High-Resolution Estimation of Cropland N₂O Emissions in China Based on Machine Learning Algorithms

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper aims to develop an inventory of agriculture-derived N₂O emissions in China using various statistical datasets. Several machine learning models are compared, and the results indicate that the Random Forest model provides the highest accuracy. The study also analyzes the distribution of emission factors (EFs) estimated using the developed model.

The paper appears to align well with the scope of the journal. It is generally well-structured, and the methodology does not exhibit major flaws. However, as outlined below, there are some aspects of the methodology that are not described with sufficient clarity, which may hinder readers’ understanding. Addressing these points will improve the transparency and reproducibility of the work.

For these reasons, I recommend that the paper be accepted after minor revisions.

 

Major Comments:

Lines 108–113: The dataset for “other crops” appears to be relatively small. Is this dataset sufficient for robust training and validation of the model? Additionally, it would be helpful to provide information on the coverage of the dataset for variables beyond crop type (e.g., weather conditions, soil characteristics, farming practices).

 

Lines 142–145: The term “city-level scale” is somewhat ambiguous. Please clarify the spatial resolution. For instance, is this based on administrative district divisions or another form of geographic delineation?

 

Line 146: Is using annual emission estimates sufficient for the intended analysis? Given that N₂O emissions may vary seasonally with changes in fertilizer application timing and weather patterns, please justify the choice of temporal resolution.

 

Lines 275–284: Please elaborate on the reasons for the observed changes in the distribution of N₂O emissions over the different time periods mentioned. What factors or processes might account for these temporal variations?

 

Minor Comments:

Line 34 and elsewhere: Literature references are not bracketed in accordance with the journal’s style. Please review and correct all instances.

Author Response

Reply to Reviewer 1

This paper aims to develop an inventory of agriculture-derived N₂O emissions in China using various statistical datasets. Several machine learning models are compared, and the results indicate that the Random Forest model provides the highest accuracy. The study also analyzes the distribution of emission factors (EFs) estimated using the developed model.

The paper appears to align well with the scope of the journal. It is generally well-structured, and the methodology does not exhibit major flaws. However, as outlined below, there are some aspects of the methodology that are not described with sufficient clarity, which may hinder readers’ understanding. Addressing these points will improve the transparency and reproducibility of the work.

For these reasons, I recommend that the paper be accepted after minor revisions.

[Response]: We thank the reviewer’s recommendation. Below we provide a point-by-point response to the reviewer’s comments, together with proposed changes in the revised manuscript (in blue).

1. Lines 108-113: The dataset for “other crops” appears to be relatively small. Is this dataset sufficient for robust training and validation of the model? Additionally, it would be helpful to provide information on the coverage of the dataset for variables beyond crop type (e.g., weather conditions, soil characteristics, farming practices).

[Response]: Thanks for pointing out this. The 191 vegetable samples in the original text fall under the category of “other crops”. The “other crops” are defined as cash crops excluding the three major food crops (maize, wheat, and rice), and their specific types are detailed in the main text. Therefore, our dataset includes a total of 217 samples of “other crops” (191 vegetables and 26 other cash crops). We have revised the dataset of “other crops” in the revised Method Section 2.1.

In addition to crop type, the dataset also includes information on meteorological conditions, soil characteristics, and field management practices. Specifically, tillage methods were categorized as conventional tillage (1978 sets) and no-tillage (43 sets); fertilization practices include deep application (895 records) and surface application (1126 sets); fertilizer types consist of urea (834 sets), compound fertilizers (506 sets), and other fertilizers (681 sets). For meteorological variables, temperature and precipitation data were collected for each site. Soil variables, including total nitrogen, organic carbon, pH, bulk density, and clay contents were primarily obtained from original observations. All the above variables cover major agricultural ecological regions in China and can well support the model’s training needs under different geographical environments and field management methods.

2. Lines 142-145: The term “city-level scale” is somewhat ambiguous. Please clarify the spatial resolution. For instance, is this based on administrative district divisions or another form of geographic delineation?

[Response]: Thank you for pointing it out. The “city-level” is defined in accordance with China’s administrative division standards. We have clarified this definition in the Methods of the revised manuscript.

3. Line 146: Is using annual emission estimates sufficient for the intended analysis? Given that N₂O emissions may vary seasonally with changes in fertilizer application timing and weather patterns, please justify the choice of temporal resolution.

[Response]: Thank you for your comment. We acknowledge that N₂O emissions are affected by factors such as the timing of fertilizer application and meteorological conditions; however, the total amount of fertilizer applied is the primary determinant of N₂O emissions. The reason why seasonal variations are not considered in our analysis is as follows: Government-reported data on fertilizer use are only available at an annual frequency. Additionally, since this study aims to assess the long-term trends of N₂O emissions with a primary focus on interannual variations, our analysis is based on annual data.

4. Lines 275-284: Please elaborate on the reasons for the observed changes in the distribution of N₂O emissions over the different time periods mentioned. What factors or processes might account for these temporal variations?

[Response]: The changes in the distribution of N₂O emissions from China’s cropland exhibit distinct two-stage characteristics, with the core driving factors being the synergistic effects of expanding population demand and policy regulation. The specific mechanisms are as follows: From 2000 to 2015, the population increased from 1.26 billion to 1.38 billion, leading to a significant rise in food demand [10]. To ensure food supply, the government encouraged the expansion of agricultural production through policies such as raising grain purchase prices, reducing agricultural input costs, and strengthening agricultural infrastructure [2,3]. Driven by both population growth and policy guidance, the scale of agricultural production continued to expand: the crop planting area increased from 1.56 × 10⁸ha in 2000 to 1.67 × 10⁸ ha in 2015. Spatially, high-emission hotspots were concentrated in key grain-producing regions in eastern China, such as the Huang-Huai-Hai Plain and the Middle-Lower Yangtze Plain, due to their intensive agricultural systems and high fertilizer application rates. Around 2015, the government launched the “Zero Growth Action for Chemical Fertilizer Use”, which precisely controlled nitrogen fertilizer input by promoting technologies such as soil testing and formulated fertilization, and organic fertilizer substitution [4]. Data from this study show that total fertilizer use decreased by 19%, and average annual N₂O emissions fell by 5%, with a noticeable turning point in the emission trend before the implementation of the policy. Regionally, the reduction was most significant in the eastern region [5], where N₂O emissions dropped from 130 Gg in 2016 to 87.6 Gg in 2022. In contrast, the west saw a smaller reduction, from 134 Gg to 97.8 Gg over the same period, largely due to differences in policy enforcement, the pace of agricultural modernization, and natural constraints. Therefore, the observed changes in N₂O emissions align with the patterns of agricultural development in China, shaped by population food demand, economic growth, and policy measures [6-8].”.

5. Line 34 and elsewhere: Literature references are not bracketed in accordance with the journal’s style. Please review and correct all instances.

[Response]: Thanks for pointing out this. References style has been revised accordingly.

Reviewer 2 Report

Comments and Suggestions for Authors

As croplands are a significant source of N₂O emissions, focusing this study on China, where cropland occupies a large share, is a valuable approach. The attempt to compile field-based emission factor data and derive dynamic EFs based on influencing factors is highly interesting. Nevertheless, I find that several issues need to be addressed and revised.

 

General Comments

• The reference format in MDPI journals requires numerical citations in square brackets at the end of sentences. This should be corrected throughout the manuscript.
• Since this study focuses on N₂O emissions from croplands in China and is submitted to an international journal, a comparison with cases from other countries would strengthen the work.
• The objective of estimating dynamic EFs is meaningful; however, beyond the choice of machine learning model, data processing, and influencing factors presented in the Results, it would be helpful if the EF estimates derived from the model were described in more detail in the main text, rather than being confined to the Supplementary Materials.
• The 2000–2022 publications used as references for building the ML model are based on on-site measurements, but the experimental conditions were not identical. This variability should be explicitly considered in the model construction. It is also important to clarify whether the influencing factors were selected only from the parameters reported in the referenced studies. For instance, under “fertilizer conditions,” fertilizer type and nitrogen application rate are included, but if certain factors such as the mode of fertilizer application are not reported in the literature, this limitation should be acknowledged.

 

 

Section 2. Methods

1. In Section 2.1, the sum of the sample numbers given in parentheses for maize, wheat, rice, vegetables, and other crops does not match the 2022 samples mentioned in the preceding sentence. This discrepancy should be checked.
2. Since there is no Section 2.2.2, it seems unnecessary to label a subsection as 2.2.1. It may be better to renumber 2.2.1 as 2.3.
3. Explanations of abbreviations used in the Supplementary Materials are needed (e.g., the y-axis in Figure S3a).

Section 3. Results

1. In Section 3.1, when evaluating natural and anthropogenic factors, please specify which variables are included. Currently, only the results are described without sufficient explanation. While Supplementary Figure 3 shows the importance of natural factors through the ML model, there is no explanation of how anthropogenic factors (e.g., tillage practices and nitrogen application rates) were identified as key contributors. Clarification of the analytical approach is needed.
2. Line 180: It is unclear whether the statement on N₂O EF trends from 2000 to 2022 is based on the analysis of the 327 publications mentioned in Section 2.1. Since “2000–2022” is mentioned only once in the Introduction as part of the study’s aim, the context feels ambiguous and should be clarified.
3. In Section 3.2, the y-axis in Figure 1a may need adjustment. As it stands, the three fertilizer types all appear at approximately the 1.5% level, making differences difficult to discern.
4. Section 3.2 relies entirely on the three graphs in Figure 1. However, Figures 1a and 1b are shown above the text, while Figure 1c is placed below, which appears inconsistent. It would be clearer to place all figures after the explanatory text, at the end of Section 3.2.
5. Line 202: Regarding Figure S5, while it is helpful to include the graph in the Supplementary Materials, adding a simple table with descriptive statistics (mean, minimum, maximum) of the coefficients would improve reader comprehension.
6. Line 212: The claim of an inverted U-shaped trend is not sufficiently supported by the bar chart alone. Adding a trendline would strengthen the argument. Alternatively, adjusting the y-axis (e.g., truncating the 0–200 Gg range) in Figure 2a could make the inverted U-shape more visually apparent.
7. In Table 1 (Section 3.3), the year “2020” seems to have been mistakenly entered as “2015.” Please verify and correct this.

 

 

Author Response

Reply to Reviewer 2

-As croplands are a significant source of N₂O emissions, focusing this study on China, where cropland occupies a large share, is a valuable approach. The attempt to compile field-based emission factor data and derive dynamic EFs based on influencing factors is highly interesting. Nevertheless, I find that several issues need to be addressed and revised.

[Response]: Thanks for the recognition of our research. The authors appreciate the reviewer for the valuable comments and suggestions that greatly help us improve our work. Please see our response to each comment below (in blue).

General Comments

1. The reference format in MDPI journals requires numerical citations in square brackets at the end of sentences. This should be corrected throughout the manuscript.

[Response]: Thanks for pointing out this. All literature citations in the manuscript have been revised.

2. Since this study focuses on N₂O emissions from croplands in China and is submitted to an international journal, a comparison with cases from other countries would strengthen the work.

[Response]:

3. The objective of estimating dynamic EFs is meaningful; however, beyond the choice of machine learning model, data processing, and influencing factors presented in the Results, it would be helpful if the EF estimates derived from the model were described in more detail in the main text, rather than being confined to the Supplementary Materials.

[Response]: Thank you for your valuable suggestion. We have added detailed content of the model-estimated emission factors in the main text:

4. The 2000-2022 publications used as references for building the ML model are based on on-site measurements, but the experimental conditions were not identical. This variability should be explicitly considered in the model construction. It is also important to clarify whether the influencing factors were selected only from the parameters reported in the referenced studies. For instance, under “fertilizer conditions”, fertilizer type and nitrogen application rate are included, but if certain factors such as the mode of fertilizer application are not reported in the literature, this limitation should be acknowledged.

[Response]: Thanks for your question. We have added a screening criterion in Section 2.1 of the Methods: “Include complete experimental data covering fertilizer type, nitrogen application rate, mode of fertilizer application, and tillage practice.” Additionally, in terms of selecting influencing factors, these variables were chosen because they played a primary role in shaping N₂O emission patterns [8].

Section 2. Methods

1. In Section 2.1, the sum of the sample numbers given in parentheses for maize, wheat, rice, vegetables, and other crops does not match the 2022 samples mentioned in the preceding sentence. This discrepancy should be checked.

[Response]: Thanks for pointing out this. The discrepancy in sample numbers has been checked and corrected. We have corrected the sum of the sample numbers in the article.

2. Since there is no Section 2.2.2, it seems unnecessary to label a subsection as 2.2.1. It may be better to renumber 2.2.1 as 2.3.

[Response]: Thank you for your valuable suggestion. Subsection 2.2.1 has been renumbered as Section 2.3.

3. Explanations of abbreviations used in the Supplementary Materials are needed (e.g., the y-axis in Figure S3a).

[Response]: Thanks for reminder. We have corrected all abbreviations in the Supplementary Materials in the revision.

Section 3. Results

1. In Section 3.1, when evaluating natural and anthropogenic factors, please specify which variables are included. Currently, only the results are described without sufficient explanation. While Supplementary Figure 3 shows the importance of natural factors through the ML model, there is no explanation of how anthropogenic factors (e.g., tillage practices and nitrogen application rates) were identified as key contributors. Clarification of the analytical approach is needed.

[Response]: Thank you for your detailed suggestion. Anthropogenic factors include variables such as tillage practices (such as no-tillage, tillage), nitrogen application rate, and fertilization methods (deep application, shallow application). Regarding how anthropogenic factors were identified as key influencing factors, we have added the following content to the revised Supplementary Information (Text S5): “During the model input phase, anthropogenic factor variables were first standardized: tillage practices (STP) and fertilization methods (NP) were encoded using 0-1 coding (no-tillage NT=1; conventional tillage CT=0; shallow application SBC=1; deep placement DPM=0), while nitrogen application were input directly as original values (e.g., 100 kg ha^-1). The identification of key factors utilized the same machine learning model as in Figure S3. The contribution of each variable was quantified by calculating its feature importance on the model’s prediction accuracy.” Among these, tillage practices accounted for 13% of the weight, fertilization methods 11%, and nitrogen application rates 9%. Given their significantly higher weights, these three factors were identified as key anthropogenic factors.

2. Line 180: It is unclear whether the statement on N₂O EF trends from 2000 to 2022 is based on the analysis of the 327 publications mentioned in Section 2.1. Since “2000-2022” is mentioned only once in the Introduction as part of the study’s aim, the context feels ambiguous and should be clarified.

[Response]: Thank you for your careful attention. The statement on the trends of N₂O emission factors from 2000 to 2022 in Line 180 was based on the analysis of the 327 publications mentioned in Section 2.1. To clarify this, we have added the following content to the revised Supplementary Information (Text S6): “We extracted experimental data from 327 literatures to construct the database, while not distinguished interannual differences in EFs. To analyze the interannual variation trend of EFs, we input meteorological and soil parameters of each city from 2000 to 2022 (each year) into the model, which could predict the EF value corresponding to each city based on the specific environmental and meteorological background of each year. We captured the interannual variations in EFs through the interannual differences of the input predictive features, ultimately enabling us to obtain EF data corresponding to each year and each city during the period from 2000 to 2022.”.

3. In Section 3.2, the y-axis in Figure 1a may need adjustment. As it stands, the three fertilizer types all appear at approximately the 1.5% level, making differences difficult to discern.

[Response]: Thanks for the valuable suggestion. We have revised Figure 1a to better visualize the differences between fertilizer types.

Figure 1 Temporal variation of N₂O EFs in Chinese croplands from 2000 to 2022. (a) Temporal trends in N₂O EFs for different fertilizers: (urea, compound, and others). (b) Annual variations in N₂O EFs among different crop types: maize, wheat, rice, and others. (c) Average N₂O EFs across the different agricultural zones.

4. Section 3.2 relies entirely on the three graphs in Figure 1. However, Figures 1a and 1b are shown above the text, while Figure 1c is placed below, which appears inconsistent. It would be clearer to place all figures after the explanatory text, at the end of Section 3.2.

[Response]: Thanks for the careful suggestion. We have adjusted Figure 1.

5. Line 202: Regarding Figure S5, while it is helpful to include the graph in the Supplementary Materials, adding a simple table with descriptive statistics (mean, minimum, maximum) of the coefficients would improve reader comprehension.

[Response]: Thanks for the valuable suggestion. We have added a concise Table in the supplementary materials. The newly added table is presented as follows:
Table S7 Descriptive statistics (mean, minimum, and maximum) of N₂O emission factors for each crop type

Crop Type	Mean	Min	Max
Maize	1.71	1.01	2.25
Others	1.51	0.93	1.94
Rice	0.99	0.69	1.24
Wheat	1.44	0.86	1.83

6. Line 212: The claim of an inverted U-shaped trend is not sufficiently supported by the bar chart alone. Adding a trendline would strengthen the argument. Alternatively, adjusting the y-axis (e.g., truncating the 0-200 Gg range) in Figure 2a could make the inverted U-shape more visually apparent.

[Response]: Thanks for the valuable suggestion. It should be noted that due to the adjustment of the figure sequence, the original “Figure 2” has now been renumbered as “Figure 3”. We have added a nonlinear fitting trendline to Figure 3a to explicitly illustrate the inverted U-shaped trend.

Figure 3 Temporal variation in cropland N₂O emissions in China from 2000 to 2022. (a) Inter- annual trends in cropland N₂O emissions. (b) Temporal dynamics of N₂O emissions from different crops (rice, maize, wheat, and others). (c) N₂O emissions are attributed to different fertilizer types (urea, compound, and others). (d) Contributions of cropland N₂O emissions across the nine main agricultural regions.

7. In Table 1 (Section 3.3), the year “2020” seems to have been mistakenly entered as “2015”. Please verify and correct this.

[Response]: Thanks for pointing out this. The “2015” has been corrected to “2020”.

 

 

References

1. National Bureau of Statistics of China. National Statistical Database 1999-2022. Available online: http://data.stats.gov.cn/easyquery.htm?cn=C01 (accessed on 20 February 2025).
2. Ash, R.F. The Agricultural Sector in China: Performance and Policy Dilemmas during the 1990s. China Q. 1992, 131, 545–576. https://doi.org/10.1017/S0305741000046294
3. Wu, Y.; Wang, C.; Ji, R.; et al. Metricizing Policy Texts: Comprehensive Dataset on China’s Agri-Policy Intensity Spanning 1982–2023. Sci. Data 2024, 11, 528. https://doi.org/10.1038/s41597-024-03367-0
4. Ministry of Agriculture and Rural Affairs of the People’s Republic of China. Two Actions that Seek to Achieve Zero Growth in the Use of Chemical Fertilizer and Pesticides by 2020 (in Chinese). Available online: http://www.moa.gov.cn/nybgb/2015/san/201711/t20171129_5923401.htm (accessed on 20 February 2025).
5. Shang, Z.; Zhou, F.; Smith, P.; Saikawa, E.; Ciais, P.; Chang, J.; Tian, H.; Del Grosso, S.J.; Ito, A.; Chen, M. Weakened Growth of Cropland N₂O Emissions in China Associated with Nationwide Policy Interventions. Glob. Chang. Biol. 2019, 25, 3706–3719. https://doi.org/10.1111/gcb.14741
6. Xu, P.; Houlton, B.Z.; Zheng, Y.; Zhou, F.; Ma, L.; Li, B.; Liu, X.; Li, G.; Lu, H.; Quan, F. Policy-Enabled Stabilization of Nitrous Oxide Emissions from Livestock Production in China over 1978–2017. Nat. Food 2022, 3, 356–366. https://doi.org/10.1038/s43016-022-00513-y
7. Zhang, F.; Gao, X.; Wang, J.; Liu, F.; Ma, X.; Cao, H.; Chen, X.; Wang, X. Sustainable Nitrogen Management for Vegetable Production in China. Front. Agric. Sci. Eng. 2022, 9, 373–385. https://doi.org/10.15302/J-FASE-2022455
8. Cui, X., Zhou, F., Ciais, P. et al. Global mapping of crop-specific emission factors highlights hotspots of nitrous oxide mitigation. Nat. Food 2, 886–893 (2021). https://doi.org/10.1038/s43016-021-00384-9
9. Tian, H.; Xu, R.; Canadell, J.G.; et al. A Comprehensive Quantification of Global Nitrous Oxide Sources and Sinks[J]. Nature 2020, 586, 248–256.
10. Nelson, D.A. European Environment Agency. Colorado Journal of International Environmental Law and Policy 1999, 10, 153.
11. Gale, F. China’s Agricultural Trade: Issues and Prospects; Economic Research Service, USDA: Washington, DC, USA, 2007; p. 65.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I’ve confirmed. Thank you.

Author Response

Response: Thank you for your suggestion. We have improved the figures and tables in the manuscript.

Author Response File: Author Response.docx