precipitation proportion (Figure 7(b)). The moderate rainfall erosivity proportion showed a right shift with respect to the peak distribution while the PDF curve of the heavy rainfall erosivity proportion presented a left shift in the post-impoundment period (p < 0.05). The average annual proportion of moderate and heavy rainfall erosivity was 2.7% higher and 5.5% lower than that before the impoundment, respectively. The results indicated that since the reservoir impoundment, there has been a tendency to transform from heavy precipitation and erosivity to moderate category, in local erosive precipitation.

Among precipitations of different categories (Figure 9), the moderate, heavy, and very heavy rainfall events all occurred mainly in summer. The precipitation distributions in summer were all more than half of the precipitation (50.6, 56.1, and 67.5%) while there was no distribution in winter. Furthermore, the erosivity distribution was even more prominent in summer, accounting for 53.4, 57.6, and 71.8% of their total erosivity.

For the four categories of precipitation, it can be observed that the light precipitation and its erosivity in spring were increased by 5.5 and 5.0% after the reservoir was impounded (p < 0.05) while the distributions in summer and autumn were slightly reduced (by 1.8–3.4%). Although the distribution of the erosivity of the medium, heavy, and very heavy precipitation in summer decreased to varying degrees, they all increased in autumn. Among the categories, the erosivity distribution of the very heavy rainfall changed the most. It decreased by 15.0% in summer and increased by 13.6% in autumn, as the result of a decrease of 13.8% in the precipitation distribution in summer and an increase of 14.0% in autumn. The distribution of moderate precipitation shifted mainly from summer to autumn with a decrease of 8.4% in summer and an increase of 7.2% in autumn, leading to a 7.5% increase in the distribution of the erosivity in autumn. The distributions of the heavy precipitation and erosivity were mainly transferred evenly from summer to spring and autumn and were reduced by 7.2 and 6.7%, respectively, in summer.

The results showed that the transfer of summer precipitation and erosivity to autumn after the reservoir impoundment was mainly reflected in precipitation except for the low-intensity case, with the very heavy precipitation being affected the most. Therefore, the impoundment of the reservoir changed the previous highly centralized distribution of the precipitation of higher intensity in summer and made it more evenly distributed between the seasons. Meanwhile, the risk of soil erosion in summer was reduced, but the entire high-risk period was extended to span the summer and autumn.

During the pre-impoundment period from 1980 to 2002, there are four obvious peaks in Rx1day erosivity: 3,675.1 (in 1982), 969.3 (in 1989), 1,769.2 (in 1993), 1,577.2 (in 1997), and 3,038.2 (in 2000) MJ mm ha⁻¹ h⁻¹10 y⁻¹, accounting for 30.2, 18, 42.4, and 36.9% of the total erosivity in the corresponding years, respectively. Rx5day erosivity had five obvious peaks: 6,749.6 (in 1982), 2,445.9 (in 1989), 3,227.3 (in 1993), 1,927.6 (in 1997), and 3,168.8 (in 2000). The corresponding erosivity accounted for 55.5, 37.4, 33.4, 51.8, and 38.5% of the total erosivity in the corresponding years, respectively. It can be observed that in the case of extreme rainfall, the erosivity of one rainfall or several consecutive days of rainfall in the TGR area can account for nearly half or even more than half of the annual erosivity. After the reservoir was impounded (2003–2020), the extreme rainfall and its erosivity changed relatively smoothly. The maximum values of Rx1day and Rx5day (in 2020) are respectively 8.7 and 24.7% lower than the average peak values before impoundment. The relative peak values of Rx1day and Rx5day erosivity are 16.7.0 and 33.5% of the average peak values before impoundment. Therefore, the water storage of the reservoir may have the effect of removing the peak and evenly distributing the extreme rainfall. During the post-impoundment period, the extreme rainfall peak values were reduced by 12.7% on average, whereas the erosivity peak values were reduced by 25.1% on average.

Our results indicated a significant influence of the reservoir impoundment on the local precipitation in the TGR area. Distinctive changes existed in the heavy precipitation category (95th–99th percentile). It had a significant decline trend and most of the reduction caused a shift into the moderate category (85th–95th percentile). This result coincided with the findings of Wu et al. (2016) for all of China, that higher precipitation intensities had higher rates of change. The abrupt point by the M-K test was around the year 2002 (Figure 5), which is evidence of the significant effect of the reservoir impoundment (in 2003) on the local precipitation. Both the total precipitation and erosivity presented decreasing trends, which is in line with the conclusion of Wu et al. (2006) who stated that the precipitation increased in the region between Daba and Qinling Mountains, whereas the precipitation in the vicinity of the Yangtze River was reduced after the impoundment in June 2003. However, in the present findings, the mean precipitation did not differ greatly (p > 0.05) after the filling of the TGD, which confirmed the simulation by Miller et al. (2005), suggesting that there was no significant change in annual precipitation.

The changes were mainly due to the land use and land cover alterations caused by the reservoir impoundment, which altered the probability of precipitation being harvested from local evaporation in precipitation recycling (Hossain et al. 2009). After reservoir impoundment, the water surface increased by 397.1 km² (He et al. 2020), hence, it is physically plausible that the increased evaporation from the open-water surface of a reservoir will feed precipitation through a feedback mechanism (Eltahir 1989). On the other hand, the evaporating surface decreased the local surface temperature leading to sinking air (Miller et al. 2005), which limited the supply to precipitation.

Moreover, in the TGR, the land use area covered by vegetation aggregately decreased by 1,196.6 km² over the period of 1990–2015, including dry croplands, paddy fields, grasslands, and forests while the built-up area increased by 799.5 km² (He et al. 2020). The alteration in the spatial pattern of transpiration and evaporation due to the land cover change resulted in a reduction in water flux to the atmosphere, leading to less precipitation from thunderstorms and a spatial displacement in the rainfall pattern (Pielke & Walko 1999). Therefore, although large-scale surface evaporation can trigger changes in precipitation recycling, the reduction of transpiration and evaporation may dominate the feedback mechanism. As a result, the annual precipitation demonstrated a declining trend. Meanwhile, the reduction of the water flux into the atmosphere after the reservoir impoundment also led to a reduction in the rainfall of a single rainstorm. Therefore, the heavy rainfall peaks were significantly eliminated and mostly shifted into moderate rainfalls. However, the erosivity of heavy precipitation has not been transferred into the moderate category to the same extent, the former decreased by 24.3%, and the latter increased by only 8.2% after the water was impounded. Hence, the reservoir impoundment obviously mitigated the erosion threat of the heavy rainfalls in the TGR and also reduced the risk of annual erosion.

The monsoon season of the TGR occurs from May to September, when the temperature is high and rainfall is necessary to compensate for intense evaporation (Fang et al. 2010). Therefore, most precipitation, as well as the heaviest rainfall events, occur in summer. However, due to the predominant effect of the reduced area of plant transpiration after the impoundment, especially in summer when the transpiration is the strongest, summer precipitation decreased. This is similar to the results presented by the study of Zhang et al. (2015) in Guangdong Province, China, indicating that not only precipitation in summer was decreasing overall, but also that a wetting tendency of the winter, spring, and autumn existed due to the water reservoirs. Among the four intensity categories, the very heavy precipitation contributed the most (decreased by 15.0% in proportion in summer) to the decline of summer precipitation. The weaker rainstorm led to a weaker fluctuation of the summer precipitation, which was also due to the reduction of the water flux into the atmosphere after the reservoir impoundment.

The replenishment of atmosphere water flux by reservoir surface evaporation in autumn was obviously greater than that before impoundment. Therefore, autumn precipitation, especially for very heavy precipitation cases, increased. The unimodal distribution of monthly rainfall before the impoundment had also changed to a bimodal distribution, with the second peak in September (Figure 8). As a result of the seasonal precipitation variation after the water was impounded, rainfall erosivity was no longer highly concentrated in summer, causing a very high level of erosion risk in July. Instead, the erosivity was evenly distributed in the summer and autumn (May–September) and the erosion risk period was less threatening than before the impoundment. Therefore, the change from one erosivity peak in July before impoundment to two peaks in July and September indicates that the risk of soil erosion is reduced, but now there are two risk periods to focus on each year.

After the reservoir impoundment, it is physically plausible that the intense evaporation from the vast water surface will alter extreme precipitation patterns through a feedback mechanism (Hossain et al. 2009). In our results, after the water was impounded, the most notable change in extreme rainfall was the flattening of fluctuations. The peak values of the Rx1day and Rx5day were reduced by 8.7 and 24.7%, respectively, verifying the reduction of the water flux into the atmosphere after the reservoir impoundment.

Besides, in the present study, very heavy precipitation is also counted as one extreme indicator since it is classified as the top 1% of the rainfall events, which is the R99p among the commonly used indices (Wang et al. 2017). Likewise, the fluctuation degree of the very heavy precipitation was reduced obviously after the impoundment (Figure 4(a)), except for an extreme value in 2017. In the autumn of 2017, the subtropical high has been maintained along the Yangtze River in Chongqing and the middle and lower reaches of the Yangtze River for a long time. The middle and high latitudes maintained a double resistance type, slowing the western low trough to move eastward, which resulted in frequent abnormally heavy precipitation in Chongqing from September to October (Tang et al. 2019). Therefore, if excluding the impact of extreme climate in 2017, the fluctuating degree reduced by 42.2%. That is, the flattening and peak-cutting effects of reservoir impoundment on extreme precipitation were more outstanding.

However, many studies demonstrated that an increasing trend of extreme precipitation has emerged under global warming (Trenberth et al. 2015; Donat et al. 2017). The present study indicates that, under the influence of the reservoir impoundment, the change of trend of the extreme rainfall was not specific. This is consistent with the finding of Song et al. (2020) indicating that the TGR impoundment has not significantly affected the storm patterns.

The precipitation from the very heavy category (99th percentile) is characterized by relatively short duration and high intensity that occurred approximately twice a year and contributed to more than one-third of the annual erosivity in the TGR. The same situation existed in Rx1day and Rx5day. Previous studies noted that soil loss was highly sensitive to precipitation types with short durations and very high intensities, which would result in severe land degradation (Martin et al. 2016; Strohmeier et al. 2016). However, after the reservoir impoundment, the previous extreme rainfalls were reduced and smoothed in the TGR, suggesting that the reservoir impoundment eliminated the threat from a few extreme precipitation events. Although the narrow spatial constraints of the investigation may limit the generality of the results and conclusions, this study provides evidence of a significant effect of TGR impoundment on local precipitation.

The influence of the reservoir impoundment of the TGR since 2003 on local precipitation and rainfall erosivity was investigated in the present study. The impoundment has had a significant impact on local precipitation, which is mainly reflected in the heavy precipitation (95th–99th percentile) and the relative erosivity. Both of these parameters displayed a significant decline trend from 1980 to 2020 (p < 0.05). The abrupt point indicated by the M-K test was around the year 2002, which is evidence of the significant effect of the reservoir impoundment on the local precipitation. Reservoir impoundment redistributed the intensity levels of erosive precipitation in which the precipitation and erosivity both shifted from the heavy to the moderate category. After the impoundment, the annual heavy precipitation was reduced by 24.9% and the relative erosivity decreased by 24.3% averagely (p < 0.05). The moderate precipitation and erosivity both increased by 11.8 and 8.2%, respectively.

The impoundment of the TGR also caused a trend of precipitation shift from summer to autumn. As a result, the unimodal distribution of monthly precipitation and erosivity had been altered to a bimodal distribution, resulting in a longer but lower-risky erosion period of high concern. The most significant effect of reservoir impoundment on extreme rainfall lies in the peak cutting and flattening effect. The interannual fluctuations of Rx1day and Rx5day have been significantly reduced after reservoir impoundment, with a peak clipping of 12.7%, and correspondingly 25.1% on the rainfall erosivity on average.

This work was supported by the Natural Science Foundation of China (U2040207) and the Sichuan Science and Technology Program (2020YFQ0002/2022YFS0493).

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.