Next Article in Journal
Development of a Diagnostic Algorithm for Detecting Freezing Precipitation from ERA5 Dataset: An Adjustment to the Far East
Previous Article in Journal
“Taking Action in Community Is Much, Much Preferable to Doing It Alone”: An Examination of Multi-Level Facilitators of and Barriers to Sustained Collective Climate Change Activism Among US Residents
Previous Article in Special Issue
A Comparison between Radar Variables and Hail Pads for a Twenty-Year Period
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Climate
Volume 12
Issue 12
10.3390/cli12120223
climate-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

Mapping Methane—The Impact of Dairy Farm Practices on Emissions Through Satellite Data and Machine Learning

by
Hanqing Bi
1,2 and
Suresh Neethirajan
2,3,*
1
Faculty of Computer Science, Dalhousie University, 6050 University Avenue, Halifax, NS B3H 4R2, Canada
2
Faculty of Mathematics, University of Waterloo, 200 University Ave W, Waterloo, ON N2L 3G1, Canada
3
Faculty of Agriculture, Agricultural Campus, Dalhousie University, P.O. Box 550, Truro, NS B2N 5E3, Canada
*
Author to whom correspondence should be addressed.
Climate 2024, 12(12), 223; https://doi.org/10.3390/cli12120223
Submission received: 13 November 2024 / Revised: 6 December 2024 / Accepted: 12 December 2024 / Published: 15 December 2024
(This article belongs to the Special Issue Applications of Smart Technologies in Climate Risk and Adaptation)
Browse Figures
Versions Notes

Abstract

:
Methane emissions from dairy farms are a significant driver of climate change, yet their relationship with farm-specific practices remains poorly understood. This study employs Sentinel-5P satellite-derived methane column concentrations as a proxy to examine emission dynamics across 11 dairy farms in Eastern Canada, using data collected between January 2020 and December 2022. By integrating advanced analytics, we identified key drivers of methane concentrations, including herd genetics, feeding practices, and management strategies. Statistical tools such as Variance Inflation Factor (VIF) and Principal Component Analysis (PCA) addressed multicollinearity, stabilizing predictive models. Machine learning approaches—Random Forest and Neural Networks—revealed a strong negative correlation between methane concentrations and the Estimated Breeding Value (EBV) for protein percentage, demonstrating the potential of genetic selection for emissions mitigation. Our approach refined concentration estimates by integrating satellite data with localized atmospheric modeling, enhancing accuracy and spatial resolution. These findings highlight the transformative potential of combining satellite observations, machine learning, and farm-level characteristics to advance sustainable dairy farming. This research underscores the importance of targeted breeding programs and management strategies to optimize environmental and economic outcomes. Future work should expand datasets and apply inversion modeling for finer-scale emission quantification, advancing scalable solutions that balance productivity with ecological sustainability.

1. Introduction

The escalating threat of climate change has propelled greenhouse gas emissions to the forefront of global environmental concerns. Among these, methane has emerged as a particularly potent contributor to global warming, with its impact significantly exceeding that of carbon dioxide on a per-molecule basis [1,2]. The urgency to mitigate methane emissions has intensified, with atmospheric concentrations rising alarmingly, reflecting a fourfold increase in global emissions over recent decades [3,4]. This surge poses a critical challenge to international climate goals, particularly those outlined in the Paris Agreement.
The dairy sector is identified as a substantial source of anthropogenic methane emissions, significantly contributing to the agricultural sector’s greenhouse gas footprint [5,6]. The impact of methane from dairy production is particularly concerning due to its global warming potential—approximately 28 times that of carbon dioxide over a 100-year period [7,8]. This necessitates targeted mitigation strategies within the industry.
Primary methane sources in dairy farming include enteric fermentation and manure management practices [9,10,11]. The biological complexities underlying these emissions present both challenges and opportunities for mitigation. Understanding and addressing these sources are essential not only for climate change mitigation but also for promoting sustainable agricultural practices that meet global food demands while minimizing environmental impact.
The intricate relationship between dairy farm characteristics and methane emissions demands a multifaceted mitigation approach. Factors such as herd genetics, feeding strategies, and management practices significantly influence methane output, yet the relative importance and interactions among these factors remain poorly understood, hindering the development of effective, targeted mitigation strategies. Recent advancements in satellite technology and data analytics have opened new avenues for monitoring and quantifying methane emissions at unprecedented scales. The launch of the Sentinel-5P satellite, equipped with the Tropospheric Monitoring Instrument (TROPOMI), has revolutionized our capacity to detect and measure atmospheric methane concentrations with high spatial and temporal resolution [12,13]. This technology offers a unique opportunity to bridge the gap between farm-level practices and regional methane concentrations, potentially unveiling patterns previously obscured by the limitations of ground-based measurements.
The integration of satellite data with ground-based measurements and atmospheric transport models represents a significant advancement in creating comprehensive methane emission estimates [4,14,15]. This approach enhances our understanding of methane dynamics across various spatial scales and provides a powerful tool for validating and refining emission inventories. The capability to detect and quantify emissions from space offers unprecedented potential for identifying and addressing methane hotspots, particularly in dairy farming regions.
The advent of advanced machine learning techniques further amplifies the potential of satellite-based methane monitoring. These sophisticated algorithms can analyze vast datasets to identify patterns and anomalies that may indicate emission sources or trends [16,17]. Applying machine learning to satellite data holds promise for enhancing the detection and quantification of methane emissions, potentially revealing insights that traditional analytical methods might overlook. Inversion modeling techniques have emerged as vital tools in translating satellite-observed methane concentrations into actual emission estimates [18]. This methodology allows researchers to infer emission strengths and locations from atmospheric measurements, providing valuable insights into the relationship between farm practices and regional methane concentrations. The synergy between multiple satellite platforms, such as Sentinel-5P and GOSAT, could further enhance monitoring capabilities [19,20].
Continuous satellite monitoring enables the detection of long-term methane emission trends, essential for assessing the effectiveness of mitigation strategies and policies. This capability is particularly relevant for the dairy sector, where changes in farming practices may take time to manifest in observable emission reductions. Long-term satellite observations can provide critical data for evaluating various interventions and guiding future policy decisions.
The rise of artificial intelligence and machine learning offers new possibilities for analyzing emission structures and optimizing mitigation strategies in the dairy industry. By assigning varying importance to different farm characteristics, these technologies can potentially identify innovative solutions for reducing methane emissions from dairy farms [21,22,23]. This data-driven approach could lead to more targeted and effective interventions tailored to specific conditions of individual farms or regions.
Despite these advancements, challenges persist in accurately attributing methane emissions to specific sources and in developing effective mitigation strategies for the dairy sector. The complex interplay among farm management practices, animal genetics, and environmental factors complicates efforts to pinpoint the most impactful interventions. Moreover, potential trade-offs between methane reduction strategies and other aspects of dairy sustainability—such as animal welfare and economic viability—must be carefully considered.
This study aims to address these challenges by exploring the correlation between dairy farm factors and satellite-derived methane concentrations in Eastern Canada. By leveraging advanced data analytics and machine learning techniques, we seek to uncover insights that can inform breeding programs, management practices, and policy decisions aimed at reducing the dairy sector’s environmental impact. Our approach involves a comprehensive analysis of various farm characteristics, including herd genetics, feeding practices, and management strategies, to identify the most significant factors influencing methane emissions.

2. Related Work

Recent advancements in satellite technology and data analytics have revolutionized the monitoring and analysis of methane emissions from the dairy sector. The Mooanalytica research group at Dalhousie University has made notable contributions in this field, leveraging satellite data to assess methane emissions from Canadian dairy farms and processors.

2.1. Satellite-Based Methane Monitoring

Satellite data collection has emerged as a reliable and cost-effective method for gathering large-scale environmental data [24,25]. The Sentinel-5P satellite, equipped with TROPOMI, has proven particularly instrumental in this regard, accurately collecting data on various atmospheric gases, including methane, through advanced spectroscopy methods [13]. Orbiting at an altitude of 824 km, Sentinel-5P provides daily global coverage with a spatial resolution of 7 km × 7 km, ensuring consistent data collection [26,27]. This technology has been applied in numerous studies, particularly in greenhouse gas emissions research, with recent efforts focusing on integrating artificial intelligence for enhanced analysis [28,29].

2.2. Application to the Canadian Dairy Sector

Recent studies utilizing Sentinel-5P data have analyzed methane emissions from over 575 dairy farms and 384 dairy processors across Canada over an eight-year period, revealing significant seasonal and provincial variations in emissions, with autumn emissions peaking and Ontario exhibiting the highest overall emissions [30]. Building on this work, an AI-driven benchmarking tool for emission reduction in Canadian dairy farms has been developed, integrating satellite-derived methane emission data with advanced machine learning technologies and geospatial analysis [31,32]. The integration of satellite data with machine learning models has significantly enhanced the detection and quantification of methane emissions (Figure 1). Studies have demonstrated how advanced machine learning techniques applied to satellite data improve the identification of emission patterns and anomalies [16,17]. Inversion modeling techniques have been crucial in translating satellite-observed methane concentrations into actual emission estimates [18], while the synergy between multiple satellite platforms has further bolstered methane monitoring capabilities [19,33].

2.3. Artificial Intelligence in Emission Analysis

The rise of artificial intelligence has opened new avenues for emission structure analysis in the dairy industry [34]. Recent studies have demonstrated how AI can assign varying importance to different farm characteristics, potentially identifying novel solutions for reducing methane emissions [21,22,23]. The Mooanalytica group’s work exemplifies this approach, providing actionable insights for implementing effective emission reduction strategies. Despite these advancements, challenges remain in accurately attributing methane emissions to specific sources and in developing effective mitigation strategies. The complex interplay between farm management practices, animal genetics, and environmental factors complicates efforts to identify the most impactful interventions [35].

3. Materials and Methods

3.1. Data Collection and Integration

Data were collected from 11 dairy farms (Table 1) located in Atlantic Canada, specifically in Nova Scotia, Prince Edward Island, and New Brunswick. This dataset was obtained through Lactanet, which provided comprehensive insights into various operational aspects of these farms. Although the data were not collected simultaneously across all farms, the information gathered offered valuable insights into the factors influencing methane emissions. A total of 36 factors related to dairy farm operations were identified and categorized into four main groups: basic information, production performance, reproduction performance, culling performance, and genetics information.
Basic information included the province code, Lactanet herd ID, and the date the herd began milk recording. Production performance metrics encompassed average milk yield at test date (kg/day), average milk fat yield at test date (kg/day), average milk protein yield at test date (kg/day), average somatic cell count (SCC) at test date (×1000 cells/mL), and average milk urea nitrogen at test date (mg/dL). Additionally, MTP predicted rolling 12-month herd averages for 305-day milk yield, fat yield, and protein yield were included.
Reproduction performance factors comprised rolling 12-month herd averages for gestation length (days), dry period length (days), days open (days), calving interval (days), days to first service (days), and total number of breeding services per year. Culling performance data included percentages of cows culled for voluntary reasons in the last 12 months, involuntary reasons, cows left for various reasons, mortality rates, and percentages sold or left for reproduction or health issues.
Genetics information encompassed the count of animals in the herd with a genetic evaluation, EBVs for milk yield, fat yield, protein yield, fat percentage, protein percentage, relative breeding value for SCC linear score, lifetime performance index, and Pro$ index.
To complement the farm-level data, we utilized Sentinel-5P satellite data to obtain methane column volume mixing ratio dry air bias-corrected measurements corresponding to each farm’s location during the data collection period. This integration allowed us to analyze methane emissions in relation to specific farm characteristics.

3.2. Data Collection Time Frame and Temporal Alignment

Farm-level data were systematically collected from January 2020 to December 2022, encompassing a three-year period. Monthly data acquisition was employed to capture seasonal variations influencing methane emissions, such as alterations in feed composition or herd management practices. This temporal frequency enabled the evaluation of patterns and seasonal impacts on methane production. To ensure synchronization with Sentinel-5P satellite observations, farm data collection dates were aligned with the satellite’s overpass schedules. Sentinel-5P orbits the Earth at an altitude of 824 km with a 100-min cycle, facilitating daily data capture over identical locations at a spatial resolution of 7 × 7 km2. Methane concentration measurements were extracted for each farm’s geographic coordinates on corresponding dates. When multiple satellite observations coincided with a single farm data point, methane concentration values were averaged to obtain representative measurements. Potential errors or biases in correlating ground-level farm data with satellite-derived methane concentrations were meticulously considered. Factors such as atmospheric conditions, cloud cover, and the satellite’s spatial resolution could influence data accuracy. To address these issues, satellite data acquired under unfavorable atmospheric conditions were excluded based on quality flags provided in the Sentinel-5P dataset. The farms in this study were selected for their isolated rural locations, minimizing the likelihood of multiple dairies within the 7 × 7 km satellite footprint. Additionally, farm locations were cross-referenced with land use data to identify and account for other potential methane sources within the satellite’s footprint. This involved pinpointing farm locations using geographic coordinates and obtaining surrounding land use information from governmental or environmental databases. Subsequent analysis identified any non-farm methane sources within the 7 km × 7 km footprint of the Sentinel-5P TROPOMI instrument. Land use data were used to identify and exclude footprints containing non-dairy methane sources. The TROPOMI algorithm retrieves column-averaged methane concentrations (XCH4) using bias correction and quality filters, representing the total methane content in a vertical atmospheric column. Although these measurements do not directly correspond to ground-level emissions, they offer valuable insights into the spatial and temporal variability of methane dynamics. If additional methane sources were detected within a farm’s satellite footprint, these were accounted for in the analysis to ensure precise attribution of methane emissions to the specific farm under study. This cross-referencing process was essential in mitigating potential errors or biases when matching ground-level farm data with satellite-derived methane concentrations. The methodology was particularly crucial given the expansive size of the TROPOMI pixels relative to typical dairy farm dimensions, enabling researchers to more accurately isolate methane emissions attributable to each dairy farm in this study. Farm locations were meticulously cross-referenced with land use data to identify non-dairy methane sources, such as wetlands or landfills. Any satellite data potentially influenced by these sources were excluded to preserve the integrity of the correlation analysis.

3.3. Data Processing and Analysis

Data processing was a critical step in transforming raw information into actionable insights. The farm-level and satellite data were organized and cleaned using Python scripts to prepare them for subsequent analyses. Specifically, preprocessing included the handling of missing data, outlier detection, and normalization. Given the large number of features (36 factors related to farm operations), feature engineering was performed using Variance Inflation Factor (VIF) and Principal Component Analysis (PCA) to address multicollinearity and reduce dimensionality, respectively. VIF was applied to identify and remove features with high multicollinearity. This step was crucial for improving the stability and interpretability of the subsequent models, especially when training Random Forest. Meanwhile, PCA was used as a dimensionality reduction tool to summarize the data into fewer components that still captured most of the variance, which helped in reducing the complexity of Neural Network training. These preprocessing steps provided a robust basis for machine learning analysis, enabling better performance and interpretability [36,37].
In our study, we employed two key techniques for feature selection and dimensionality reduction: Variance Inflation Factor (VIF) and Principal Component Analysis (PCA). These methods were crucial in addressing multicollinearity and reducing the complexity of our dataset while preserving essential information. We utilized VIF to identify and mitigate multicollinearity among our independent variables. The process involved calculating VIF for each independent variable using the formula V I F i = 1 1 − R i 2   , where R i 2 is the coefficient of determination for regressing the ith independent variable on the remaining ones. We interpreted VIF values as follows: VIF = 1 indicated no correlation, 1 < VIF < 10 suggested moderate correlation, and VIF > 0 signaled high correlation. For variables with VIF > 0, we carefully examined their relevance to this study, retaining control variables with high VIF if variables of interest showed low multicollinearity. Variables with VIF > 10 were considered for removal or combination with related variables.
Our PCA implementation involved standardizing the data, computing the covariance matrix, calculating eigenvectors and eigenvalues, sorting eigenvectors by decreasing eigenvalues, selecting the top k eigenvectors to form a transformation matrix, and transforming the original dataset onto the new 10-dimensional feature subspace. We chose the number of principal components (10) based on the cumulative explained variance ratio, ensuring that we retained at least 95% of the original variance in the dataset. By using VIF and PCA, we effectively addressed multicollinearity issues and reduced the dimensionality of our dataset, improving computational efficiency and enhancing the interpretability of our results by focusing on the most significant features and their linear combinations.

3.4. Emissions Profiling and Trend Analysis

We employed two machine learning models to predict methane emissions: Random Forest (RF) and Neural Network (NN). Random Forest Regressor implemented using the RandomForestRegressor class from the scikit-learn library. It consists of 100 decision trees, where each tree is independently trained on a random subset of the training dataset through bootstrap aggregation. During training, each tree considers a random subset of features for splitting at each node, which helps in capturing complex interactions and non-linear relationships within the data while reducing the risk of overfitting. The random state is set to 42 to ensure reproducibility of results across multiple runs. The NN model used in this study is a fully connected feedforward neural network, implemented using the Keras Sequential API. The architecture comprises three hidden layers with 128, 64, and 32 units, respectively, and an output layer with a single unit to predict the target variable. Each hidden layer employs the ReLU activation function, enabling the model to capture non-linear relationships in the data effectively. The input layer accepts features with a shape corresponding to the number of input variables. The output layer is linear, appropriate for regression tasks where the target is a continuous value. The model is compiled using the Adam optimizer, which provides efficient and robust optimization by combining adaptive learning rates and momentum, and the mean squared error loss function, which minimizes the squared differences between predicted and actual values. Both models were trained using the preprocessed dataset, with PCA-transformed data being used primarily for NN to optimize its training efficiency by reducing input dimensionality. RF, in contrast, was more effective when trained on the dataset after handling multicollinearity via VIF, preserving interpretability of feature contributions. The Random Forest model was chosen due to its capability to handle a large number of features and its robustness against overfitting. Additionally, RF’s ability to rank feature importance provided insights into the most significant factors affecting methane emissions. The Neural Network model was utilized to capture complex, non-linear relationships between the farm characteristics and methane emissions. By using PCA-reduced input data, the NN achieved improved training speed and efficiency. However, the trade-off with NN is its lower interpretability compared to RF, making it more challenging to explain the specific relationships it identifies.

3.5. Predictive Analytics for Emission Reduction

The evaluation of both Random Forest and Neural Network models was conducted using 5-fold cross-validation techniques to assess their predictive performance, where the dataset was divided into five equally sized subsets. During each iteration, four subsets were used for training, and the remaining subset was used for validation, ensuring that every data point was included in the validation set exactly once. Metrics such as R-squared and mean squared error (MSE) were used to quantify model accuracy and reliability. These metrics allowed us to compare the models’ capabilities in predicting methane emissions and determine which model provided the best balance of performance and interpretability. The Random Forest model was ultimately selected for further analysis based on its high accuracy and ability to provide feature importance rankings, which are crucial for understanding the influence of different farm characteristics on methane emissions. The Neural Network model provided valuable insights into potential non-linear relationships but was less interpretable, which limited its use in understanding specific contributing factors.

4. Results

4.1. Correlation Analysis of Dairy Farm Factors

Figure 2 illustrates the correlations among various dairy farm factors, with color intensity indicating the strength of these correlations. The analysis reveals significant multicollinearity primarily within genetic information, likely due to the biological relationship between milk composition and cow age. Notably, the average production metrics—Average milk yield at test date (kg/day), Average milk fat yield at test date (kg/day), and Average milk protein yield at test date (kg/day)—exhibit strong positive correlations. Similarly, the rolling herd averages for milk, fat, and protein (MTP predicted rolling 12-mo herd average for 305-day milk yield based (kg), MTP predicted rolling 12-mo herd average for 305-day milk fat yield (kg), MTP predicted rolling 12-mo herd average for 305-day milk protein yield (kg)) demonstrate high interdependence, reflecting their shared biological traits. Conversely, negative correlations emerge among Rolling 12-mo herd average for days open (days), Rolling 12-mo herd average for calving interval (days), and Rolling 12-mo herd average for days to first service (days) with multiple genetic characteristics. This intriguing biological relationship suggests that adjustments in these features could effectively influence methane emissions. Understanding how these factors interact provides valuable insights for targeted emission reduction strategies.

4.2. Correlation of Dairy Farm Factors with Methane Concentration

Figure 3 presents the correlation coefficients between individual dairy farm characteristics and methane concentration (ppb). The correlation values range between 0.25 and −0.25, indicating no direct linear relationship between most features and methane concentration. The highest correlation coefficient is observed for the percentage of cows left for other or unknown reasons in the last 12 months, followed closely by province code, percentage of cows left in the last 12 months, and rolling 12-month herd totals for breeding services—all exceeding 0.2. These findings suggest that production metrics and mortality rates have a weak positive correlation with methane concentration. The province code’s notable position in positive correlation indicates significant regional differences in methane concentrations, likely due to varying baseline emissions across provinces. In contrast, only three characteristics exhibit a negative correlation with methane concentration: Lactanet herd ID, EBV for protein percent, and rolling 12-month herd average for days to first service. This suggests that while milk quality and production metrics have weak linear relationships with methane emissions, their influence is nonetheless relevant.

4.3. Performance Comparison of Machine Learning Models

Figure 4 compares the performance of different machine learning models for predicting methane emissions. The Random Forest model applied to data processed through Variance Inflation Factor (VIF) achieved an R-squared value of 0.97 and a mean squared error (MSE) of 51,000—indicating superior performance compared to other configurations. In contrast, the Random Forest model using PCA-processed data yielded lower performance metrics due to PCA’s dimensionality reduction compressing information into fewer dimensions. The neural network model demonstrated enhanced capability in capturing potential information from PCA-processed data compared to Random Forest; however, it lacked interpretability due to its black-box nature. Ultimately, we selected the Random Forest model applied to VIF-processed data for further analysis of feature importance due to its balance of performance and interpretability.

4.4. Feature Importance in Predicting Methane Emissions

Figure 5 ranks dairy farm characteristics based on their importance in predicting methane emissions as determined by the Random Forest model. The test date emerged as a critical feature; methane concentrations have shown a gradual increase over time, potentially reflecting rising emissions from dairy operations. Key features influencing methane emissions include the percentage of cows left for other or unknown reasons in the last 12 months and the percentage of cows left in the last 12 months—all positively correlated with increased methane concentrations. The EBV for protein percent also plays a significant role; its strong negative correlation with methane concentration suggests that improving this trait could effectively mitigate emissions by influencing feed composition and manure management practices.

5. Discussion

5.1. Integration of Atmospheric Transport Models for Accurate Emission Estimates

Accurate methane emission estimates are essential for effective climate action, necessitating the integration of atmospheric transport models such as GEOS-Chem and WRF-Chem. These models simulate methane dispersion by incorporating crucial meteorological data, including wind patterns, temperature, and atmospheric stability [38]. By combining satellite observations with these models, researchers can employ inverse modeling techniques to back-calculate emissions from observed concentrations. This approach enhances spatial resolution and source attribution, allowing for a clearer distinction between agricultural, industrial, and natural methane sources [39,40]. However, developing high-resolution atmospheric models to improve methane emission estimates presents significant challenges. High-resolution modeling requires detailed meteorological inputs and precise emission inventories that may not always be readily available. Furthermore, integrating local meteorological data into atmospheric transport models is complicated by variability in weather patterns across different regions. Localized data collection efforts must be robust enough to capture these variations while ensuring compatibility with broader modeling frameworks. Addressing these challenges involves refining atmospheric models to incorporate localized meteorological data and validating these models against ground-based measurements [41]. The development of such models can significantly enhance the accuracy of methane emissions estimates and contribute to more effective mitigation strategies. Background methane concentrations (~2 ppm) and meteorological influences such as wind patterns and temperature variability were not explicitly subtracted but are recognized as potential confounders. Future studies should integrate high-resolution atmospheric transport models to address these factors

5.2. Negative Correlation Between EBV for Protein Percentage and Methane Emissions

A compelling finding of this study is the significant negative correlation between the Estimated Breeding Value (EBV) for protein percentage and methane concentrations. In the Random Forest model, EBV for protein percentage emerged as the second most influential feature affecting methane emissions, following the test date. This correlation indicates that cows with higher genetic potential for milk protein production are associated with lower methane emissions. Several biological mechanisms may explain this relationship. Cows with elevated EBVs for protein percentages likely exhibit greater efficiency in converting feed into milk protein, leading to reduced fermentation in the rumen and consequently lower methane production [42]. Genetic factors that enhance protein production could also influence rumen microbiota composition, favoring microbial communities that produce less methane as a byproduct of fermentation [43,44]. Additionally, cows bred for higher protein production may allocate more energy toward milk synthesis rather than maintenance, potentially decreasing overall methane emissions per unit of milk produced.
Dietary factors further complicate this relationship. The composition and quality of feed can significantly influence both milk protein content and methane emissions. Diets formulated to enhance milk protein levels may alter rumen fermentation patterns, thereby affecting methane production. For instance, higher-quality feeds that promote efficient digestion could lead to reduced methane emissions while simultaneously boosting protein yields. Comparative studies reinforce our findings; van Lingen et al. identified relationships between milk fatty acid profiles and methane production in dairy cattle, suggesting that milk composition serves as an indicator of methane emissions [20]. Similarly, de Haas et al. discussed genetic selection’s potential to mitigate enteric methane emissions, aligning with our results [45]. The negative correlation implies that selecting higher milk protein percentages in breeding programs could effectively reduce methane emissions from dairy farms. However, the intricate relationship between milk protein genetics and methane emissions warrants further investigation to disentangle genetic influences from management and dietary factors.

5.3. Management Factors Influencing Methane Emissions

Bedding types play a pivotal role in influencing methane emissions on dairy farms due to their effects on manure management, microbial activity, and the decomposition of organic materials. Organic bedding, such as wood shavings or straw, can decompose under anaerobic conditions, promoting microbial activity that produces methane. This process is particularly influenced by the bedding’s moisture content, carbon-to-nitrogen ratio, and temperature. In contrast, inorganic bedding like sand does not decompose and therefore does not directly contribute to methane emissions. However, sand’s low moisture retention can suppress microbial activity, indirectly reducing methane production. Bedding also impacts manure management; for example, sand facilitates the separation of manure, which can decrease emissions during storage. Additionally, bedding that retains moisture can exacerbate anaerobic digestion in manure, increasing methane emissions, whereas drier, inorganic materials help mitigate this effect. Beyond emissions, bedding choice affects cow comfort and health, which in turn influence milk production efficiency and emissions intensity per unit of milk produced. To minimize emissions, a holistic approach that considers bedding type, manure management, and cow welfare is essential.

5.4. Trade-Offs Between Management Practices

Implementing strategies to reduce methane emissions often involves trade-offs that can impact milk yield, quality, and overall farm economics. For instance, altering feed composition—such as increasing dietary fats—can effectively lower methane emissions but may compromise feed intake, nutrient digestibility, and milk composition [46,47]. Additionally, emission reduction strategies may incur increased costs due to more expensive feed additives or changes in management practices, potentially affecting farm profitability [48,49]. Focusing exclusively on mitigating methane emissions risks overlooking other environmental issues such as nitrogen excretion leading to nitrate pollution or increased carbon dioxide emissions from feed production. A holistic approach is imperative to ensure that mitigation strategies do not inadvertently exacerbate other environmental challenges [50]. Evaluating emission reduction strategies within the broader context of sustainable dairy production is essential. Life cycle assessments can quantify the net environmental benefits and economic feasibility of various practices [51,52].
Overall, while our findings underscore the potential for breeding strategies focused on enhancing milk protein content to mitigate methane emissions effectively, they also highlight the complexity of interactions among genetics, diet, management practices, and environmental factors. Future research should aim to refine these insights further by integrating high-resolution atmospheric modeling with robust local data collection efforts to develop targeted strategies that balance economic viability with environmental sustainability in dairy farming.

6. Conclusions

This study highlights the transformative potential of integrating satellite observations with farm-level data and advanced modeling techniques to revolutionize methane emission monitoring in the dairy sector. By employing a rigorous analytical framework that included correlation analysis alongside Variance Inflation Factor (VIF) and Principal Component Analysis (PCA), we effectively captured methane emission dynamics using Random Forest and Neural Network methodologies. Our findings revealed a strong positive correlation between methane concentrations and dairy cow production and mortality rates, reflecting the influence of herd turnover and nutrient utilization on emissions. Conversely, the significant negative correlation between the Estimated Breeding Value (EBV) for protein percentage and methane emissions underscores the critical role of genetics in shaping emission outcomes. This result demonstrates the potential of targeted breeding strategies to enhance protein efficiency and reduce methane output, providing an innovative pathway for mitigation. However, the intricate interactions between milk protein genetics, dietary inputs, and emission profiles warrant further investigation to disentangle these overlapping influences and refine actionable interventions.
Incorporating atmospheric transport models into the analytical framework offers the promise of significantly improving emission estimates by accounting for dispersion processes and regional variations. These models can provide greater spatial and temporal resolution, enabling precise attribution of methane emissions to specific sources. Nevertheless, applying such models at the farm-specific scale remains a considerable challenge, requiring high-resolution input data and substantial computational resources. The limited scope of this study, involving data from only 11 dairy farms, highlights the necessity for expanding datasets to achieve more comprehensive and statistically robust analyses. Future research should prioritize developing high-resolution atmospheric models that integrate local meteorological data while validating predictions against ground-based measurements to bridge the gap between observation and actionable insight.
The findings of this study emphasize the importance of balancing emission reduction strategies with economic viability and broader environmental impacts. Trade-offs between management practices, milk production, and environmental sustainability must be critically assessed to ensure that interventions are both effective and equitable. Adopting a holistic approach that integrates technical, biological, and economic dimensions will be vital for achieving meaningful progress in sustainable dairy farming.
Ultimately, this research provides valuable insights into the multifaceted relationship between dairy farming practices and methane emissions, forming a robust foundation for informed decision-making. By harnessing the power of advanced analytics and machine learning tools, stakeholders can develop targeted strategies to mitigate emissions while supporting sustainable agricultural practices. These insights not only contribute to reducing the environmental footprint of dairy production in Canada but also offer a scalable framework for addressing similar challenges globally. This work calls for interdisciplinary collaboration and sustained innovation to align dairy farming with global climate goals, ensuring a resilient and sustainable future for the industry.

Author Contributions

Conceptualization, S.N; methodology, S.N.; software, H.B.; validation, H.B.; formal analysis, H.B.; investigation, S.N., and H.B., resources, S.N.; writing—original draft preparation, H.B.; writing—review and editing, S.N.; visualization, H.B.; supervision, S.N.; project administration, S.N.; funding acquisition, S.N. All authors have read and agreed to the published version of the manuscript.

Funding

This work is kindly sponsored by the Natural Sciences and Engineering Research Council of Canada (RGPIN 2024-04450), Net Zero Atlantic Canada Agency (300700018), Mitacs Canada (IT36514), and the Department of New Brunswick Agriculture, Aquaculture and Fisheries (NB2425-0025).

Data Availability Statement

The data are available from the corresponding author upon reasonable request.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Howarth, R. Methane emissions and climatic warming risk from hydraulic fracturing and shale gas development: Implications for policy. Energy Emiss. Control. Technol. 2015, 3, 45–54. [Google Scholar] [CrossRef]
  2. Jain, P. Greenhouse effect and climate change: Scientific basis and overview. Renew. Energy 1993, 3, 403–420. [Google Scholar] [CrossRef]
  3. Nisbet, E.G.; Manning, M.R.; Dlugokencky, E.J.; Fisher, R.E.; Lowry, D.; Michel, S.E.; Myhre, C.L.; Platt, S.M.; Allen, G.; Bousquet, P.; et al. Very Strong Atmospheric Methane Growth in the 4 Years 2014–2017: Implications for the Paris Agreement. Glob. Biogeochem. Cycles 2019, 33, 318–342. [Google Scholar] [CrossRef]
  4. Zhang, S.; Ma, J.; Zhang, X.; Guo, C. Atmospheric remote sensing for anthropogenic methane emissions: Applications and research opportunities. Sci. Total. Environ. 2023, 893, 164701. [Google Scholar] [CrossRef]
  5. Bačėninaitė, D.; Džermeikaitė, K.; Antanaitis, R. Global Warming and Dairy Cattle: How to Control and Reduce Methane Emission. Animals 2022, 12, 2687. [Google Scholar] [CrossRef]
  6. Wattiaux, M.; Uddin, M.; Letelier, P.; Jackson, R.; Larson, R. Invited Review: Emission and mitigation of greenhouse gases from dairy farms: The cow, the manure, and the field. Appl. Anim. Sci. 2019, 35, 238–254. [Google Scholar] [CrossRef]
  7. Skytt, T.; Nielsen, S.N.; Jonsson, B.-G. Global warming potential and absolute global temperature change potential from carbon dioxide and methane fluxes as indicators of regional sustainability—A case study of Jämtland, Sweden. Ecol. Indic. 2020, 110. [Google Scholar] [CrossRef]
  8. Derwent, R.G. Global Warming Potential (GWP) for Methane: Monte Carlo Analysis of the Uncertainties in Global Tropospheric Model Predictions. Atmosphere 2020, 11, 486. [Google Scholar] [CrossRef]
  9. Ghahremanloo, M.; Choi, Y.; Singh, D. Deep learning bias correction of GEMS tropospheric NO2: A comparative validation of NO2 from GEMS and TROPOMI using Pandora observations. Environ. Int. 2024, 190, 108818. [Google Scholar] [CrossRef]
  10. Tongwane, M.I.; Moeletsi, M.E. Provincial cattle carbon emissions from enteric fermentation and manure management in South Africa. Environ. Res. 2021, 195, 110833. [Google Scholar] [CrossRef]
  11. Vörösmarty, G.; Dobos, I. Green purchasing frameworks considering firm size: A multicollinearity analysis using variance inflation factor. Supply Chain Forum Int. J. 2020, 21, 290–301. [Google Scholar] [CrossRef]
  12. Grzybowski, P.T.; Markowicz, K.M.; Musiał, J.P. Estimations of the Ground-Level NO2 Concentrations Based on the Sentinel-5P NO2 Tropospheric Column Number Density Product. Remote. Sens. 2023, 15, 378. [Google Scholar] [CrossRef]
  13. Veefkind, J.P.; Aben, I.; McMullan, K.; Förster, H.; de Vries, J.; Otter, G.; Claas, J.; Eskes, H.J.; de Haan, J.F.; Kleipool, Q.; et al. TROPOMI on the ESA Sentinel-5 Precursor: A GMES mission for global observations of the atmospheric composition for climate, air quality and ozone layer applications. Remote Sens. Environ. 2012, 120, 70–83. [Google Scholar] [CrossRef]
  14. Zhang, W.-F.; Dou, Z.-X.; He, P.; Ju, X.-T.; Powlson, D.; Chadwick, D.; Norse, D.; Lu, Y.-L.; Zhang, Y.; Wu, L.; et al. New technologies reduce greenhouse gas emissions from nitrogenous fertilizer in China. Proc. Natl. Acad. Sci. USA 2013, 110, 8375–8380. [Google Scholar] [CrossRef]
  15. Zou, M.; Xiong, X.; Wu, Z.; Li, S.; Zhang, Y.; Chen, L. Increase of Atmospheric Methane Observed from Space-Borne and Ground-Based Measurements. Remote. Sens. 2019, 11, 964. [Google Scholar] [CrossRef]
  16. Schuit, B.J.; Maasakkers, J.D.; Bijl, P.; Mahapatra, G.; Berg, A.-W.v.D.; Pandey, S.; Lorente, A.; Borsdorff, T.; Houweling, S.; Varon, D.J.; et al. Automated detection and monitoring of methane super-emitters using satellite data. Atmos. Meas. Tech. 2023, 23, 9071–9098. [Google Scholar] [CrossRef]
  17. Joyce, P.; Villena, C.R.; Huang, Y.; Webb, A.; Gloor, M.; Wagner, F.H.; Chipperfield, M.P.; Guilló, R.B.; Wilson, C.; Boesch, H. Using a deep neural network to detect methane point sources and quantify emissions from PRISMA hyperspectral satellite images. Atmos. Meas. Tech. 2023, 16, 2627–2640. [Google Scholar] [CrossRef]
  18. Janardanan, R.; Maksyutov, S.; Tsuruta, A.; Wang, F.; Tiwari, Y.K.; Valsala, V.; Ito, A.; Yoshida, Y.; Kaiser, J.W.; Janssens-Maenhout, G.; et al. Country-Scale Analysis of Methane Emissions with a High-Resolution Inverse Model Using GOSAT and Surface Observations. Remote. Sens. 2020, 12, 375. [Google Scholar] [CrossRef]
  19. Fan, L.; Wan, Y.; Dai, Y. Development of a Multi-Source Satellite Fusion Method for XCH4 Product Generation in Oil and Gas Production Areas. Appl. Sci. 2024, 14, 11100. [Google Scholar] [CrossRef]
  20. Vanselow, S.; Schneising, O.; Buchwitz, M.; Reuter, M.; Bovensmann, H.; Boesch, H.; Burrows, J.P. Automated detection of regions with persistently enhanced methane concentrations using Sentinel-5 Precursor satellite data. Atmos. Meas. Tech. 2024, 24, 10441–10473. [Google Scholar] [CrossRef]
  21. Pramod, A.; Naicker, H.S.; Tyagi, A.K. Machine learning and deep learning: Open issues and future research directions for the next 10 years. In Computational Analysis and Deep Learning for Medical Care: Principles, Methods, and Applications, 18th ed.; Tyagi, A.K., Ed.; John Wiley & Sons: Hoboken, NJ, USA, 2021. [Google Scholar] [CrossRef]
  22. Guan, K.; Jin, Z.; Peng, B.; Tang, J.; DeLucia, E.H.; West, P.C.; Jiang, C.; Wang, S.; Kim, T.; Zhou, W.; et al. A scalable framework for quantifying field-level agricultural carbon outcomes. Earth-Science Rev. 2023, 243. [Google Scholar] [CrossRef]
  23. Liu, L.; Zhou, W.; Guan, K.; Peng, B.; Xu, S.; Tang, J.; Zhu, Q.; Till, J.; Jia, X.; Jiang, C.; et al. Knowledge-guided machine learning can improve carbon cycle quantification in agroecosystems. Nat. Commun. 2024, 15, 1–15. [Google Scholar] [CrossRef] [PubMed]
  24. Hu, H.; Landgraf, J.; Detmers, R.; Borsdorff, T.; de Brugh, J.A.; Aben, I.; Butz, A.; Hasekamp, O. Toward Global Mapping of Methane With TROPOMI: First Results and Intersatellite Comparison to GOSAT. Geophys. Res. Lett. 2018, 45, 3682–3689. [Google Scholar] [CrossRef]
  25. Jacob, D.J.; Turner, A.J.; Maasakkers, J.D.; Sheng, J.; Sun, K.; Liu, X.; Chance, K.; Aben, I.; McKeever, J.; Frankenberg, C. Satellite observations of atmospheric methane and their value for quantifying methane emissions. Atmos. Meas. Tech. 2016, 16, 14371–14396. [Google Scholar] [CrossRef]
  26. Loew, A.; Bell, W.; Brocca, L.; Bulgin, C.E.; Burdanowitz, J.; Calbet, X.; Donner, R.V.; Ghent, D.; Gruber, A.; Kaminski, T.; et al. Validation practices for satellite-based Earth observation data across communities. Rev. Geophys. 2017, 55, 779–817. [Google Scholar] [CrossRef]
  27. Saarela, M.; Jauhiainen, S. Comparison of feature importance measures as explanations for classification models. SN Appl. Sci. 2021, 3, 1–12. [Google Scholar] [CrossRef]
  28. Kherif, F.; Latypova, A. Principal component analysis. In Machine Learning; Elsevier: Amsterdam, The Netherlands, 2020; pp. 209–225. [Google Scholar] [CrossRef]
  29. Shetty, S.; Schneider, P.; Stebel, K.; Hamer, P.D.; Kylling, A.; Berntsen, T.K. Estimating surface NO2 concentrations over Europe using Sentinel-5P TROPOMI observations and Machine Learning. Remote. Sens. Environ. 2024, 312. [Google Scholar] [CrossRef]
  30. Bi, H.; Neethirajan, S. Satellite Data and Machine Learning for Benchmarking Methane Concentrations in the Canadian Dairy Industry. Sustainability 2024, 16, 10400. [Google Scholar] [CrossRef]
  31. Barré, J.; Aben, I.; Agustí-Panareda, A.; Balsamo, G.; Bousserez, N.; Dueben, P.; Engelen, R.; Inness, A.; Lorente, A.; McNorton, J.; et al. Systematic detection of local CH4 anomalies by combining satellite measurements with high-resolution forecasts. Atmos. Meas. Tech. 2021, 21, 5117–5136. [Google Scholar] [CrossRef]
  32. Parmar, M.; Vaishya, A.; Ojha, N.; Pandya, M.R.; Girach, I.A. Addressing Observational Gaps in Aerosol Parameters using Machine Learning: Implications to Aerosol Radiative Forcing. Atmos. Environ. 2024, 333, 120661. [Google Scholar] [CrossRef]
  33. Balasus, N.; Jacob, D.J.; Lorente, A.; Maasakkers, J.D.; Parker, R.J.; Boesch, H.; Chen, Z.; Kelp, M.M.; Nesser, H.; Varon, D.J. A blended TROPOMI+GOSAT satellite data product for atmospheric methane using machine learning to correct retrieval biases. Atmos. Meas. Tech. 2023, 16, 3787–3807. [Google Scholar] [CrossRef]
  34. Kort, E.A.; Frankenberg, C.; Costigan, K.R.; Lindenmaier, R.; Dubey, M.K.; Wunch, D. Four corners: The largest US methane anomaly viewed from space. Geophys. Res. Lett. 2014, 41, 6898–6903. [Google Scholar] [CrossRef]
  35. A Rolph, K.; E Forest, C.; Ruark, M.D. The role of non-CO2 mitigation options within the dairy industry for pursuing climate change targets. Environ. Res. Lett. 2019, 14, 084039. [Google Scholar] [CrossRef]
  36. Cheng, L.; De Vos, J.; Zhao, P.; Yang, M.; Witlox, F. Examining non-linear built environment effects on elderly’s walking: A random forest approach. Transp. Res. Part D Transp. Environ. 2020, 88, 102552. [Google Scholar] [CrossRef]
  37. Cui, F.; Kim, M.; Park, C.; Kim, D.; Mo, K.; Kim, M. Application of principal component analysis (PCA) to the assessment of parameter correlations in the partial-nitrification process using aerobic granular sludge. J. Environ. Manag. 2021, 288, 112408. [Google Scholar] [CrossRef]
  38. Feng, X.; Lin, H.; Fu, T.-M.; Sulprizio, M.P.; Zhuang, J.; Jacob, D.J.; Tian, H.; Ma, Y.; Zhang, L.; Wang, X.; et al. WRF-GC (v2.0): Online two-way coupling of WRF (v3.9.1.1) and GEOS-Chem (v12.7.2) for modeling regional atmospheric chemistry–meteorology interactions. Geosci. Model Dev. 2021, 14, 3741–3768. [Google Scholar] [CrossRef]
  39. Maazallahi, H.; Fernandez, J.M.; Menoud, M.; Zavala-Araiza, D.; Weller, Z.D.; Schwietzke, S.; von Fischer, J.C.; van der Gon, H.D.; Röckmann, T. Methane mapping, emission quantification, and attribution in two European cities: Utrecht (NL) and Hamburg (DE). Atmos. Meas. Tech. 2020, 20, 14717–14740. [Google Scholar] [CrossRef]
  40. Santos, R.; Costa, A.A.; Silvestre, J.D.; Pyl, L. Development of a BIM-based Environmental and Economic Life Cycle Assessment tool. J. Clean. Prod. 2020, 265, 121705. [Google Scholar] [CrossRef]
  41. Viggiano, M.; Busetto, L.; Cimini, D.; Di Paola, F.; Geraldi, E.; Ranghetti, L.; Ricciardelli, E.; Romano, F. A new spatial modeling and interpolation approach for high-resolution temperature maps combining reanalysis data and ground measurements. Agric. For. Meteorol. 2019, 276–277, 107590. [Google Scholar] [CrossRef]
  42. Beauchemin, K.A.; Ungerfeld, E.M.; Eckard, R.J.; Wang, M. Review: Fifty years of research on rumen methanogenesis: Lessons learned and future challenges for mitigation. Animal 2020, 14, s2–s16. [Google Scholar] [CrossRef]
  43. Firkins, J.; Mitchell, K. Invited review: Rumen modifiers in today’s dairy rations. J. Dairy Sci. 2023, 106, 3053–3071. [Google Scholar] [CrossRef]
  44. Pitta, D.; Indugu, N.; Narayan, K.; Hennessy, M. Symposium review: Understanding the role of the rumen microbiome in enteric methane mitigation and productivity in dairy cows. J. Dairy Sci. 2022, 105, 8569–8585. [Google Scholar] [CrossRef] [PubMed]
  45. de Haas, Y.; Veerkamp, R.; de Jong, G.; Aldridge, M. Selective breeding as a mitigation tool for methane emissions from dairy cattle. Animal 2021, 15, 100294. [Google Scholar] [CrossRef] [PubMed]
  46. Arndt, C.; Hristov, A.N.; Price, W.J.; McClelland, S.C.; Pelaez, A.M.; Cueva, S.F.; Oh, J.; Dijkstra, J.; Bannink, A.; Bayat, A.R.; et al. Full adoption of the most effective strategies to mitigate methane emissions by ruminants can help meet the 1.5 °C target by 2030 but not 2050. Proc. Natl. Acad. Sci. USA 2022, 119, e2111294119. [Google Scholar] [CrossRef] [PubMed]
  47. Min, B.-R.; Lee, S.; Jung, H.; Miller, D.N.; Chen, R. Enteric Methane Emissions and Animal Performance in Dairy and Beef Cattle Production: Strategies, Opportunities, and Impact of Reducing Emissions. Animals 2022, 12, 948. [Google Scholar] [CrossRef]
  48. Bilotto, F.; Recavarren, P.; Vibart, R.; Machado, C.F. Backgrounding strategy effects on farm productivity, profitability and greenhouse gas emissions of cow-calf systems in the Flooding Pampas of Argentina. Agric. Syst. 2019, 176. [Google Scholar] [CrossRef]
  49. Giamouri, E.; Zisis, F.; Mitsiopoulou, C.; Christodoulou, C.; Pappas, A.C.; Simitzis, P.E.; Kamilaris, C.; Galliou, F.; Manios, T.; Mavrommatis, A.; et al. Sustainable Strategies for Greenhouse Gas Emission Reduction in Small Ruminants Farming. Sustainability 2023, 15, 4118. [Google Scholar] [CrossRef]
  50. Mishra, S.; Anderson, K.; Miller, B.; Boyer, K.; Warren, A. Microgrid resilience: A holistic approach for assessing threats, identifying vulnerabilities, and designing corresponding mitigation strategies. Appl. Energy 2020, 264, 114726. [Google Scholar] [CrossRef]
  51. Hamedani, S.R.; Kuppens, T.; Malina, R.; Bocci, E.; Colantoni, A.; Villarini, M. Life Cycle Assessment and Environmental Valuation of Biochar Production: Two Case Studies in Belgium. Energies 2019, 12, 2166. [Google Scholar] [CrossRef]
  52. Yu, X.; Millet, D.B.; Henze, D.K.; Turner, A.J.; Delgado, A.L.; Bloom, A.A.; Sheng, J. A high-resolution satellite-based map of global methane emissions reveals missing wetland, fossil fuel, and monsoon sources. Atmos. Meas. Tech. 2023, 23, 3325–3346. [Google Scholar] [CrossRef]
Climate 12 00223 g001
Figure 1. Schematic representation of the data analysis process, including data collection, feature analysis, and machine learning modeling to assess the relationship between dairy farm characteristics and methane emissions.
Figure 1. Schematic representation of the data analysis process, including data collection, feature analysis, and machine learning modeling to assess the relationship between dairy farm characteristics and methane emissions.
Climate 12 00223 g001
Climate 12 00223 g002
Figure 2. Visual representation of the correlations between various dairy farm factors. The color intensity indicates the strength of correlation, with darker colors representing stronger correlations.
Figure 2. Visual representation of the correlations between various dairy farm factors. The color intensity indicates the strength of correlation, with darker colors representing stronger correlations.
Climate 12 00223 g002
Climate 12 00223 g003
Figure 3. Correlation of dairy farm factors with methane concentration. Bar chart showing the correlation coefficients between individual dairy farm characteristics and methane concentration (ppb). Positive values indicate positive correlations, while negative values indicate inverse relationships.
Figure 3. Correlation of dairy farm factors with methane concentration. Bar chart showing the correlation coefficients between individual dairy farm characteristics and methane concentration (ppb). Positive values indicate positive correlations, while negative values indicate inverse relationships.
Climate 12 00223 g003
Climate 12 00223 g004
Figure 4. Performance comparison of machine learning models for methane prediction. Comparison of R-squared values and mean squared errors (MSEs) for Random Forest and Neural Network models applied to datasets processed with Variance Inflation Factor (VIF) and Principal Component Analysis (PCA).
Figure 4. Performance comparison of machine learning models for methane prediction. Comparison of R-squared values and mean squared errors (MSEs) for Random Forest and Neural Network models applied to datasets processed with Variance Inflation Factor (VIF) and Principal Component Analysis (PCA).
Climate 12 00223 g004
Climate 12 00223 g005
Figure 5. Feature importance in predicting methane emissions from dairy farms. Ranking of dairy farm characteristics based on their importance in predicting methane emissions, as determined by the Random Forest model. Higher values indicate greater importance in the prediction model.
Figure 5. Feature importance in predicting methane emissions from dairy farms. Ranking of dairy farm characteristics based on their importance in predicting methane emissions, as determined by the Random Forest model. Higher values indicate greater importance in the prediction model.
Climate 12 00223 g005
Table 1. Key characteristics of dairy farms analyzed in Eastern Canada: insights into farm size, bedding, and milking practices.
Table 1. Key characteristics of dairy farms analyzed in Eastern Canada: insights into farm size, bedding, and milking practices.
Farm IDProvinceNumber of CowsBedding TypeMilking Method
Farm 1New Brunswick115Free stall barn,
Wood shavings 		Herringbone milking system
Farm 2New Brunswick250Free stall barn,
Bedded on sand		Double 12 milking parlour
Farm 3New Brunswick60Wood shaving Voluntary Automatic Milking System (DeLaval VMS)
Farm 4New Brunswick70Wood ShavingHerringbone milking system
Farm 5New Brunswick385Wood ShavingRotary Robotic Parlour
Farm 6New Brunswick130SandDouble 12 milking parlour
Farm 7New Brunswick200SandRotary Robotic Parlour
Farm 8Nova Scotia100Wood Shaving Voluntary Automatic Milking System
Farm 9Nova Scotia65SandHerringbone milking system
Farm 10Nova Scotia85SandVoluntary Automatic Milking System
Farm 11Prince Edward Island200Wood ShavingRobotic Rotary Parlour
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

Bi, H.; Neethirajan, S. Mapping Methane—The Impact of Dairy Farm Practices on Emissions Through Satellite Data and Machine Learning. Climate 2024, 12, 223. https://doi.org/10.3390/cli12120223

AMA Style

Bi H, Neethirajan S. Mapping Methane—The Impact of Dairy Farm Practices on Emissions Through Satellite Data and Machine Learning. Climate. 2024; 12(12):223. https://doi.org/10.3390/cli12120223

Chicago/Turabian Style

Bi, Hanqing, and Suresh Neethirajan. 2024. "Mapping Methane—The Impact of Dairy Farm Practices on Emissions Through Satellite Data and Machine Learning" Climate 12, no. 12: 223. https://doi.org/10.3390/cli12120223

APA Style

Bi, H., & Neethirajan, S. (2024). Mapping Methane—The Impact of Dairy Farm Practices on Emissions Through Satellite Data and Machine Learning. Climate, 12(12), 223. https://doi.org/10.3390/cli12120223

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