Next Article in Journal
Electrifying the Road to Net-Zero: Implications of Electric Vehicles and Carbon Emission Coefficient Factors in European Power Systems
Previous Article in Journal
Integration of Methane Reforming and Chemical Looping Technologies for Power Generation from Waste Plastic: Technical and Economic Assessment
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 16
Issue 12
10.3390/su16125083
sustainability-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Dynamic Prediction of Total N and P Contents in Slurry from Dairy Farms under Different Treatment Processes Using Near-Infrared Spectroscopy

by
Mengting Li
,
Di Sun
,
Shengbo Liu
,
Keqiang Zhang
,
Run Zhao
* and
Zengjun Yang
*
Agro-Environmental Protection Institute, Ministry of Agriculture and Rural Affairs, Tianjin 300191, China
*
Authors to whom correspondence should be addressed.
Sustainability 2024, 16(12), 5083; https://doi.org/10.3390/su16125083
Submission received: 27 April 2024 / Revised: 28 May 2024 / Accepted: 3 June 2024 / Published: 14 June 2024
Browse Figures
Versions Notes

Abstract

:
Nutrient content fluctuation in dairy production slurry is highly influenced by the various treatment processes applied in the Chinese dairy sector. The dynamic measurement of these contents is critical for the practical and efficient field application of slurry subjected to various processes. In the study, a total of 715 slurry samples were collected from 24 intensive dairy farms in Tianjin subjected to three typical treatment processes. Descriptive statistical analysis, principal component analysis, and partial least square regression were used to investigate the variation in total nitrogen (TN) and total phosphorus (TP) contents, spectral characteristics, and the performance of the prediction model of the slurry under the processes, respectively. Results revealed significant differences in both TN and TP contents along with the spectra for the slurry subjected to different treatment processes. All the inter-process models showed poor performance, and the results were worse compared to the intra-process models. Among the intra-process models for TN, the optimally performing models were the Pac fusion model (R2pred = 0.82; RPD = 2.38) and the single model Pa (R2pred = 0.83; RPD = 2.31). Among the intra-process models for TP, the optimum results were seen for Pab (R2pred = 0.77; RPD = 2.07) and Pa (R2pred = 0.79; RPD = 2.30). Taking different treatment processes into consideration is essential to establish flexible models that can be adaptive for diversified scenarios. This would be helpful to improve the tracking monitor measures, efficiently guide the land application of slurry, and support the sustainable development of animal farming and environmental conversation.

1. Introduction

There has been an increase in several environmental issues along with the ever-increasing intensification and modernization of the dairy farming industry in China. Among these, the issue of massive amounts of slurry in high concentrations has been extremely prominent. Calculating in terms of both the Manual on Accounting Methods and Coefficients of Pollutant Discharge from the Statistical Survey of Emission Sources and the Chinese Animal Husbandry and Veterinary Yearbook, the production of slurry from dairy farms in 2021 is 0.86b m3 [1,2]. Importantly, both nitrogen (N) and phosphorus (P) in the slurry are crucial indicators for land application. However, the nitrogen use efficiency (NUE) and phosphorus use efficiency (PUE) are only 33% and 50%, respectively, which is far below those in Western countries [3]. In order to augment the NUE and PUE, obtaining N and P contents in the slurry at the optimal time can be useful for slurry application in fields in reasonable amounts. In Western countries, slurry flows from the dairy barn to the storage facility, eventually returning to the farmland without reversing [4]. Domestic slurry treatment processes are different from these approaches; they are complex and variable, often moving in circles with longer chains and multiple links [5,6]. Therefore, it is imperative to develop an efficient approach for obtaining these vital nutrient contents throughout the slurry processing chain.
Currently, dynamic measurements of N and P content in livestock and poultry slurry mainly include direct determination methods based on electrochemical sensing technology [7,8,9] and indirect prediction methods based on spectroscopic technology. However, due to the high N content and complex composition of slurry filled with high nutrient contents, sample dilution is often required when using electrochemical methods, resulting in the relatively weak stability of detection results and limited measurable substances. Taking into account the accuracy, stability, and cost, recent studies have increasingly preferred the use of near-infrared spectroscopy (NIRS), which offers rapid measurement without preprocessing, the simultaneous detection of multiple components, and cost-effectiveness [10,11,12]. For instance, Tolleson and Angerer [13] employed NIRS and successfully confirmed fecal N and P from six ruminant herbivore species. In another report, Cabassi et al. [14] compared the performances of four different NIR spectrometers by using 99 dairy cattle slurries collected from 94 heterogeneous livestock farms. The comparison of the results showed generally good prediction accuracies. Given these findings, it can be seen that NIRS is fully applicable for the practical and dynamic acquisition of slurry nutrients before their field application. However, some researchers have also implied that the precision and robustness of an NIRS-based model would fluctuate under complex conditions [14,15].
Given many similar findings, many current studies have shifted their aim to the spectral response and model performance driven by a broad spectrum of factors based on the samples [16,17,18]. For instance, Finzi et al. [16] compared the influence of three sample preparations and two reading set-ups during the NIR scanning on 36 digestates and livestock slurries. The predicted effect of filtration and homogenization was better than that of the original samples, while the spectra obtained by using a Petri dish as the sample cell were slightly more accurate than those obtained from an optical fiber. In addition to these microscopic conditions, external environmental factors were additionally taken into account to meet the requirement of its practical application. On these lines, our team illustrated the differential performance of NIRS based on fertilizing in different seasons [15]. Some reports have focused on nutrient migration and transformation in the slurry under different treatment processes in lab-scale and pilot-scale studies [19,20]. Hence, it is essential to further clarify any impact and the performance of the spectra and models as a response to the processes.
The objective of this paper is to (1) unveil the distribution characteristics and variation regularity of TN and TP contents under three different treatment processes; (2) explore the difference in the spectra under three treatment processes; and (3) establish predictive models for TN and TP contents within and among processes utilizing NIRS. Taking both the regularity of chemical values and spectra into consideration, as well as the comparison of inter-model performance, the response mechanism of slurry TN and TP towards different processes will be collectively revealed. Further, diversified, flexible, customized models that can meet the demands of different practical scenes will be creatively established. This will be beneficial for not only enhancing the sustainable monitoring measures but also providing effective guidance for the land application of slurry.

2. Materials and Methods

2.1. Study Area

Tianjin is among the three foremost dairy production areas of China and has been a leading metropolis center both in dairy farming and processing over the years. Apart from crucial indicators like the annual milk yield, per unit output, and protein and fat content, the somatic cell count precedes the majority of Chinese mega-cities. Further, intensification, mechanization, and normalization have become typical characteristics of dairy industrial development in the coastlands. Since 2014, the government has initiated various slurry treatment approaches due to land limitations and environmental pressure. A series of state-of-the-art technologies and facilities have been thoroughly established across the whole dairy farms. In terms of several baseline conditions involved in slurry collection and transportation, corresponding farmlands with crop cultivation, 24 representative dairy farms from dominant producing areas and from the whole range of the scale were selected (Figure 1). The basic profile is shown in Table 1, representing the general situation of dairy farming in Tianjin.

2.2. Process Selection and Sample Collection

Based on a previous investigation, three commonly used processes that were adopted by approximately 78% of the total dairy farms in Tianjin were selected [21]. These three processes were solid–liquid separation + anaerobic fermentation + biogas slurry storage + lagoon (Pa), solid–liquid separation + biogas slurry storage + lagoon (Pb), and solid–liquid separation + biogas slurry storage (Pc) (Figure 2). Both separation and multi-level storage are used in the rotation system, which is attributed to the shortage of land and funds. Additionally, the backwash system from the lagoon, storage tank, and even separation tank to the manure collection gutter for reducing the gross wastewater was also reported as satisfactory. However, appropriate adjustments were indicated depending on the weather or unexpected emergencies. In such cases, running processes of slurry flow are considerably complicated and flexible.
As shown in Figure 2, based on the manure collection gutter in dairy cow barns, the slurry flows through the slurry tank and separation system before dividing into three lines. These three are the anaerobic digestion reactor plus the storage tank and lagoon, the storage tank and lagoon, and the storage tank. The supernate from the last facility was recalled to wash away the gutter mixture to the slurry tank at regular intervals. Except for closed anaerobic digestion, other uncovered facilities included the manure collection gutter, slurry tank, separation tank, storage tanks, and lagoons, which were used as sampling points (Figure 2). Comprehensive and representative samples are a basic requirement in line with NIRS. Owing to the large span of concentration and drastic fluctuation, sample collection should run throughout almost every link of the treatment process, reconciling horizontal and vertical sites. There are 3 sampling steps as follows: 3 points of the same facilities were randomly selected to collect the slurry samples and mixed thoroughly in a 19 L bucket. Next, 400 mL samples were taken out and put into 500 mL bottles; then, the bottles were placed in a 52 L sample incubator and immediately sent back to the laboratory for TN and TP content testing and spectral scanning. A total of 715 slurry samples were artificially collected from October 2018 to January 2020 according to the national monitoring standard. The distribution of sampling amounts in the districts has also been shown in Figure 1.

2.3. Determination of Total Nitrogen and Total Phosphorus

The TN and TP (reference contents) were determined from slurry samples according to GB/T 11891-1989 [22] (Determination of Kjeldahl nitrogen) and GB/T 11893-1989 [23] (Ammonium Molybdate Spectrophotometric), respectively. The TN content was measured by an Automatic Kjeldahl nitrogen determination apparatus (Foss kjeltecTM 8400, Hørsholm, Denmark) after the sample was dissolved with acid at a high temperature. The TP content was calculated by absorbance after the sample was boiled in a high-pressure steam sterilization pot at a high temperature and detected at 700 nm by an ultraviolet–visible spectrophotometer (722E, Beijing, China).

2.4. Capture of Near-Infrared Spectroscopy Spectra

The NIR spectra were collected using a Fourier transform near-infrared spectrometer (FT-NIR) from the PerkinElmer Company of Wellesley, MA, USA. The FT-NIR spectrophotometer was equipped with an integrating sphere accessory and an InGaAs detector. The spectral range was between 4000 and 12,000 cm−1 with the following parameters: resolution of 8 cm−1, scanning interval of 2 cm−1, and 64 scanning times. The NIR spectrometer was equipped with a dedicated sample cup, ensuring that each sample occupies approximately two-thirds of the cup (2–3 mL) each time. Due to variations in solid content among different slurry samples, prior to sample analysis, the slurry sample in the collection bottle was thoroughly mixed. Using a disposable pipette with a 3 mL capacity, 2–3 mL of the sample was extracted from the middle of the collection bottle and placed in the sample cup. The sample cell with slurry was placed on the rotating platform of the integrating sphere and the spectra were measured.

2.5. Chemometrics Analysis

The spectra underwent pre-processing for normalization, baseline, and detrending to reduce spectral noise. Outliers were excluded from the dataset through the utilization of both leverage values and studentized residuals. Following this, the concentration gradient method, employing a one-in-four ratio, was employed to partition the data into calibration and validation sets, with 3/4 allocated to the calibration set and 1/4 to the validation set. The allocation of sample sizes for the calibration and validation sets for each model was delineated in Table 2. The model was constructed utilizing partial least squares (PLS) regression [24,25,26], and internal cross-validation was conducted using the Venetian blinds method. The optimal number of factors for modeling was determined based on the root mean square error of cross-validation (RMSECV). The chemometric analysis was performed by matlab PLS toolbox in 2016, and the figures in the paper were drawn by origin 2024. In general, the model performance of the validation set was evaluated by the root mean square error of prediction (RMSEP) and R2pred [27,28]. The RPD (the ratio of the standard deviation to the RMSEP) was also proposed to address the different ranges among measured samples. The ranges used and their inferences were as follows: RPD < 1.0, very poor; RPD = 1.0–1.4, poor; RPD = 1.4–1.8, fair; RPD = 1.8–2.0, good; RPD = 2.0–2.5, very good; RPD > 2.5, excellent [29,30].
To investigate the potential influence of the treatment process on model performance, two categories of models were devised: intra-process models and inter-process models (Figure 3). The former consisted of three single-process models (Pa, Pb, and Pc), three double-process combined models (Pab, Pac, and Pbc), and a triple-process fusion model (Pabc).
The remaining process could be predicted by utilizing data from two of the single-process models and merging them with data from both processes. Specifically, the corrective and validation datasets stemmed from distinct processes. For instance, in predicting Pa, the corrective dataset fused the data from both Pb and Pc, along with individual datasets from Pb and Pc, respectively.

3. Results

3.1. Total Nitrogen and Total Phosphorus Profiles of the Slurry from the Three Treatment Processes

The descriptive statistical analysis of TN and TP contents from the three treatment processes has been listed in Table 3 and Table 4, respectively. Looking at TN, the mean and standard deviation (SD) values of Pa (1943.0 ± 1321.2 mg L−1) exceeded those of Pb (1525.4 ± 1171.0 mg L−1) and Pc (1780.8 ± 1247.4 mg L−1). The Tukey variance analysis showed no significant difference between Pa and Pc or Pb and Pc, while the disparity between Pa and Pb was marked as significant (p < 0.05). This showed that the process influenced the TN content of the slurry. When the TP was analyzed, the mean and SD values of both Pa (92.2 ± 44.7 mg L−1) and Pc (92.9 ± 49.4 mg L−1) exceeded those of Pb (75.1 ± 39.1 mg L−1). Moreover, the content range of Pa and Pc exceeded that of Pb (6.1–190.2 mg L−1) as well. The Tukey test (p < 0.05) similarly identified no significant difference between Pa and Pc. Nonetheless, the contrasts between Pb and both Pa and Pc were pronounced. This illuminated the variation regularities of N and P contents in the slurry under different treatment processes.
Overall, the coefficient of variation (CV) for TN (>60%) was greater than that of TP (<60%) under the three different processes used. This may be attributed to the mass emission of ammonia in the open storage system while most of the TP in the sediments remains relatively stable [21]. Simultaneously, under the Pa treatment, the CV values for both TN (68%) and TP (48%) in the slurry were lower than the other two processes. This indicated that during the Pa process, there was the least variability in the TN and TP in the slurry, which made the Pa process the most stable. The CV value for the Pc process (70%) was lower than that of the Pb process (77%), indicating that the Pc process was more stable than Pb. Considering the treatment sites included in these processes, the biogas plant in Pa could retain nutrients in comparison to the fully open Pb process and Pc process, while the Pb model had one more open lagoon compared to the Pa model.
The above results showed that the three treatment processes significantly influenced the TN and TP contents in the slurry. These differences may have led to variations in the NIRS results of the slurry, thereby impacting the predictive performance of the NIRS model. Therefore, we next delved into the influence of the process used on the spectroscopy analysis of the slurry.

3.2. Near-Infrared Spectroscopy Spectral Characteristics of Slurry Samples under the Three Treatment Processes

Figure 4 illustrates the original NIRS of 715 samples in the range of 4000 to 12,000 cm−1. Four peaks emerged, positioned at 5180 cm−1, 6890 cm−1, 8660 cm−1, and 10,400 cm−1, accompanied by two valleys at 4512 cm−1 and 5920 cm−1, which corresponded to the vibrational responses of C-H and O-H multiplicity and combination frequencies in the slurry [31,32]. The mean spectra of the three different processes shown in Figure 4b were subjected to a Kruskal–Wallis test. The results revealed significant disparities between Pa and Pb, while differences between Pa and Pc and Pa and Pb were deemed non-significant (p < 0.05). Building upon the pronounced differences in the representative spectra from the three processes, we employed principal component analysis (PCA) to comprehensively analyze all spectra among the different processes. This further allowed for a more refined understanding of the influence exerted by the processes on the spectra obtained.

3.3. Principal Component Analysis of Slurry Spectra from the Three Treatment Processes

The NIR spectra of the slurry samples were further analyzed by PCA to comprehend the spectral responses from the variations in the TN and TP under the three treatment processes. In the principal component score chart, the longer the samples’ distances are, the more different the properties and components, and vice versa. Based on the sample distribution on the score charts, the variations in the component content of slurry samples can be inferred along with three different treatment processes (Figure 5). It was found that the initial two principal components, namely PC1 and PC2, included 98.4%, 98.6%, 99.2%, and 96.9% of the variance information in Pa (Figure 5b), Pb (Figure 5c), Pc (Figure 5d), and Pabc (mixed data, Figure 5a), respectively. This showed that the spectral information extracted varied among three different processes. For instance, PC1 of Pc, at 93.1%, surpassed the PC1 of Pb and Pa. This indicated that the spectral data of the Pc process exhibited greater similarity in comparison to the Pa and Pb processes. Upon amalgamating the spectral data from the three processes, PC1 declined to 85.1%, while PC2 increased to 11.8%. This indicated that spectral differences were present among the various processes, and the spectral fusion magnified these differences.
One intriguing observation was that the variations in PC1 could be correlated with the actual process links. The order of PC1 was Pc > Pb > Pa. Except for the identical front-end links in the three processes, Pc had only one subsequent treatment link, Pb involved two treatment links, and Pa had three treatment links. This observation can possibly be attributed to the greater number of links involved in the treatment process, leading to a more intense transformation of the form of the slurry. Consequently, a more diverse range of spectra was captured, resulting in larger differences among the obtained spectra, which ultimately yielded a smaller extracted PC1.
The above results indicated disparities in the spectral features of the slurry under different processes. Considering the process influence was crucial for the accuracy of NIRS predictions of the TN and TP contents in the slurry. This factor may also contribute to the subsequent alternations in the performances of the model.

3.4. Near-Infrared Spectroscopy Model Performances of Slurry TN and TP

The different slurry treatment processes documented variations in the contents of TN, TP, and the slurry spectra. It was crucial to gain a comprehensive understanding of the degree of the influence on the predictive performance from various processes to establish NIRS models capable of efficiently predicting the TN and TP contents in the slurry under different treatment processes.

3.4.1. Intra-Process Near-Infrared Spectroscopy Model (Intra-Model)

The prediction results of the TN and TP by single or combined processes of Pa, Pb, and Pc are shown in Figure 6 and Figure 7, respectively. As is shown in Figure 6a, Pc—Pc means that we use the data under process Pc to build a model and predict the data under process Pc. This shows that the data for the calibration set and validation set come from the same process. The performance of the prediction had an RMSEP = 469.23 mg/L, an R2pred = 0.81, and an RPD = 2.28.
For the TN, each model presented a very good performance with the following ranges: 0.78–0.83 for R2pred, 406.73–570.73 mg L−1 for RMSEP, and 2.10–2.38 for RPD. Further, the prediction results of both the Pac and Pa were optimal with R2pred between 0.82 and 0.83, RMSEP between 491.49 and 497.54 mg L−1, and RPD between 2.38 and 2.31, respectively. These results demonstrated that the models were applicable for predicting the TN of the slurry. However, variations in the predictive performance among different models were observed due to the influence of the processes. For the TP, the ranges of evaluative indicators were as follows: the R2pred was 0.55–0.79, the RMSEP was 19.24–27.09 mg L−1, and the RPD was 1.31–2.30, respectively. Subsequently, the prediction results of Pab and Pa were found to be optimal, with an R2pred of 0.77 and 0.80, an RMSEP of 20.49 and 19.24 mg L−1, and an RPD of 2.07 and 2.30, respectively.
The Pabc prediction results of both TN and TP performed relatively well. The performance of TN as evaluated by the factors was as follows. The R2pred was 0.80, RMSEP was 521.17 mg L−1, and RPD was 2.20. The model was deemed to be good. However, the prediction results of TP were relatively poor with the following results: R2pred = 0.63, RMSEP = 26.84 mg L−1, and RPD = 1.64. The model was applicable just for screening. The different intra-models not only simply mitigated the impact of the processes but also met the demand for different slurry movement pathways. It was essential to take the treatment processes into account when employing NIRS in practical applications.

3.4.2. Inter-Process Near-Infrared Spectroscopy Model (Inter-Model)

Figure 8 and Figure 9 compare the prediction results of these inter-models for the TN and TP, respectively. As is shown in Figure 8a, Pc—Pa means that we use the data under process Pc to build a model and predict the data under process Pa. This indicates that the data for the calibration set and validation set come from different processes. The performance of the prediction had an RMSEP = 817.67 mg/L, an R2pred = 0.52, and an RPD = 1.4.
For the TN, the predictive performance of all the models was unsatisfactory with the following results: R2pred = 0.44–0.80, RMSEP = 450.35–1005.14 mg L−1, and RPD = 1.14–2.06. These models were found to be inadequate for the practical detection of TN in the slurry. The prediction effects for TP were even worse when compared to the TN results with the following results: R2pred was 0.32–0.65, RMSEP was 29.95–38.21 mg L−1, and RPD was 0.91–1.50. To summarize, the combined prediction showed a poor performance, which showed its limited suitability for practical implementation.

4. Discussion

This study showed that the intra-models documented superior outcomes compared to the constructed inter-models. This indicated that the effect caused by the treatment processes could be addressed and alleviated. Compared to the effects of single-process models, the model established by Pa had the best prediction effect on N and P. The reason for this result may be that the sample size of Pa (358) was more than that of Pb (225) and Pc (132). The larger the sample size of the model, the more useful information can be extracted and the better the prediction effect of the model prediction. Comparing the effect of both double-process combined models and the triple-process model to the single-process models, the prediction effect of the latter was better than that of the former. But the sample size of the latter was smaller than that of the former, which indicated that the quality of the dataset of the model will also affect the prediction performance of the model.
Li et al. [33] and Mathieu et al. [34] mentioned the concept of local models and global models. The former referred to the calibration of samples with relatively constant composition, such as water samples from the same site as the wastewater treatment plant [35] or surface water [36]. The latter referred to the calibration of a model sampled from different sampling points in the target area, which should be generalized to apply one of the models to more monitoring sites. This was equivalent to the single-process model we built as the local model, while the model built by the triple-process fusion model was a global model. The larger the sample size of the local model was, the better the model would be. However, after the fusion of slurry samples under different processes, the prediction effect of the model became worse, indicating that the difference in slurry treatment processes would affect the prediction performance of the N and P models. Although the fusion of different processes lowers the prediction accuracy of the single-process model, it expands the applicability of the model. Therefore, diversified models become necessary to meet the requirements of different customers. Accordingly, considering the treatment processes would also be in favor of deriving multiplex models to be available for complex scenarios in the field. That would be conducive to efficient and sustainable use of slurry resources.
Based on the results of the intra-models for TN, the Pa process in single-process models performed optimally. Among the double-process models, Pac was the best, while Pbc was the poorest. Considering the results of the intra-models for TP, Pa in single-process models was the best, while Pb was the worst performer. Among the double-process models, for this, Pab showed the best result, while Pbc showed the poorest result. It should be noted that Pa exhibited the best performance for both TN and TP models compared to the Pbc, which showed the poorest outcome. It could be inferred that the performance of the NIRS models may be related to the stability of the slurry treatment system. Further, nutrients could be retained during the biogas phase of the Pa process in comparison to the fully open systems of Pb and Pc. This finding was also in line with the values of CV at 68% and 48% for the TN and TP contents, respectively, which were smaller than those of Pb and Pc. Further, Pb contained two open links while Pc included only one and the latter was relatively more stable than the former. When the slurry treatment process was more stable, the CV was much smaller and the NIRS model displayed a better performance. Meanwhile, combined with the PCA results of slurry spectra under different treatment processes, it can be found that the size order on PC1 was Pc > Pb > Pa. Combined with the above discussion, it can be seen that the size of PC1 was related to the length of slurry treatment flow. The more sites involved in the treatment system, the smaller PC1 extracted, and the better the model performed.
Additionally, TN models showed a better profile than TP models among all the models. While data regarding the nitrogen in the slurry could be obtained based on the characteristics of NIRS, various forms of phosphorus in the slurry were either not absorbed in the NIRS region or exhibited a low absorption amount. Previous studies also showed that the detection of phosphorus by NIRS was indirectly predicted by selecting plausible correlations between phosphorus and other substances [37,38]. For instance, Cabassi et al. [14] and Sørensen et al. [39] both demonstrated that the predictive performance of the NIRS method for TP was worse than that for TN in cattle slurry. These findings align with the outcomes of our current study. Hence, incorporating the feature band extraction and other modeling algorithms may further improve the predictive performance of TP models.

5. Conclusions

The study demonstrated that different treatment processes remarkably influenced TN and TP contents, spectra, and NIRS model performances of the processed slurry. Different intra-models were constructed to not only mitigate such effects but also meet the requirement of long-term stable monitoring in intricate scenes. The optimal fusion model and single model were Pac and Pa among the TN intra-models, respectively. For the TP, the optimum results were shown by Pab’s fusion model and Pa’s single model, respectively. Process stability was closely connected with the nutrient variation and the predicted precision. Overall, the performance of the TN models was better than that of the TP models. Diversified model establishment provides highly alternative technologies for tracking, monitoring, and evaluating the slurry treatment process and its effects. Taking the treatment processes into account will be conducive to not simply promoting sustainable monitoring and measurement but also guiding land application in the field, precisely and efficiently. This will be helpful for improving resource utilization efficiency and agricultural environmental conservation.

Author Contributions

Conceptualization, R.Z.; methodology, M.L. and Z.Y.; validation, R.Z. and Z.Y.; formal analysis, M.L. and S.L.; investigation, D.S. and S.L.; data curation, M.L. and D.S.; writing—original draft preparation, M.L. and D.S.; writing—review and editing, M.L. and R.Z.; visualization, M.L.; supervision, Z.Y.; project administration, K.Z.; funding acquisition, K.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Yunnan Science & Technology Planning Project, 202402AE090032; Innovation Team of Tianjin Dairy (Sheep) Research System, ITTCRS202100007; Central Public-interest Scientific Institution Basal Research Fund, Y2022CG09; and Innovation Project of Agricultural Science and Technology by CAAS.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The raw data used in this study are publicly available and can be obtained upon reasonable request to the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Ministry of Ecology and Environment of the People’s Republic of China; National Bureau of Statistics; Ministry of Agriculture and Rural Affairs of the People’s Republic of China. Manual on Accounting Methods and Coefficients of Pollutant Discharge from Statistical Survey of Emission Sources; Ministry of Ecology and Environment of the People’s Republic of China: Beijing, China, 2021.
  2. Nie, S.; Hu, L. China Animal Husbandry and Veterinary Yearbook; China Agriculture Press: Beijing, China, 2021. [Google Scholar]
  3. Bai, Z.H.; Ma, L.; Oenema, O.; Chen, Q.; Zhang, F.S. Nitrogen and phosphorus use efficiencies in dairy production in China. J. Environ. Qual. 2013, 42, 990–1001. [Google Scholar] [CrossRef] [PubMed]
  4. Hou, Y.; Velthof, G.L.; Lesschen, J.P.; Staritsky, I.G.; Oenema, O. Nutrient recovery and emissions of ammonia, nitrous oxide, and methane from animal manure in Europe: Effects of manure treatment technologies. Environ. Sci. Technol. 2017, 51, 375–383. [Google Scholar] [CrossRef] [PubMed]
  5. Guo, R.; Jin, H.; Wu, H.; Huang, H.; Ye, X.; Xu, Y.; Zhang, M. Total content of heavy metals and their chemical form changes in multilevel wastewater treatment system in intensive swine farm. Trans. Chin. Soc. Agric. Eng. 2018, 34, 210–216. [Google Scholar]
  6. Li, D.; Hu, J.; Dong, Y.; Jing, H. Life Cycle Assessment of Two Manure Treatment Modes in Intensive Dairy Farms. J. Ecol. Rural Environ. 2021, 37, 257–264. [Google Scholar] [CrossRef]
  7. Wang, Q.H.; Yu, L.J.; Liu, Y.; Lin, L.; Lu, R.G.; Zhu, J.P.; He, L.; Lu, Z.L. Methods for the detection and determination of nitrite and nitrate: A review. Talanta 2017, 165, 709–720. [Google Scholar] [CrossRef] [PubMed]
  8. Li, D.; Wang, T.; Li, Z.; Xu, X.B.; Wang, C.; Duan, Y.Q. Application of graphene-based materials for detection of nitrate and nitrite in water—A review. Sensors 2020, 20, 54. [Google Scholar] [CrossRef]
  9. Ribeiro, J.A.; Silva, F.; Pereira, C.M. Electrochemical sensing of ammonium ion at the water/1,6-dichlorohexane interface. Talanta 2012, 88, 54–60. [Google Scholar] [CrossRef]
  10. Althaus, B.; Papke, G.; Sundrum, A. Use of near infrared reflectance spectroscopy to assess nitrogen and carbon fractions in dairy cow feces. Anim. Feed Sci. Technol. 2013, 185, 53–59. [Google Scholar] [CrossRef]
  11. Jancewicz, L.J.; Swift, M.L.; Penner, G.B.; Beauchemin, K.A.; Koenig, K.M.; Chibisa, G.E.; He, M.L.; McKinnon, J.J.; Yang, W.Z.; McAllister, T.A. Development of NIRS calibrations to estimate fecal composition and nutrient digestibility in beef cattle. Can. J. Anim. 2017, 97, 51–64. [Google Scholar] [CrossRef]
  12. Shen, X.; Yang, Z.; Huang, G.; Han, L. Near infrared spectroscopy detection of copper in pig manure and the spectral basis of the analysis. J. Near Infrared Spectrosc. 2014, 22, 305–312. [Google Scholar] [CrossRef]
  13. Tolleson, D.R.; Angerer, J.P. The application of near infrared spectroscopy to predict faecal nitrogen and phosphorus in multiple ruminant herbivore species. Rangel. J. 2021, 42, 415–423. [Google Scholar] [CrossRef]
  14. Cabassi, G.; Cavalli, D.; Fuccella, R.; Gallina, P.M. Evaluation of four NIR spectrometers in the analysis of cattle slurry. Biosyst. Eng. 2015, 133, 1–13. [Google Scholar] [CrossRef]
  15. Sun, D.; Yang, R.; Li, M.; Mou, M.; Zhang, K.; Zhao, R. Influences of spring and autumn on the nitrogen and phosphorus contents of the slurry predicted by near-infrared spectrum model on dairy farms. Trans. Chin. Soc. Agric. Eng. 2020, 36, 197–205. [Google Scholar] [CrossRef]
  16. Finzi, A.; Oberti, R.; Negri, A.S.; Perazzolo, F.; Cocolo, G.; Tambone, F.; Cabassi, G.; Provolo, G. Effects of measurement technique and sample preparation on NIR spectroscopy analysis of livestock slurry and digestates. Biosyst. Eng. 2015, 134, 42–54. [Google Scholar] [CrossRef]
  17. Kuda-Malwathumullage, C.P.; Small, G.W. Temperature correction strategy for improving concentration predictions with near-infrared spectra of aqueous-based samples. Anal. Chim. Acta 2020, 1095, 20–29. [Google Scholar] [CrossRef] [PubMed]
  18. Tamburini, E.; Castaldelli, G.; Ferrari, G.; Marchetti, M.G.; Pedrini, P.; Aschonitis, V.G. Onsite and online FT-NIR spectroscopy for the estimation of total nitrogen and moisture content in poultry manure. Environ. Technol. 2015, 36, 2285–2294. [Google Scholar] [CrossRef] [PubMed]
  19. Khalil, T.M.; Stöckle, C.O.; Carlson, B.R.; Uslar-Valle, N.; Nelson, R.L.; Frear, C.S.; Ma, J.; Higgins, S.S.; Leytem, A.B.; Dungan, R.S. Dairy-CropSyst: Gaseous emissions and nutrient fate modeling tool. Comput. Electron. Agric. 2019, 162, 962–978. [Google Scholar] [CrossRef]
  20. Zhang, P.; Ding, J.; Meng, H.; Zhao, L.; Shen, Y.; Liu, H. Characteristics of nitrogen transformation in the process of acidification and storage of cow slurry. Trans. Chin. Soc. Agric. Eng. 2020, 36, 212–218. [Google Scholar] [CrossRef]
  21. Li, M.; Sun, D.; Mou, M.; Liu, H.; Zhang, K.; Yang, R.; Zhao, R. Variation characteristics and rules of nitrogen and phosphorus contents throughout the slurry movement from scaled dairy farms in Tianjin. Trans. Chin. Soc. Agric. Eng. 2020, 36, 27–33. [Google Scholar] [CrossRef]
  22. GB/T 11891-1989; Water Quality-Determination of Kjeldahl Nitrogen. China Quality Inspection Press: Beijing, China, 1989.
  23. GB/T 11893-1989; Water Quality-Determination of Total Phosphorus-Ammonium Molybdate Spectrophotometric Method. China Environmental Publishing Co., Ltd.: Beijing, China, 1989.
  24. Alassali, A.; Fiore, S.; Kuchta, K. Assessment of plastic waste materials degradation through near infrared spectroscopy. Waste Manag. 2018, 82, 71–81. [Google Scholar] [CrossRef]
  25. Fitamo, T.; Triolo, J.M.; Boldrin, A.; Scheutz, C. Rapid biochemical methane potential prediction of urban organic waste with near-infrared reflectance spectroscopy. Water Res. 2017, 119, 242–251. [Google Scholar] [CrossRef] [PubMed]
  26. Liu, J.; Jin, S.; Bao, C.; Sun, Y.; Li, W. Rapid determination of lignocellulose in corn stover based on near-infrared reflectance spectroscopy and chemometrics methods. Bioresour. Technol. 2021, 321, 124449. [Google Scholar] [CrossRef] [PubMed]
  27. Malley, D.F.; McClure, C.; Martin, P.D.; Buckley, K.; McCaughey, W.P. Compositional analysis of cattle manure during composting using a field-portable near-infrared spectrometer. Commun. Soil Sci. Plant Anal. 2005, 36, 455–475. [Google Scholar] [CrossRef]
  28. Vaudour, E.; Gilliot, J.M.; Bel, L.; Lefevre, J.; Chehdi, K. Regional prediction of soil organic carbon content over temperate croplands using visible near-infrared airborne hyperspectral imagery and synchronous field spectra. Int. J. Appl. Earth Obs. Geoinf. 2016, 49, 24–38. [Google Scholar] [CrossRef]
  29. Rossel, R.A.V.; Walvoort, D.J.J.; McBratney, A.B.; Janik, L.J.; Skjemstad, J.O. Visible, near infrared, mid infrared or combined diffuse reflectance spectroscopy for simultaneous assessment of various soil properties. Geoderma 2006, 131, 59–75. [Google Scholar] [CrossRef]
  30. Douglas, R.K.; Nawar, S.; Alamar, M.C.; Mouazen, A.M.; Coulon, F. Rapid prediction of total petroleum hydrocarbons concentration in contaminated soil using vis-NIR spectroscopy and regression techniques. Sci. Total Environ. 2018, 616, 147–155. [Google Scholar] [CrossRef] [PubMed]
  31. Han, L.; Sun, Y.; Wang, Y.; Fu, H.; Duan, C.; Wang, M.; Cai, W.; Shao, X. Ultra-high resolution near-infrared spectrum by wavelet packet transform revealing the hydrogen bond interactions. Spectrochim. Acta Part A 2023, 289, 122233. [Google Scholar] [CrossRef] [PubMed]
  32. Maeda, H.; Ozaki, Y.; Tanaka, M.; Hayashi, N.; Kojima, T. Near infrared spectroscopy and chemometrics studies of temperature-dependent spectral variations of water: Relationship between spectral changes and hydrogen bonds. J. Near Infrared Spectrosc. 1995, 3, 191–201. [Google Scholar] [CrossRef]
  33. Li, P.; Qu, J.; He, Y.; Bo, Z.; Pei, M. Global calibration model of UV-Vis spectroscopy for COD estimation in the effluent of rural sewage treatment facilities. RSC Adv. 2020, 10, 20691–20700. [Google Scholar] [CrossRef]
  34. Lepot, M.; Torres, A.; Hofer, T.; Caradot, N.; Gruber, G.; Aubin, J.B.; Bertrand-Krajewski, J.L. Calibration of UV/Vis spectrophotometers: A review and comparison of different methods to estimate TSS and total and dissolved COD concentrations in sewers, WWTPs and rivers. Water Res. 2016, 101, 519–534. [Google Scholar] [CrossRef]
  35. Carré, E.; Pérot, J.; Jauzein, V.; Lin, L.; Lopez-Ferber, M. Estimation of water quality by UV/Vis spectrometry in the framework of treated wastewater reuse. Water Sci. Technol. 2017, 76, 633–641. [Google Scholar] [CrossRef]
  36. Li, J.; Tong, Y.; Guan, L.; Wu, S.; Li, D. Optimization of COD determination by UV–vis spectroscopy using PLS chemometrics algorithms. Optik 2018, 174, 591–599. [Google Scholar] [CrossRef]
  37. Chen, L.; Xing, L.; Han, L. Review of the Application of Near-Infrared Spectroscopy Technology to Determine the Chemical Composition of Animal Manure. J. Environ. Qual. 2013, 42, 1015–1028. [Google Scholar] [CrossRef]
  38. Saeys, W.; Mouazen, A.M.; Ramon, H. Potential for onsite and online analysis of pig manure using visible and near infrared reflectance spectroscopy. Biosyst. Eng. 2005, 91, 393–402. [Google Scholar] [CrossRef]
  39. Sørensen, L.K.; Sørensen, P.; Birkmose, T.S. Application of reflectance near infrared spectroscopy for animal slurry analyses. Soil Sci. Soc. Am. J. 2007, 71, 1398–1405. [Google Scholar] [CrossRef]
Sustainability 16 05083 g001
Figure 1. Distribution of dairy farms and sampling amounts.
Figure 1. Distribution of dairy farms and sampling amounts.
Sustainability 16 05083 g001
Sustainability 16 05083 g002
Figure 2. Flow charts of slurry treatment processes in 24 dairy farms. (The manure collection gutter functioned by aggregating the manure inside the barn. The slurry tank acted as the central point for collecting all the slurry in the dairy farm. The separation tank served as a temporary storage facility for the slurry after solid–liquid separation. Storage tanks and lagoons functioned as the final storage facilities prior to the land application of slurry).
Figure 2. Flow charts of slurry treatment processes in 24 dairy farms. (The manure collection gutter functioned by aggregating the manure inside the barn. The slurry tank acted as the central point for collecting all the slurry in the dairy farm. The separation tank served as a temporary storage facility for the slurry after solid–liquid separation. Storage tanks and lagoons functioned as the final storage facilities prior to the land application of slurry).
Sustainability 16 05083 g002
Sustainability 16 05083 g003
Figure 3. The strategy employed to establish the model.
Figure 3. The strategy employed to establish the model.
Sustainability 16 05083 g003
Sustainability 16 05083 g004
Figure 4. Near-infrared diffuse reflection spectra of slurry samples ((a) all spectra of 715 samples; (b) the mean spectra of the three processes, respectively).
Figure 4. Near-infrared diffuse reflection spectra of slurry samples ((a) all spectra of 715 samples; (b) the mean spectra of the three processes, respectively).
Sustainability 16 05083 g004
Sustainability 16 05083 g005
Figure 5. The principal component analysis of slurry spectra under three treatment processes.
Figure 5. The principal component analysis of slurry spectra under three treatment processes.
Sustainability 16 05083 g005
Sustainability 16 05083 g006
Figure 6. NIRS prediction results of TN intra-models. ((a) Three single-process models; (b) three double-process combined models and a triple-process fusion model).
Figure 6. NIRS prediction results of TN intra-models. ((a) Three single-process models; (b) three double-process combined models and a triple-process fusion model).
Sustainability 16 05083 g006
Sustainability 16 05083 g007
Figure 7. NIRS prediction results of TP intra-model. ((a) Three single-process models; (b) three double-process combined models and a triple-process fusion model).
Figure 7. NIRS prediction results of TP intra-model. ((a) Three single-process models; (b) three double-process combined models and a triple-process fusion model).
Sustainability 16 05083 g007
Sustainability 16 05083 g008
Figure 8. NIRS prediction results of TN inter-model. ((a) Three models for predicting Pa; (b) three models for predicting Pb; (c) three models for predicting Pc).
Figure 8. NIRS prediction results of TN inter-model. ((a) Three models for predicting Pa; (b) three models for predicting Pb; (c) three models for predicting Pc).
Sustainability 16 05083 g008
Sustainability 16 05083 g009
Figure 9. NIRS prediction results of TP inter-model. ((a) Three models for predicting Pa; (b) three models for predicting Pb; (c) three models for predicting Pc).
Figure 9. NIRS prediction results of TP inter-model. ((a) Three models for predicting Pa; (b) three models for predicting Pb; (c) three models for predicting Pc).
Sustainability 16 05083 g009
Table 1. Overview of 24 typical dairy farms in Tianjin.
Table 1. Overview of 24 typical dairy farms in Tianjin.
DistrictFarm NumberBreeding Stock/HeadageSlurry Treatment Process
Ninghe3850–2830Pa 1; Pb 2; Pc 3
Binhai5450–2640Pa 1; Pb 2
Wuqing8360–1680Pa 1; Pb 2; Pc 3
Jinghai22280–2850Pa 1; Pb 2
Beichen1300–600Pb 2
Baodi5440–5040Pa 1; Pb 2; Pc 3
1 Pa represents solid–liquid separation + anaerobic fermentation + biogas slurry storage + lagoon; 2 Pb represents solid–liquid separation + biogas slurry storage + lagoon; 3 Pc represents solid–liquid separation + biogas slurry storage.
Table 2. Sample information of correction set and validation set divided by concentration gradient method (the number in the table represents the sample number of sets).
Table 2. Sample information of correction set and validation set divided by concentration gradient method (the number in the table represents the sample number of sets).
IndicatorTNTP
ModelsCalibration SetValidation SetCalibration SetValidation Set
Pa—Pa2498125484
Pb—Pb1524816252
Pc—Pc90299732
Pab—Pab40313540084
Pac—Pac340112356118
Pbc—Pbc2568625284
Pabc—Pabc486159488161
Pbc—Pa2568125284
Pb—Pa1528116284
Pc—Pa90819784
Pac—Pb3404835652
Pa—Pb2494825452
Pc—Pb90489752
Pab—Pc4032940032
Pa—Pc2492925432
Pb—Pc1522916232
Table 3. Descriptive statistical analysis of TN contents.
Table 3. Descriptive statistical analysis of TN contents.
Reference
Contents (TN)		NumberMean ± Std (mg L−1)Min–Max (mg L−1)Median (mg L−1 L)Coefficient of Variation (CV)KurtosisSkewness
Pa3581943.0 ± 1321.245.1–6768.31643.868%0.210.83
Pb2251525.4 ± 1171.041.4–5262.31335.277%0.720.99
Pc1321780.8 ± 1247.420.7–5135.51692.970%−0.720.31
Table 4. Descriptive statistical analysis of TP contents.
Table 4. Descriptive statistical analysis of TP contents.
Reference
Contents (TP)		NumberMean ± Std (mg L−1)Min–Max (mg L−1)Median (mg L−1)Coefficient of
Variation (CV)		KurtosisSkewness
Pa35892.2 ± 44.74.8–209.788.548%−0.980.11
Pb22575.1 ± 39.16.1–190.267.752%−0.330.46
Pc13292.9 ± 49.40.9–208.195.053%−0.850.13
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Li, M.; Sun, D.; Liu, S.; Zhang, K.; Zhao, R.; Yang, Z. Dynamic Prediction of Total N and P Contents in Slurry from Dairy Farms under Different Treatment Processes Using Near-Infrared Spectroscopy. Sustainability 2024, 16, 5083. https://doi.org/10.3390/su16125083

AMA Style

Li M, Sun D, Liu S, Zhang K, Zhao R, Yang Z. Dynamic Prediction of Total N and P Contents in Slurry from Dairy Farms under Different Treatment Processes Using Near-Infrared Spectroscopy. Sustainability. 2024; 16(12):5083. https://doi.org/10.3390/su16125083

Chicago/Turabian Style

Li, Mengting, Di Sun, Shengbo Liu, Keqiang Zhang, Run Zhao, and Zengjun Yang. 2024. "Dynamic Prediction of Total N and P Contents in Slurry from Dairy Farms under Different Treatment Processes Using Near-Infrared Spectroscopy" Sustainability 16, no. 12: 5083. https://doi.org/10.3390/su16125083

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop