A method for dividing runoff components of a watershed based on microbial information

Abstract

The application discloses a watershed runoff component division method based on microbial information and belongs to the technical field of watershed hydrology and water resource management. The method collects microbial species abundance data of each potential runoff source end member and rainfall runoff mixed sample; takes the mixed sample abundance data as a benchmark to screen an advantage species set A; takes the set A as a benchmark to screen a specific advantage species set B which belongs to the set A and has higher relative abundance in the corresponding end member in each source end member; performs principal component analysis on the set B, calculates comprehensive contribution scores of each species and selects a candidate tracer factor set C; performs significance test on the set C to screen a final tracer factor; and inputs the final tracer factor into a Bayesian mixture model to calculate the contribution proportion of each source. The application realizes accurate quantitative division of the runoff source, has low cost, simple operation and reliable result.

Description

Technical Field

[0001] This invention relates to the field of watershed hydrology and water resources management technology, specifically a method for classifying watershed runoff components based on microbial information. Background Technology

[0002] In the fields of watershed hydrology and water resources management, accurately identifying the source components of rainfall-runoff is a crucial foundation for revealing runoff generation and confluence mechanisms and formulating pollution prevention and control strategies. Under natural conditions, rainfall-runoff is typically formed by the mixture of three water sources: direct rainfall, soil water, and groundwater. Quantitatively clarifying the contribution ratio of each source is essential for flood early warning and pollution source tracing. In recent years, with the significant decrease in the cost of high-throughput sequencing, acquiring information on aquatic microbial communities has become efficient and economical. This data covers multidimensional information such as species composition and functional genes, and sample preservation conditions are relatively flexible. It can fully reflect the unique microbial community fingerprints formed by different water sources due to environmental differences, providing a new source of tracer information for runoff source identification.

[0003] However, when microbial sequencing data is directly used for quantitative classification of runoff sources, the raw sequencing data usually contains hundreds or thousands of microbial species, many of which are common species or rare species with low abundance that are widely present in different sources. Even if high abundance species are screened out, not all high abundance species have source indicative significance. Species with insignificant abundance differences between different sources also cannot effectively distinguish endmembers. If unsystematically screened species data are directly input into quantitative mixture models, not only will the model be difficult to converge due to dimensionality explosion, but the combined interference of noisy species and invalid tracer factors will also severely reduce the accuracy and reliability of source contribution estimation. Summary of the Invention

[0004] The purpose of this invention is to provide a method for classifying watershed runoff components based on microbial information, in order to solve the above-mentioned problems.

[0005] The technical solution of this invention is: A method for classifying watershed runoff components based on microbial information includes the following steps: Microbial samples from potential runoff source endmembers in the target watershed and mixed samples of rainfall and runoff were collected, and microbial sequencing was performed to obtain microbial species abundance data for each sample. Obtaining abundance data for each sample through microbial sequencing provides a data foundation for subsequent runoff source identification, without relying on the stringent sampling and preservation conditions required by traditional isotope testing, thus reducing experimental costs and operational difficulties.

[0006] Based on the microbial species abundance data of the mixed rainfall-runoff sample, species whose cumulative relative abundance contribution in the mixed sample reaches a preset threshold or whose relative abundance ranks among the top are selected as the dominant species set A. Based on the dominant species set A, species that belong to the dominant species set A and whose relative abundance in the corresponding endmembers meets the endmember representativeness condition are selected from each runoff source endmember as the specific dominant species set B for each source. Through a two-step progressive strategy of first selecting dominant species in the mixed sample and then selecting source-specific dominant species, common species and low-abundance interfering species that are irrelevant to source differentiation are effectively eliminated, so that the remaining species are both dominant in the mixed runoff and representative in the specific source endmembers, thereby improving the source specificity of the tracer factor and reducing data noise.

[0007] Abundance data of species in each endmember sample from the dominant species set B from each source were organized into a data matrix. Principal component analysis (PCA) was performed on the data matrix to calculate the loadings of each species in the principal component space and the variance contribution rate of each principal component. The squares of each loading were weighted and summed using the variance contribution rate of each principal component to obtain the comprehensive contribution score of each species. Species whose comprehensive contribution scores met the preset requirements were selected as the candidate tracer factor set C. Through PCA loading calculation and variance contribution rate weighted summation, the ability of each species to distinguish endmembers was quantitatively evaluated and ranked. While further compressing the data dimensionality, the key information that best distinguishes different sources was retained, improving the stability and computational efficiency of subsequent models.

[0008] The abundance differences of each species in the candidate tracer factor set C among endmembers of different runoff sources are tested for significance. Species with significant abundance differences among different sources are selected as the final tracer factors. Species with statistically significant differences among different sources are selected through significance testing to ensure that the discrimination ability of the final tracer factors is statistically reliable and to avoid introducing randomly fluctuating species into the quantitative model.

[0009] The abundance data of the final tracer factor in each runoff source endmember and the mixed sample of rainfall and runoff are input into a Bayesian mixture model to calculate the contribution ratio of each runoff source to rainfall and runoff. The optimized high-specificity tracer factor is then input into the Bayesian mixture model, which outputs the contribution ratio of each source while also providing the posterior uncertainty interval. This gives the partitioning results a clear and reliable characterization, overcoming the shortcomings of traditional methods in interpreting complex multi-endmember mixtures and lacking uncertainty quantification.

[0010] Furthermore, the microbial sequencing is 16S rRNA gene high-throughput sequencing, and the microbial species abundance data is a relative abundance table of species obtained based on OTU clustering or ASV noise reduction. OTU or ASV abundance data obtained through 16S rRNA high-throughput sequencing can objectively reflect the species composition and relative abundance information of the microbial community in the sample, providing an accurate and quantifiable data basis for subsequent screening of tracer factors, and the analysis and testing costs are significantly lower than traditional isotope tracing methods.

[0011] Furthermore, before using the microbial species abundance data from the mixed rainfall-runoff samples as a benchmark, the abundance data is preprocessed: species with a relative abundance mean below 0.01% or an occurrence frequency below 10% in all mixed samples are removed, and the species abundance in each sample is normalized to relative abundance. By removing low-abundance rare species and normalizing the abundance, the dimensional influence caused by differences in sequencing depth between different samples is effectively eliminated, and random noise signals are removed, allowing subsequent screening to focus on ecologically significant dominant species and improving the reliability of the screening results.

[0012] Furthermore, the preset threshold is that the cumulative contribution of relative abundance reaches 80% to 90% of the total abundance, and the "ranking at the top" refers to species ranking in the top 50 to 200 in relative abundance. While retaining species that play a dominant role in the structure of mixed runoff communities, this effectively excludes low-abundance background species and rare, accidental species, improving the comprehensiveness and representativeness of the dominant species screening.

[0013] Furthermore, the representativeness criteria for the endmembers are: the average relative abundance of the species in the endmember ranks within the top 30% to 50% of all species in that endmember, or is higher than the median average abundance of all species in that endmember. This ensures that the selected species have a genuine abundance advantage in each source endmember, avoids misclassifying low-abundance or occasionally occurring species as source characteristic species, and improves the accuracy and robustness of screening for specific dominant species.

[0014] Furthermore, before performing principal component analysis on the data matrix, the abundance data in the matrix are first standardized using Z-scores. During principal component analysis, the top k principal components with a cumulative contribution rate of 70%–80% are selected. This ensures that the principal component analysis results accurately reflect the relative variation relationships between species. Selecting the top k principal components with a cumulative contribution rate of 70%–80% further reduces the data dimensionality while preserving the main variation information.

[0015] Furthermore, the comprehensive contribution score meets the preset requirement of being ranked among the top 20-50 species. This ensures a sufficient number of candidate species for subsequent significance testing while significantly reducing the data dimensionality, resulting in tracer factors that possess both high contribution and statistical significance, thus improving the convergence speed and estimation accuracy of the Bayesian mixture model.

[0016] Furthermore, the significance test was performed using one-way ANOVA, with P < 0.05 as the significance criterion. One-way ANOVA screened out species with significant abundance differences among different sources from the candidate tracer factors, ensuring the statistical reliability of the final tracer factors. Combined with the strict threshold of P < 0.05, this effectively avoided misjudging random abundance fluctuations between sources as characteristic differences, thus improving the reliability of the tracer factors.

[0017] Furthermore, the Bayesian mixture model employs Markov chain Monte Carlo sampling to obtain the posterior distribution of the contribution proportions of each source, outputting the mean and uncertainty interval of the contribution proportions of each source. Obtaining the posterior distribution of the contribution proportions through MCMC sampling not only outputs the mean but also provides the uncertainty interval, giving the runoff source classification results a clear characterization of reliability and overcoming the deficiency of traditional endmember mixture analysis, which only outputs a single value and cannot evaluate the reliability of the results.

[0018] Furthermore, the potential runoff source end-units include at least three categories: rainwater, soil water, and groundwater. This covers the runoff composition of the vast majority of rainfall events, meeting the practical application requirements for watershed runoff source classification.

[0019] Compared with the prior art, the beneficial effects of the present invention are: This invention employs a two-step progressive strategy: first screening for dominant species in mixed samples, and then screening for source-specific dominant species. This effectively eliminates common species and low-abundance interfering species irrelevant to source differentiation, significantly reducing data noise and improving the source specificity of tracer factors. Furthermore, principal component analysis is used to calculate the loadings of each species and weighted summation based on variance contribution rates to quantitatively assess and rank the endmember discrimination ability of each species. Combined with significance tests, specific tracer factors with both statistical differences and high contribution are finally selected, effectively overcoming the dimensionality explosion and signal inundation problems caused by directly inputting all sequenced species into the model. Finally, the selected tracer factors are input into a Bayesian mixture model, outputting the contribution ratio of each source while providing its posterior uncertainty interval, giving the classification results a clear and reliable characterization. Moreover, this invention only requires microbial sampling and sequencing to complete runoff source classification, eliminating the need for stringent sample preservation conditions and complex auxiliary data and model parameter calibration. This significantly reduces the difficulty and cost of fieldwork and achieves accurate quantitative classification of rainfall-runoff sources, demonstrating promising prospects for widespread application. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention.

[0021] Figure 2 This is a comparison chart of the microbial tracer factors screened in this invention and the results of the classification of runoff sources using hydrogen and oxygen isotopes. Detailed Implementation

[0022] The following is combined with Figures 1 to 2 The specific embodiments of the present invention will be described in detail below.

[0023] Example This embodiment takes a watershed in a county of a province as the research object and classifies the runoff sources of the watershed. The total area of ​​the watershed is 2.19 km², and the terrain is mainly low mountains and hills, with an elevation ranging from 44.72 to 109.18 m, with an average of 68.90 m. The study area belongs to the subtropical monsoon climate, characterized by mild and humid conditions, abundant rainfall, distinct seasons, cold winters and hot summers, and uneven seasonal distribution of rainfall, with rainy springs and summers and dry autumns and winters. The basin exhibits significant meteorological and hydrological characteristics: the average annual precipitation is 1337.0 mm, with 60% of the rainfall concentrated between April and September; the maximum annual precipitation is 1985.4 mm, and the minimum is 900 mm; the average annual temperature is 18.3℃, with extreme minimum temperatures of 5.0℃ and maximum temperatures of 42.5℃; the frost-free period averages 314 days per year; there is abundant sunshine, with an annual sunshine duration of 1432.5 hours; and there is ample heat, with an annual total solar radiation of 108.6 KJ / cm². These climatic characteristics lead to significant differences in the severity and seasonality of rainfall events in the basin, providing an ideal natural setting for conducting runoff source apportionment studies under different rainfall conditions.

[0024] This embodiment classifies potential runoff source end-units into three categories: rainwater, soil water, and groundwater.

[0025] like Figure 1 As shown, a method for classifying watershed runoff components based on microbial information includes the following steps: Microbial samples from potential runoff source endmembers in the target watershed and mixed samples of rainfall and runoff were collected, and microbial sequencing was performed to obtain microbial species abundance data for each sample.

[0026] Within the target watershed, rainwater end-member samples, soil water end-member samples, groundwater end-member samples, and mixed rainfall-runoff water samples were collected.

[0027] Rainwater end-member samples were collected using a rainwater collector or a temporary rainwater sampling bottle, with at least 3 to 5 parallel samples collected from independent events.

[0028] Soil water end-member samples were collected from 0 to 60 cm of soil using a soil water extraction device. At least 3 to 5 parallel samples from independent events were collected, and the sampling points should cover the entire watershed space and its different land use types as much as possible.

[0029] Groundwater end-member samples were collected through monitoring wells, with at least three replicates collected.

[0030] Rainfall-runoff mixed water samples were collected at the watershed outlet or control section to capture complete rainfall-runoff process samples. After rainfall began, samples were collected every 15 minutes for the first hour; then every 30 minutes for the next two hours; subsequently, samples were collected four times at 1-hour intervals, four times at 2-hour intervals, and so on, until a 4-hour sampling interval was reached. The sampling frequency could be adjusted according to the duration and amount of rainfall. If the rainfall increased sharply, the sampling interval was adjusted to 15 or 30 minutes. After rainfall stopped, sampling continued at 4-hour and 6-hour intervals until the flow returned to pre-rainfall levels, at which point sampling ceased.

[0031] All samples were immediately frozen (-20℃) after collection and then filtered uniformly in the laboratory (using a 0.22μm filter membrane) to extract total microbial DNA. High-throughput sequencing was performed using universal primers for the V3-V4 region of the 16S rRNA gene to obtain the original sequences. After quality control, chimera removal, and OTU clustering (97% similarity) for noise reduction, an OTU abundance table was obtained (rows = species, columns = samples, elements are relative abundance).

[0032] Based on the microbial species abundance data of the mixed rainfall-runoff sample, species whose cumulative relative abundance contribution in the mixed sample reaches a preset threshold or whose relative abundance ranks among the top are selected as the dominant species set A. Based on the dominant species set A, species that belong to the dominant species set A and whose relative abundance in the corresponding endmembers meets the endmember representativeness condition are selected from the endmembers of each runoff source and are selected as the specific dominant species set B of each source.

[0033] OTU abundance data from rainfall-runoff mixed samples were preprocessed: the OTU abundance of each sample was converted to relative abundance (summed to 1 or 100%), and rare species with a mean relative abundance of <0.01% in all mixed samples were removed. Then, the mean relative abundance of each species in all mixed samples was calculated, and the species were sorted in descending order. Species whose cumulative contribution to the total abundance reached 80%–90% were selected to form the dominant species set A.

[0034] Using dominant species set A as a baseline, specific dominant species were screened from each source endmember (rainwater, soil water, groundwater). For each source endmember, the average relative abundance of each species in set A was calculated among all parallel samples of that endmember. Species that belong to set A and whose average relative abundance ranks in the top 30% to 50% of all species in that endmember were selected to form specific dominant species sets B for each source.

[0035] Abundance data of species in each endmember sample from the dominant species set B from each source are organized into a data matrix. Principal component analysis is performed on the data matrix to calculate the loadings of each species in the principal component space and the variance contribution rate of each principal component. The squares of each loading are weighted and summed using the variance contribution rate of each principal component as weights to obtain the comprehensive contribution score of each species. Species whose comprehensive contribution scores meet the preset requirements are selected as the candidate tracer factor set C.

[0036] The sets of dominant species from each source, B, are merged into a unified species list, and a data matrix M is constructed: rows = selected species, columns = all endmember samples (rainwater, soil water, and groundwater horizontal samples), and elements represent relative abundance.

[0037] Perform Z-score normalization on matrix M to eliminate the influence of dimensions.

[0038] .

[0039] Where x ij Let x be the relative abundance of the i-th species in the j-th sample. - i and S i and are the mean and standard deviation of the i-th species, respectively.

[0040] Perform principal component analysis (PCA) on the standardized matrix: calculate the eigenvalues ​​and eigenvectors of the covariance matrix to obtain the principal components PC1, PC2, ... (usually the first k principal components with a cumulative contribution rate of ≥70%~80%).

[0041] The specific calculation method for relative contribution (loading): the loading l of each species on the m-th principal component im This corresponds to the i-th component of the eigenvector. To integrate multi-principal component information, the overall contribution score for each species is calculated:

[0042] .

[0043] Where λ m This represents the eigenvalue (or variance contribution rate) of the m-th principal component. According to C... i Arrange the species in descending order and select the top 20-50 species with the highest contribution scores as the candidate tracer set C.

[0044] The abundance differences of each species in the candidate tracer factor set C among different runoff source endmembers are tested for significance, and species with significant abundance differences among different sources are selected as the final tracer factors.

[0045] For each species in the candidate tracer set C, a one-way ANOVA was performed, using source type (rainwater, soil water, groundwater) as the grouping factor and the relative abundance of the species in each endmember parallel sample as the dependent variable to test whether there were significant differences among different sources. Species with significant differences in abundance among different sources were selected using P < 0.05 as the statistical significance criterion. Finally, based on the ranking of comprehensive contribution scores, the top three species with high contribution and significant differences were determined as the final tracer factors.

[0046] The formula for the ANOVA model is: .

[0047] Among them, Y jk Let τ be the relative abundance of the k-th parallel sample from the j-th source. j For the source effect, ε jk This is random error.

[0048] Calculate the F-statistic and its corresponding P-value. Use P < 0.05 as the criterion for statistical significance (this can be combined with an effect size η² > 0.14 as an auxiliary criterion). For species with significant differences, further Tukey HSD post-hoc multiple comparisons can be performed to confirm the differences between each pair of sources.

[0049] Finally, from set C, select those with P < 0.05 and a comprehensive contribution score C. i The top-ranked species were selected, and the top X species with the highest contributions and significant differences were identified as the final tracer factors.

[0050] ANOVA is closely linked to the aforementioned two-step screening and PCA contribution quantification, forming a progressive screening chain of abundance advantage → source specificity → contribution quantification → statistical significance, which serves to optimize the selection of microbial tracer factors and significantly improves the factor discrimination ability.

[0051] In this embodiment, the three final tracer factors selected are: E1 is a combination of Fuscovulum, Chryseobacterium, and Cypionkella; E2 is a combination of Aurantimicrobium, Janthinobacterium lividum, and Paenarthrobacter; and E3 is a combination of Sphingorhabdus, Brevundimonasvesicularis, and Massilia.

[0052] The abundance data of the final tracer factor in each runoff source endmember and the mixed sample of rainfall and runoff are input into the Bayesian mixture model to calculate the contribution ratio of each runoff source to rainfall and runoff.

[0053] A Bayesian mixture model was constructed using the MixSIAR package in R. The model employed an information-free prior distribution, enabling only the residual error term to characterize observational uncertainty while disabling the process error term. The relative abundance data of the three selected final tracer factors in each source endmember and the mixture sample were compiled into an input matrix. The contribution proportions were based on a Dirichlet prior, and the endmember means were based on a multivariate normal distribution prior. Posterior distributions of the contribution proportions from each source were obtained through Markov chain Monte Carlo sampling (10,000 iterations during the burning period to eliminate the influence of initial values, and 50,000 iterations during sampling). The posterior means and confidence intervals were then tabulated to quantify the components of watershed rainfall-runoff.

[0054] In this embodiment, the contribution ratios of rainwater, soil water, and groundwater are shown in Table 1.

[0055] Table 1. Bayesian microbial runoff classification results E1 0.255 0.341 0.404 E2 0.185 0.622 0.193 E3 0.168 0.489 0.343 As shown in Table 1 and Figure 2 As shown, the contribution ratios of rainwater, soil water, and groundwater are as follows: 25.5%, 34.1%, and 40.4% based on E1; 18.5%, 62.2%, and 19.3% based on E2; and 16.8%, 48.9%, and 34.3% based on E3. The results calculated using the three tracer factors corroborate each other, and a comparison with traditional hydrogen and oxygen isotope partitioning results indicates that the method in this embodiment is consistent with actual hydrological processes and can stably output quantitative results.

[0056] The above-disclosed embodiments are merely preferred embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for classifying watershed runoff components based on microbial information, characterized in that, Includes the following steps: Microbial samples from endmembers of potential runoff sources in the target watershed and mixed samples of rainfall and runoff were collected, and microbial sequencing was performed to obtain microbial species abundance data for each sample. Based on the microbial species abundance data of the mixed rainfall-runoff sample, species whose cumulative relative abundance contribution in the mixed sample reaches a preset threshold or whose relative abundance ranks among the top are selected as the dominant species set A. Based on the dominant species set A, species that belong to the dominant species set A and whose relative abundance in the corresponding endmembers meets the endmember representativeness condition are selected from the endmembers of each runoff source and are selected as the specific dominant species set B of each source. Abundance data of species in each endmember sample from the specific dominant species set B from each source are organized into a data matrix. Principal component analysis is performed on the data matrix to calculate the loading of each species in the principal component space and the variance contribution rate of each principal component. The squares of each loading are weighted and summed with the variance contribution rate of each principal component as weights to obtain the comprehensive contribution score of each species. Species whose comprehensive contribution scores meet the preset requirements are selected as the candidate tracer factor set C. The abundance differences of each species in the candidate tracer factor set C among different runoff source endmembers are tested for significance, and species with significant abundance differences among different sources are selected as the final tracer factors. The abundance data of the final tracer factor in each runoff source endmember and the mixed sample of rainfall and runoff are input into the Bayesian mixture model to calculate the contribution ratio of each runoff source to rainfall and runoff.

2. The method for classifying watershed runoff components based on microbial information according to claim 1, characterized in that, The microbial species abundance data are relative abundance tables of species obtained based on OTU clustering or ASV noise reduction.

3. The method for classifying watershed runoff components based on microbial information according to claim 1, characterized in that, Before using the microbial species abundance data of the mixed rainfall-runoff samples as a benchmark, the abundance data is preprocessed: species with a relative abundance mean of less than 0.01% or an occurrence frequency of less than 10% in all mixed samples are removed, and the species abundance in each sample is normalized to relative abundance.

4. The method for classifying watershed runoff components based on microbial information according to claim 1, characterized in that, The preset threshold is that the cumulative contribution of relative abundance reaches 80% to 90% of the total abundance, and the ranking of top species is that the species are ranked in the top 50 to 200 in relative abundance.

5. The method for classifying watershed runoff components based on microbial information according to claim 1, characterized in that, The representativeness criteria for the endmember are: the average relative abundance of the species in the endmember ranks in the top 30% to 50% of all species in the endmember, or is higher than the median average abundance of all species in the endmember.

6. The method for classifying watershed runoff components based on microbial information according to claim 1, characterized in that, Before performing principal component analysis on the data matrix, the abundance data in the matrix are first standardized by Z-score; during principal component analysis, the top k principal components with a cumulative contribution rate of 70% to 80% are selected.

7. The method for classifying watershed runoff components based on microbial information according to claim 1, characterized in that, The overall contribution score must meet the preset requirement as follows: Species ranked in the top 20-50 in terms of overall contribution score.

8. The method for classifying watershed runoff components based on microbial information according to claim 1, characterized in that, The significance test was performed using a one-way ANOVA, with P < 0.05 as the criterion for significance.

9. The method for classifying watershed runoff components based on microbial information according to claim 1, characterized in that, The Bayesian mixture model uses Markov chain Monte Carlo sampling to obtain the posterior distribution of the contribution ratio of each source, and outputs the mean of the contribution ratio of each source and its uncertainty interval.

10. A method for classifying watershed runoff components based on microbial information according to claim 1, characterized in that, The potential runoff source end-units include at least three categories: rainwater, soil water, and groundwater.

Images