Effects of nitrogen and phosphorus on chlorophyll a in lakes of China: a meta-analysis

We selected experimental studies that assessed the effects of nutrients (total phosphorus, total nitrogen), TN/TP on Chla of the lake in China. Firstly, we conducted a literature search using Web of Science with the following sequence of key terms: (TS = (phosphorus*OR 'Total phosphorus' OR TP OR nitrogen* OR 'Total nitrogen' OR TN) AND TS = eutrophication AND TS = (Chl-a OR 'Chlorophyll a') AND AD = China). This search yielded 805 hits.

We complied with the following criteria to avoid bias in selecting publications. First, we included only field experiment data. Second, all nitrogen limits in eutrophication contained at least TN and TP. Third, Chla had to be explicitly reported with the published data. If the data were presented in a graph, we digitized the data using the GetData Graph Digitizer (version 2.24, Russian Federation). Finally, we included only reports on the lake in china. For accuracy, our analysis required the following details in the selected publications: location (longitude/latitude), the research location, nutrition (TN, TP, and TN/TP). A total of 61 peer-reviewed papers were produced during the screening process based on abstracts, charts, tables, and full text, as well as the removal of duplicates (figure 1, supplementary information (SI) references). Representing 42 sites made up the final dataset (figure 2, SI table S1).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In a meta-analysis, results in the form of one or more effect sizes are extracted from each study. Effect sizes are designed to combine the results of different studies on the same scale, and use a set of indicators for comparison (Gurevitch et al 2018). This is essential for past effect size indicators. In this study, we used the response ratio (RR) as the effect size. The natural-logarithm-transformed response ratio $\ln {\text{RR}}$ is generally used as an indicator in the field of ecology (Hedges et al 1999) By using the $\ln {\text{RR}}$, we can give a more robust prediction of how Chla varies with a change in a given nutrient driver. All statistical analyses were performed with R v3.5.4 (R Core Team 2020).

According to (Lajeunesse 2015), calculate the response rate $\ln {\text{RR}}$ and its variance (v) of the natural logarithmic transformation of individual experimental units:

$$\begin{equation}\ln \left( {{\text{RR}}} \right) = \ln \left( {\frac{{{Y_{\text{t}}}}}{{{Y_{\text{r}}}}}} \right) + \frac{1}{2}\left[ {\frac{{S_{\text{t}}^2}}{{{N_{\text{t}}} \cdot Y_{\text{t}}^{\,2}}} - \frac{{S_{\text{r}}^{\,2}}}{{{N_{\text{r}}} \cdot Y_{\text{r}}^2}}} \right]\end{equation} \tag{ 1 }$$

$$\begin{align}v = \frac{{S_{\text{t}}^2}}{{{N_{\text{t}}} \cdot Y_{\text{t}}^{\,2}}} + \frac{{S_{\text{r}}^2}}{{{N_{\text{r}}} \cdot Y_{\text{r}}^{\,2}}} + \frac{1}{2}\left[ {\frac{{S_{\text{t}}^{\,4}}}{{N_{\text{t}}^{\,2} \cdot Y_{\text{t}}^{\,4}}} - \frac{{S_{\text{r}}^{\,4}}}{{N_{\text{r}}^{\,2} \cdot Y_{\text{r}}^{\,4}}}} \right]\end{align} \tag{ 2 }$$

where Y denotes mean Chla, S is the standard deviation of that mean, and N is the sample size for the treatment (subscript t) and the reference (subscript r) groups.

We calculated the weighted mean $\ln {\text{RR}}$ using the mixed-effects model with the R package metaphor.

The weighted mean (M) and its variance (V) were calculated as

$$\begin{equation}M = \frac{{\mathop \sum \nolimits_{j = 1}^k {W_j}{M_j}}}{{\mathop \sum \nolimits_{j = 1}^k {W_j}}}\end{equation} \tag{ 3 }$$

$$\begin{equation}V = \frac{1}{{\mathop \sum \nolimits_{j = 1}^k {W_j}}}\end{equation} \tag{ 4 }$$

where k is the total number of experimental units, ${M_j}$ is effect size $\ln {\text{RR}}$ in experimental unit j, and ${W_j}$ is the weighting factor, which is the inverse of the variance. The 95% confidence interval (CI) for the weighted mean was computed as:

$$\begin{equation}{\text{CI}} = M \pm 1.96 \times \sqrt V .\end{equation} \tag{ 5 }$$

In the presentation of this article, the value of $\ln {\text{RR}}$ is reversely converted, and the percentage change of Chla caused by nutrients was calculated based on the weighted response rate (equation (6))

$$\begin{align}E\left( \% \right) = & \,\frac{{{X_e} - {X_r}}}{{{X_r}}} \times 100\% = \left( {\frac{{{X_e}}}{{{X_r}}} - 1} \right) \times 100\% \nonumber \\ = & \,\left( {{e^{\overline {\ln \left( {{\text{RR}}} \right)} }} - 1} \right) \times 100\% .\end{align} \tag{ 6 }$$

They are considered statistically significant if the 95% CI do not overlap with zero.

Our results indicated that the response of TN and TP to Chla was positive in the lakes of China (figure 3) with RR values of 5.5% and 272.0% respectively. As showed in the figure 3, TN and TP mainly enter the ecosystem through runoff from land, which leads to water blooms, or great concentrations of algae, and even fish die. This study revealed the effect of TN and TP on Chla at three concentration levels: lower, medium to high (TN: 64.0%, 99.0%, and 66.0%, TP: 28.0%, 170.0%, and 46.0%).

Zoom In Zoom Out Reset image size

3.1.1. Effects of TN on Chla

The average effect in the TN of the lakes in China was 5.5% (figure 3). Our model quantified the response of RR to changes in TN was positive and significant across studies (figure 4(b)). We observed a positive relationship between the TN and Chla (figure 5(a), SI figure S1) (Phillips et al 2008, Magumba et al 2014). Whilst analyzing individual studies, we observed that TN (0.75 < TN ⩽ 1.4 mg l⁻¹) have the largest RR flexibility (r = 99% [60, 147], n = 602, P < 0.001) followed by TN > 1.4 mg l⁻¹ (r = 66% [34, 106], n = 250, P < 0.001) and TN (0.75 < TN ⩽ 1.4 mg l⁻¹) (r = 64% [27, 113], n = 172, P < 0.001) in lake (figure 4(b), SI table S2). Generally, the effect of TN on Chla in the lakes showed a positive relationship trend in China. Detailed RR of Chla in TN from different concentrations were listed in table S2.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

3.1.2. Effects of TP on Chla

3.1.3. Effects of TN/TP on Chla

Response of Chla trophic states under different combinations of TN and TP concentrations are shown in figure 6. Firstly, we compare the response of the Chla trophic state when both nutrients are at low levels (figure 6(a)) and when both nutrients are at middle levels (figure 6(e)) or high levels of nutrient states (figure 6(i)). When both TN and TP are at low levels of nutrient states, the $\ln R$ of Chla Eutrophic is high (2.41) and the $\ln R$ of Chla oligo-mesotrophic is low (0.83). However, if the concentration state of TN becomes middle or high levels, the $\ln R$ of Chla oligo-mesotrophic decreases greatly to 0.65 and 0.64, respectively. On the one hand, the response of oligo-mesotrophic Chla is 0.66 when both nutrients (TN and TP) are at middle levels and 1.28 for eutrophic Chla. On the other hand, when the TN and TP are at high levels the $\ln R$ increased with the Chla increased (figure 6(i)). Therefore, the levels of TN and TP are very important for the effect of lakes Chla. This shows that Chla is limited by nutrients at macroscales.

Zoom In Zoom Out Reset image size

Next, we discuss the question of whether or not a single (TN or TP) or both (TN and TP) nutrient affects Chla trophic state, because we have established that the nutrients are important for the response of Chla. We kept the levels of the other nutrient constant so that we can explore the effect of one nutrient independent of the other. For example, we can determine the effect of TN on Chla by comparing figures 6(a)–(c). When we set the TP levels to be lower, changing the TN levels from lower (figure 6(a)) to the middle (figure 6(b)) or high (figure 6(c)) will lead to a decrease in the $\ln R$ of Chla (a decline from 0.83 when TP is lower to 0.64 when TN is high). Concurrently, we see an increase in the $\ln R$ of Chla (an increase from 2.41 when TN is lower to 2.49 when TN is middle), while the data is less convincing when TN concentration is at a high level. When we keep the TP state at the middle (figures 6(d)–(f)) or hyper (figures 6(g)–(i)) levels, we obtain similar results to the results that the effect of TN on Chla. To determine the effect of TP on Chla, we hold the TN concentration levels. If TN concentration levels are lower (figures 6(a), (d) and (g)), changing the levels of TP from lower to middle or hyper will cause a change effect to Chla based on oligo-mesotrophic and eutrophic trophic states (0.83 when TP is at lower levels, 0.91 when TP is middle levels, and 0.10 when TP is high levels). If TN concentration levels are middle (figures 6(b), (e) and (h)), the oligo-mesotrophic trophic states of Chla changes are little (0.65 when TP is at lower levels, 0.66 when TP is at middle levels, and 0.77 when TP is high levels). If TN concentration levels are high (figures 6(c), (f) and (i)), the eutrophic trophic states of Chla changes are decreased (2.49 > 1.28 > 1.21). Therefore, according to the effect of TN or TP on different Chla trophic states, both TN and TP could influence the Chla $\ln R$, showing that both TN and TP could be limiting.

To determine the relative importance of TN and TP, as both nutrients can affect Chla. We assume that both TN and TP concentrations are at lower levels, and then shift either concentration levels to middle or high concentration. The changes in the TP concentration to a middle or high level will to 72.9% (0.825–0.096) of the oligo-mesotrophic trophic states of Chla and 88.0% (2.41–1.53) of the eutrophic trophic states of Chla (figures 6(a), (d) and (g)). In contrast, the $\ln R$ of the TN concentration to a middle or high level only causes a change for 18.5% (0.825–0.64) and 8.0% (2.49–2.41) of the Chla (figures 6(a)–(c)). Moreover, we do have not enough data to evaluate changes in $\ln R$ for hypereutrophic trophic states of Chla. Therefore, although TN and TP influence the Chla trophic state, TP (72.9% > 18.5%, and 88.0% > 8.0%) is more important than TN. Considering the huge changes of $\ln R$ between the TN and TP effect and the large effect of TP on Chla, TP generally dominates in determining Chla trophic state, indicating that TP is more important than TN in limiting Chla.

Finally, we examined whether TN and TP could interactively affect the Chla trophic state. We found that when the TN concentration was at lower levels, changing the TP concentration from lower to middle or high levels would cause an increase in the change of Chla being hypereutrophic (figures 6(a), (d) and (g)). However, when the TN concentration levels were middle or high, the response of Chla being eutrophic or hypereutrophic increased when changing the TP concentration levels from lower to middle or high (figures 6(d)–(f)) in a middle level, figures 6(g)–(i) in high levels). That is, the impact of the TP concentration on the Chla is much larger when the TN concentration levels are middle or high, indicating that TN and TP have a positive interaction in determining the eutrophic or hypereutrophic of Chla. There is a relative decrease in the change to Chla being hypereutrophic when the TN or TP changes from middle or high levels to lower levels. Therefore, both nutrients (TN, TP) are important and suggest co-limitation, when the Chla is eutrophic or hypereutrophic.

We summarized whether the Chla of lakes in China is limited by TN, TP, or both on a macro scale. Although both the TN and TP affect the Chla, TP generally plays a more important role (272% > 5.5%), which is consistent with most of the earlier synthesis studies (Xu et al 2013, 2015, Schindler et al 2016). Early stages of eutrophication in lakes can be managed by reducing phosphorus alone (Guildford and Hecky 2000, Ma et al 2015). Our results showed that the effects of N are limited when Chla trophic state is oligo-mesotrophic (TP: 64.4% vs TN: 9.6%) and co-limitation of P and N when Chla trophic state is eutrophic (TP: 110% vs TN: 121%) or hypereutrophic (TP: 240% vs TN: 173%). As shown in table 2, a finding consistent with the findings of many earlier studies.

As a representative index of eutrophication, the Chla level has always been a crucial indicator monitored by environmental managers. Importantly, TN/TP is one of the most widely used indicators to explain the nutrient limitation for phytoplankton (Liebig 1842, Redfield 1958, Liang et al 2020). In most cases, the relationships between Chla and TN/TP are fundamental to guiding the lake's eutrophication management. Our results showed a significant decreasing trend of the TN/TP with increasing Chla concentration for the lakes in our database (linear regression line fitted to figure 7). As shown in figure 7, $\ln \left( {{\text{Chla}}} \right)$ and $\ln \left( {{\text{TN}}/{\text{TP}}} \right)$have a linear decrease relationship. The linear relationships between logarithmically transformed data of TN/TP and Chla have been widely used by environmental managers seeking eutrophication governance (Tong et al 2018). The nutrient limitation condition shifts from P-limitation to that P and N co-limitation, when the TN/TP approaches the Redfield Ration (7.2 by mass, as shown in figure 7 dashed horizontal line). Three eutrophication trophic were digitized with two solid lines which set the Chla concentration to be 7 μg l⁻¹ and 30 μg l⁻¹ (oligo-mesotrophic and eutrophic, eutrophic and hypereutrophic). We estimated TN/TP is 16.7 (oligo-mesotrophic and eutrophic), which is close to the proposal of 16 by (Tong et al 2018). At this time, the estimated TN/TP is 24.4 (eutrophic and hypereutrophic), which is close to the proposal of 22 by (Guildford and Hecky 2000).

Zoom In Zoom Out Reset image size

Generally, P and N are most related to the Chla trophic state in the lake. The release of P from the sediment and decomposition of phytoplankton was the key process to the release of P. However, the N cycle has natural removal mechanisms at the waterbody, such as denitrification, which can decrease N inputs. The denitrification increases with increased water eutrophication because of the increased organic matter. So, P inputs are often retained at higher proportions than N inputs. Globally, TN/TP is decreasing trend with increasing Chla concentration. Yan et al (2016) found a similar negative relationship between the TN/TP and Chla from 157 publications data. In addition, Liang et al (2020) found a similar negative relationship between the TN/TP and Chla analyzed data from 1382 lakes in 17 US states with a probabilistic machine learning method, Bayesian Network. Similarly, the TN/TP decreased with the eutrophication in lakes such as Taihu, Dianchi, and some American or Canadian lakes (Dove and Chapra 2015, Yan et al 2016, Tong et al 2018).

Overall, our meta-analysis was performed using 61 individual research data that a wide range of trophic states, and because our results are supported by single lake studies. Therefore, our findings are significant to understand the limitations of TN or TP and provide some reference to eutrophication managers. For example, we can decide whether to limit N or P under different Chla trophic states according to the concentration threshold of TN and TP.

As shown in figure 8, we quantitatively analyzed the response of Chla to different TN and TP concentrations. However, considering the temporal and spatial variations of lake information, it is impossible to have a strategy that can be suitable for the eutrophication management of all lakes. Therefore, our results are not omnipotent but can apply to general guidelines for many lakes. It can provide important prior information for eutrophication management, especially for lakes that have little data.

Zoom In Zoom Out Reset image size

On the one hand, when we keep the TN concentration unchanged. Our results showed, decreasing the concentration of TN and TP is useful for the recovery of hypereutrophic lakes. The response of Chla being hypereutrophic reduced a large when TP concentration from high to middle level (54.0%, 2.4–1.56, figure 8(a)). Similarly, the $\ln R$ of Chla being eutrophic reduced a large when TP concentration from high to middle level (29.0%, 1.39–1.1, figure 8(a)). For the recovery of hypereutrophic lakes, decreasing concentrations of TN also can reduce the $\ln R$ of Chla being hypereutrophic when we keep the TP concentration unchanged. The $\ln R$ of Chla being hypereutrophic would be reduced by 17.0% (1.73–1.56, figure 8(b)) when the TN concentration becomes middle level. In other words, the eutrophic or hypereutrophic lake is more suitable for simultaneously controlling the input of N and P, or reducing the input of P alone.

The novelty of our research is examining effects of N and P across 41 lakes (61 studies) in China. We emphasize the importance of using a dataset with a large number of lakes and a variety of ecological contexts, and then quantified the effect of TN and TP on Chla with meta-analysis methods. The analysis of our national lake eutrophication database revealed that different concentrations of TN and TP influences Chla status, although the consistency and magnitude of the effect varies across the lake.

Our work also highlights the innovative application of meta-analysis in exploring the effect of TN and TP on Chla at macroscales. As shown in figures 4 and 5, under certain concentration levels, the $\ln R$ of Chla is not deterministic. Fortunately, meta-analysis can obtain quantitative results by collecting peer-reviewed articles. As Hedges et al (1999) said, meta-analysis is the quantitative, scientific synthesis of research results. We also argue that the meta-analysis could be encouraged as an effective tool for use in macrosystem studies. Firstly, the variance between groups in the mixed-effects model explains the impact of the spatial heterogeneity of the ecological environment and reduces the bias. Secondly, our findings can provide targeted strategies for eutrophication management, but we still need to be cautious because some data are rare. Finally, open science and big data have emphasized unbiased access to research data, which is of longstanding importance in meta-analysis. On this basis, we can get more peer-review meta-data, so that the results are more robust. This will provide an important reference for formulating eutrophication strategies from a macro perspective.

Despite its current utility and future potential, meta-analysis has various limitations as a tool. For example, this study evaluated the effects of TN, TP, and TN/TP on the Chla trophic state in lakes at macroscales in China. The predictions of their net effects are difficult to generalize because these factors may depend on the Secchi depth, vegetation type/composition, and time scale. Our analysis did not address the impact of these additional factors. Therefore, future work needs to be carried out more factors.