The sampling locations are shown in Figure 1 and the sampling periods are listed in Table 1. About 2–5 L of water samples were collected at each site. Every sample was divided into two subsamples, one for sediment (A) and the other for P (B) analysis in laboratory. Sub-sample A was further divided into two parts, one for sediment concentration (measured according to China national standard GB11901-89) and the other for the grain size distributions (measured using the HORIBA LA-920 Laser Particle Size Analyzer). Sub-sample B was also subdivided into one unfiltered part and the other filtered through a 0.45-μm filter, all of the subsamples were digested with potassium persulfate to analyze P using the ammonium metamolybdate spectro-photometric method (Environment Protection Administration of China (EPA China), 1989).

We define TP concentration (C _tp , mg/L) and DP concentration (C _dp , mg/L) as the measured value in the unfiltered samples and the value after filtration through a 0.45-μm filter, respectively; and particulate P (PP) concentration (C _pp , mg/L) as the difference between C _tp and C _dp (i.e., C _pp = C _tp - C _dp ) (Zhou et al., 2013). For the sediment properties, S (g/L) denotes the sediment concentration; D ₅₀ (μm) is the median particle size of sediment; C _a (m²/L) is the surface area concentration of sediment ( C a = 6 × 10 − 2 S / γ s ∑ p i / d i ), where γ _s is the specific weight of the sediment (kg/m³), and p _i (%) and d _i (mm) are the percentage and grain diameter of the fractions, respectively (Zhou et al., 2015); d _pp = C _pp /S (mg/g) is the P content per Gram of sediment, called the PP mass density (Zhou et al., 2013); and d _a = C _pp /C _a (mg/m²) is the density of PP on per unit area of sediment grain surface, called the PP areal density (Zhou et al., 2015).

In addition to field surveys, we also collected long-term data for flow and sediment at key hydrological stations, the Yichang and Datong stations in the Yangtze River as well as the Huayuankou and Lijin stations in the Yellow River (Figure 1). This included daily average data of the flow discharge and sediment concentration (1950–2020) and monthly average data for the sediment grain size distribution (1960–2019), which was provided by the Ministry of Water Resources of China (MWR).

In the laboratory, P adsorption experiments were tested to verify the results from the rivers. Three sediment samples collected from the Yellow River were used: fine sediment (D ₅₀ = 4.61 μm), median sediment (D ₅₀ = 20.89 μm), and coarse sediment (D ₅₀ = 51.56 μm) (grain size distribution of these three sediment samples see Supplementary Figure S1). These sediment samples were naturally air-dried and their native P content (d _pp0 ) was 0.59, 0.49 and 0.27 mg/g, respectively. A batch of strictly composed samples was prepared by adding a different amount of sediment (0.3, 0.6, 1.5, 3, 6, 15, 30, 60, 150 or 300 mg) to 30 mL of a P solution with initial P concentrations (denoted as C _dp0 ) of 0.001, 0.01, 0.1, 0.5, 1, and 10 mg/L. These samples were then processed in an incubator shaker and shaken continuously for 24 h to achieve an equilibrium adsorption (Li et al., 2021). Next, the C _dp under equilibrium conditions was measured after passing each sample through a 0.45-μm filter (see section 2.1) and the C _tp was calculated according to the conservation of matter (C _tp = C _dp0 + Sd _pp0 ).

As seen in Table 1, the selected six rivers in China differs substantially in terms of their sediment properties, especially S. The long-term average S of the Yellow River (25.2 g/L) is about two orders of magnitude higher than that of the Xin’an River (0.22 g/L), and the long-term average S of the Luan River (4.44 g/L) is about one order of magnitude higher than that of the Yangtze River (0.486 g/L). As expected, the samples collected from these six rivers in this study has a wide disparity in their S (0.001–41.8 g/L) and D ₅₀ (1.49–195.62 μm) (Figure 2A). Similar to sediment, a wide disparity in the P properties is evident, especially for C _tp (0.02–21.66 mg/L) (Figure 2B). It is worth noting that the peaks of S (41.8 g/L) and C _tp (21.66 mg/L) both appear in the Yellow River, being comparable to that of previous study (47.94 g/L and 30.98 mg/L) (Yu et al., 2010). Like most rivers in the world (Zhou et al., 2018), TP is dominated by PP (the average proportion of C _dp to C _tp of all samples is 68%).

Figure 3 shows the relationships between P and sediment in the six rivers that have a wide disparity in their sediment concentration and size distribution. There is a highly consistent positive relationship between C _tp and S in these six rivers (Figure 3A), indicating that sediment is a crucial agent in the movement and fate of P. This positive pattern is consistent with other rivers like the Pearl River (Duan and Zhang, 1999), the Mississippi River (Welch et al., 2014) and Aurajoki River (Uusitalo et al., 2000), and even lakes such as Poyang Lake (Pu et al., 2020). Notably, the relationship between C _tp and S shows a clear comet shaped pattern from divergence to convergence with increasing S. For instance, the regression relationship between C _tp and S in the area with low S (like S ≤ 0.1 g/L, R ² = 0.04) is much worse than that in the area with high S (S > 0.1 g/L, R ² = 0.92) (Supplementary Figure S3). It indicates that P is mainly attached to the sediment and C _tp is controlled by sediment in high sediment laden rivers like the Yellow River and Luan River.

By contrast, the proportion of DP (C _dp /C _tp ) is negatively related to S (Figure 3C), indicating a buffering effect of sediment on DP. The comet shaped pattern is more evident from the relationship between C _dp /C _tp and S. Specifically, in those areas with high S, sediment has a strong buffering effect, thus C _dp /C _tp is close to zero (i.e., the DP level is mainly controlled by S). However, with the decreases of S, the buffering effect of sediment gradually diminishes, and C _dp /C _tp varies widely (i.e., 3.7%–93.3%), being affected mainly by other factors like the riparian P input. Values of C _dp /C _tp in areas with low S in this study (e.g., Yangtze River and Lancang River) are similar to those reported for some low-sediment water bodies, like lakes (C _dp /C _tp = 10%–97%) (Pu et al., 2020) and estuaries (C _dp /C _tp = 50%–90%) (Fang and Wang, 2020).

Sediment PP density d _pp is also inversely proportional to S (Figure 3E), which indicates that sediment has a strong adsorption potential in areas with high S or small grain size (Zhou et al., 2018), when considering the competitive mechanism between sediment particles (Müller et al., 2006). For example, the markedly lower d _pp due to high S of the Yellow River than the other rivers implies that the sediment still has a strong adsorption potential, a view supported by experimental findings that most of the Yellow River’s suspended sediment acts as a sink for additional phosphate input (Pan et al., 2013).

The highly consistent relationships between P and S from six rivers with varying S convincingly demonstrate the dependence of fluvial P on sediment (Figures 3A, C, E). The specific surface area of sediment affecting P adsorption varies with its particle size (Goldberg and Sposito, 1985; Boström et al., 1988). Therefore, we propose that C _a might better reflect the effect of sediment on P after a comprehensive consideration of S and grain size. The correlation results are shown in Figures 3B, D, F, where the effect of C _a on fluvial P displays a similar pattern to that from S, but PP has a slightly better relationship with C _a (Figure 3F) than its relationship with S (Figure 3E). However, there is no significant improvement in the relationship between TP (or DP) and sediment. One possible reason is that flocculation may affect the effective surface area of fine sediment in natural waters, which may raise inaccuracy when calculating C _a in this study. This effect is discussed in detail in the next section with experimental results.

For a better understanding of the P partition regime (denoted mainly as the ratio C _dp /C _tp ) constituted by sediment of natural water samples, strictly composed samples is analyzed in this section. Figure 4 shows the relationship between C _dp /C _tp and sediment of the strictly composed samples from the P adsorption experiments. We can see that C _dp /C _tp is overall inversely proportional to S and C _a , and it also exhibits a consistent comet shaped pattern (Figures 4A, B), corroborating the field-based results for the rivers (Figure 3C). This comet-like pattern is also supported by the equilibrium adsorption model. These general results confirm that the comet shaped buffer phenomenon in field measurements (Figures 3C, D) is a reasonable and natural law. As shown in Figures 4A, B, the relationship between C _dp /C _tp and C _a is more diffuse than the relationship between C _dp /C _tp and S. As mentioned in Section 3.1, flocculation of sediment, especially fine sediment, may raise inaccuracy when calculating C _a . Here, the effective surface area ( C a ′ ) is proposed by introducing a reduction factor that represents the effect of fine sediment flocculation (Supplementary Equation S2.1). As shown in Figure 4C, by introducing C a ′ , the experimental data from fine to coarse sediment are closely unified into the same equilibrium adsorption curve; Figure 4; Supplementary Table SA1 also indicate that C a ′ performs better in optimizing the correlation of P partition between PP and DP for graded sediment. This further suggests that the surface area is the essential determinant of the effect of sediment on P. Therefore, in order to indicate the P content of sediment in rivers, the true state of sediment in rivers needs to be considered. In this research, we suggest that C _a should be determined by sedimentation particle size (flocculated sediment as a whole).

Our results uncover negative relationship between C _dp /C _tp and S, indicating a strong buffering effect of sediment on DP. However, the divergence phenomenon in the comet tail region implies that this buffering effect is not only controlled by S. Taking the Yangtze River as an example (Figure 5), the statistical results of the measured data show that C _dp /C _tp gradually increase as the group-average D ₅₀ increase, indicating that C _a decrease. Thus, assuming a relatively constant P input level during the sampling period, C _dp /C _tp shifts from being S-controlled to sediment grain size-controlled with declining S. In particular, in areas with low S, the advantage of fine sediment over coarse sediment is most prominent in terms of the buffering effect on DP.

The outstanding buffering effect of fine sediment is due to its stronger ability to adsorb P compared to coarse sediment. As evinced by Figure 6, sediment can restrain the increase in DP caused by increasing C _dp0 through adsorption. The extent that the slope deviates from one can be used to quantify the restraining effect of sediment on C _dp levels in solution. Sediment grain size is a key factor responsible for the restraining effect. Specifically, despite the same S, the amount of phosphate remaining in solution can vary greatly among these three sediment types. Especially, for fine sediment, at the highest S (10 g/L), most of the added P is removed by adsorption (k₃ = 0.15), whereas the effect of coarse sediment under the same S is not discernible (k₃ = 0.81). This is because C _a of the latter (0.75 m²/L) is less than 10% that of the former (12.4 m²/L).

It is important to note that, for all of the three sediment size types, the relationship between C _dp and C _dp0 becomes linear with a slope of 1 as S is reduced below 0.1 g/L (Figure 6). This can further interpret the above-mentioned divergence phenomenon in the relationship between C _tp and S (S < 0.1 g/L, R ² = 0.04). These relationships all indicate that an S value of 0.1 g/L may be a critical threshold for the buffering effect of natural sediment, below which most phosphate cannot be removed by sediment. Actually, for some natural sediment collected from river bed, this threshold value of S is even higher (0.2–1.0 g/L) (Pan et al., 2013). However, in recent years, the natural S in many rivers may fail to reach this threshold. For instance, the average S of the Tuo River (0.03 g/L), Lancang River (0.06 g/L) and Xin’an River (0.02 g/L) during the sampling period in this study are all far below this threshold. This indicates that the buffering effect of sediment on DP in these rivers is gradually disappearing.

DP (especially dissolved inorganic phosphate, DIP) is easily used by primary producers (Maruo et al., 2015) and is often a pollutant for a local or stationary system. In recent years, the significant DP-elevating phenomena has been widely reported in the context of sediment reduction (Chai et al., 2009; Dai et al., 2011; Pan et al., 2013; Chen et al., 2020). Our results suggest that adsorption of dissolved phosphate onto sediment, particularly fine sediment, is a key process to keep C _dp at a relatively low level and to improve the river water quality.

To illustrate the impact of sediment reduction on P cycling in a river, the Yangtze River and the Yellow River are taken as examples for further discussion. The relationship between P and S and that between the P and C _a in the Yellow River are proposed in this study (Supplementary Figure S3). The relationship between P and S (Zhou et al., 2013) and that between the P and C _a (Zhou et al., 2015) in the Yangtze River reported in previous studies are used directly. First, according to the relationship between P and S and the sediment data, long-term variation in the cumulative TP flux (L _tp ) in the Yangtze River (Figure 7C) and Yellow River (Figure 7D) is estimated. Then, combined with sediment grading data, long-term variation in the C _tp and C _dp in the Yangtze River (Figure 7E) and Yellow River (Figure 7F) are roughly modeled using the obtained relationships between P and C _a .

Estimates show that the growth rate of cumulative L _tp (Figure 7C) and C _tp (Figure 7E) related to sediment in the Yangtze River has declined slightly since 1990 in the context of sediment reduction (Figure 7A). In particular, an obvious inflection point appears in 2003 just after the Three Gorges Reservoir (TGR) became operational. The average growth rate of cumulative L _tp through Yichang falls to 0.04 MT/a in 2003–2020, only accounting for 20% of that in 1951–1990 (0.2 MT/a) and the C _tp during wet season at Yichang falls to 0.14 mg/L in 2003–2019, only accounting for 22% of that in 1951–1990 (0.63 mg/L). Like the Yangtze River, the estimated growth rate of cumulative L _tp (Figure 7D) and C _tp (Figure 7F) related to sediment in the Yellow River has declined slightly since 1960 (Figure 7B), relating to the implementation of dams and other hydraulic projects along the river. In particular, the average growth rate of cumulative L _tp through Huayuankou falls to 0.08 MT/a in 2000–2020, being less than 7% of that in 1950–1959 (1.15 MT/a) (Figure 7D). The estimated decreasing trend of L _tp and C _tp in the Yangtze River and Yellow River in this study is consistent with other rivers in the world, such as the Madeira River (decline by 40%) (Almeida et al., 2015) and the Zambezi River (60% trapped) (Kunz et al., 2011). The above estimates for long-term TP are rough but have a physical basis, i.e., sediment has a strong affinity to P due to its great specific surface areas and active surface sites (Davis and Kent 1990; Huang et al., 2016). In this study, this physical basis is further illustrated by both the strong positive correlation between TP and S in the field measurements and the strong P adsorption for natural sediment in the laboratory experiments. Long-term measurements show that since the TGR begins operating in 2003, 93% of the sediment normally delivered to the middle and lower Yangtze River through Yichang has been trapped; it decreased to only 35 Mt/a (2003–2020) compared with 517 Mt/a over the long term (1951–1990) (Figure 7A). Similarly, a stepwise decline in sediment is observed as the implementation of dams and other hydraulic projects along the Yellow River and the average Q _s through Huayuankou in 2000–2020 (119 MT/a) constituted just 8% of that in 1950–1959 (1495 MT/a) prior to the implementation of Sanmenxia Dam (1960) and Xiaoliangdi Dam (1999) (Figure 7B). In fact, the percentage of sediment interception in these rivers far exceeds that of other major rivers globally (Vörösmarty et al., 2003). Therefore, the dramatic decrease in TP estimated in this study is reasonable, which is also consistent with the global trend (Maavara et al., 2015).

Like most rivers in the world, P in the Yangtze River and Yellow River occurs mainly in particulate form (Zhou et al., 2018). In fact, PP is likely strongly adsorbed to iron and aluminum oxide surfaces and is not readily bioavailable (Berner and Rao, 1994). However, PP can be released from sediment and may become mobilized and bioavailable after undergoing transformation under certain chemical conditions (e.g., low pH and low oxygen concentration) (Silva and Sampaio, 1998). Numerous studies have shown that PP is the key source of potentially bioavailable P (BAP) and the ratio BAP/PP is 19%–31% in the Yellow River (Yao et al., 2015) and 45.6% in the Yangtze River (Wei et al., 2010). Thus, from a long-term ecological perspective, maintaining the long-term stability of both the sediment and TP load is critical for the primary productivity of estuary and coast, where P is often the limiting nutrient for bioactivity (Wang et al., 2021). To our knowledge, the ratio of nitrogen (N) to P (N/P) has far surpassed the normal value (15–16) in either the Yangtze River (192.5–317.5) (Shen & Liu, 2009) or Yellow river (700−1480), being much higher than that of other major rivers worldwide (Pan et al., 2013; Wang et al., 2021). This very high N/P ratio means that both the two rivers are extreme P-limited ecosystems and their productivity is severely limited by P (Elser et al., 2007; Shen and Liu, 2009). On the one hand, such significant reductions in L _tp and C _tp will not only further reduce the bioavailability of P but also augment the existing high N/P, altering the nutrient structure of downstream rivers. On the other hand, dam trapping causes P accumulation in the reservoirs that changes fluvial nourishment to contamination within the river.

Conversely, evident increases in C _dp against a background of decreasing C _tp are found in both the Yangtze River (Figure 7E) and the Yellow River (Figure 7F). Especially in the dry season, C _dp increases to 0.08 mg/L at Yichang and 0.09 mg/L at Huayuankou, respectively, in recent years. The inferred trend of rising C _dp is supported further by additional observations from the Yangtze River and the Yellow River; for example, (i) recently, Zhang et al. (2022) reported that the C _dp rose from 0.05 mg/L (2003–2012) to 0.1 mg/L (2013–2017) under the background of decreasing TP at Yichang in the Yangtze River; (ii) earlier, Pan et al. (2013) found the DIP (important components of the DP) increasing from 0.29 μmol/L (1980–1992) to 0.43 μmol/L (2007) due to S decreasing in the Yellow River. The abnormally increased DP level indicates that the decreasing sediment may have nearly lost its ability to buffer P pollutant (Figure 6). In fact, our estimates of how much P cycling could be affected by sediment reduction are likely to be conservative and poorly constrained because TP and DP are not only affected by sediment but also directly affected by pollutant emissions (Zeng et al., 2022). In recent decades, anthropogenic P emissions from both point and non-point sources, particularly fertilizers applied to support agriculture, have greatly enhanced the P load to watersheds (Jarvie et al., 2006). For instance, long-term (1980–2015) net anthropogenic phosphorus input to the Yangtze River Basin has progressively increased (by ∼1.4-fold) (Hu et al., 2020). Such high phosphate input would further promote a rising DP according to our laboratory experiments results (Figure 6). Therefore, maintaining a certain degree of turbidity in rivers plays an important role in the health of water bodies. Major strategies of the Yangtze River and Yellow River require that ecological protection of rivers be put in the first place, and it is recommended for the reservoirs to properly increase the amount of fine sediment released downstream to sustain the health of the downstream ecology.