Assessing the Influence Factors of Agricultural Soils’ CH₄/N₂O Emissions Based on the Revised EDGAR Datasets over Hainan Island in China

Abstract

Global warming poses a significant environmental challenge, which is primarily driven by the increase in greenhouse gas concentrations. In this study, we aimed to investigate the factors influencing CH₄/N₂O emissions from agricultural soils over Hainan Island, China, from 2009 to 2018. To achieve this, we selected air temperature, precipitation, and solar radiation as climate factors and categorized farmland as paddy or non-paddy, using revised EDGAR greenhouse gas datasets involving the bias correction method, and geographical detector analysis, multiple linear regression models, and bias sensitivity analysis were used to quantify the sensitivity of climate and land use. The maximum air temperature emerged as the primary factor influencing CH₄ emissions, while the mean air temperature predominantly affected N₂O emissions. The ratio of paddy field area to city area emerged as the second most influential factor impacting CH₄/N₂O emissions. The mean CH₄/N₂O emission intensity from paddy fields was significantly higher (0.42 t·hm⁻²/0.0068 t·hm⁻²) compared to that of non-paddy fields (0.04 t·hm⁻²/0.002 t·hm⁻²). Changes in maximum air temperature under global warming and crop irrigation practices profoundly affect greenhouse gas emissions on Hainan Island. Specifically, the emission intensities of CH₄ and N₂O increased by 14.2% and 11.14% for each Kelvin warmer, respectively.

1. Introduction

The results of the sixth IPCC report indicate that, in high-emission scenarios, near-surface air temperature is projected to rise by +4.0 °C above pre-industrial levels [1]. This temperature increase is expected to exacerbate a range of climate-related issues, including extreme droughts and floods, heat waves, and rising sea levels that threaten the safety of coastal cities [2,3,4]. Global warming also profoundly affects the soil microbial activity and soil organic carbon [5]. Numerous studies have shown that the continuous increase in greenhouse gas emissions (GHGe) is the primary driver of global warming [6,7]. Carbon dioxide (CO₂), methane (CH₄), and nitrous oxide (N₂O) are the three most significant greenhouse gases responsible for global warming. CH₄ and N₂O, also known as non-carbon-dioxide greenhouse gases (GHG-NCO₂), have warming potentials 25 and 298 times greater than CO₂ over a 100-year period, respectively [8]. In recent years, GHG-NCO₂ have received significant attention in the fields of atmospheric science and ecology due to the challenges associated with accurately calculating emissions and the wide range of sources of these gases [9,10].

Agriculture is a significant contributor to greenhouse gas (GHG) emissions in the atmosphere [10]. Globally, agricultural activities account for 24% of total GHGe [11]. In China, the growth of GHGe from agricultural activities was higher than the global average from 1993 to 2007 [12]. The emission percentages of CH₄ and N₂O originating from agricultural activities in China were 40.2% and 59.5%, respectively [13]. Further analysis of a single species of GHGe found that 50% (10%) of agricultural CH₄ (N₂O) emissions resulted from paddy fields [14]. Paddy fields, including irrigated and rain-fed paddy, are important sources of GHGe caused by human activities. As a populous country, China experiences a constant and increasing demand for food, and paddy production accounts for approximately 43.7% of the country’s total grain production [15]. The main paddy fields are located in the south of China (southeastern coastal, Yangtze River basin) in a sub-tropical/tropical humid climate region and the northeast of China in a sub-humid and temperate climate region. Due to the different climatic conditions in China, paddy yields are higher in the tropical/sub-tropical humid region compared to the temperate region [16]. In addition, the use of irrigation methods in southern China results in higher GHGe compared to rain-fed practices in the same region, especially from paddy fields [7].

GHGe from agricultural soil are strongly correlated with temperature and precipitation [17,18]. Soil microorganisms play a critical role in driving biochemical processes that lead to GHGe and are sensitive to temperature changes [19,20]. Temperature changes can affect the microbial decomposition rate, carbon/nitrogen mineralization, bacterial activity, and other factors, resulting in corresponding changes in GHGe [21]. Global warming is expected to affect rice growing and GHGe from paddy fields [22]. Multi-site experiments in China have shown a parabolic relationship between CH₄ emissions from major paddy fields and background air temperature [23]. However, the impact of climate change on regional-scale GHGe remains unclear, and the intensity of emissions (GHG_intensity) from various crops in different regions may vary in warmer conditions [24]. Prolonged warming can hamper crop growth and increase GHGe if temperatures exceed a certain threshold, particularly in hot–humid climates [25]. Some ecosystem models estimate the relationship between GHGe and environmental factors. In the Dynamic Land Ecosystem Model (DLEM), the fluxes of CH₄ and N₂O are mainly affected by annual climate variations. Annual precipitation is an important factor in determining annual CH₄ emissions, while annual mean temperature and precipitation are strongly correlated with annual N₂O emissions [26]. GHGe from agricultural soil are also influenced by land use and management practices such as irrigation type, fertilization, and weeding. Irrigation is the second largest contributor to the total carbon footprint of crop production in China (22%) [12]. Different management practices may also influence GHGe from agricultural soil. For example, Wang et al. assessed GHGe from irrigation processes based on irrigation management practices [27], while Lamb et al. evaluated the potential to reduce GHGe through land use management [28]. Although numerous studies have investigated the effects of climatic and land use factors on GHGe, relatively few studies have investigated the combined effects of these factors.

There are two main ways to investigate GHGe. The top-down approach is based on monitoring atmospheric GHG concentrations and simulations of chemical processes [29,30,31], while the “bottom-up” approach involves point/field-scale ground observations [32]. The Emission Database for Global Atmospheric Research (EDGAR) bottom-up emissions inventories developed by the Joint Research Centre (JRC) is a well-known and widely used emissions inventory that provides monthly emission grid data with a resolution of 0.1° for each sector and country, making it ideal for atmospheric modeling and GHG emissions studies [33]. EDGAR can be used to assess the effects of high emissions and is also used as input for air quality modeling software [34]. However, previous studies have shown that the EDGAR data may not be accurate; thus, bias correction is required [35].

Although a series of studies on GHGe have been conducted, the specific factors influencing GHG_intensity in paddy and non-paddy fields in tropical regions remain unclear. This study aims to address this gap by investigating GHG_intensity in agricultural soils using EDGAR data on Hainan Island. The specific objectives of this study are as follows: (1) to investigate the extent to which climate and land use factors influence GHGe in agricultural soils; (2) to use a multiple linear regression (MLR) method to construct a model between paddy/non-paddy areas and GHG_intensity; and (3) to identify the primary climate drivers affecting GHG_intensity in paddy fields.

2. Materials and Methods

2.1. Study Domain

Located at the southernmost point of China, Hainan Island spans geographical coordinates from 18°10′ to 20°07′ N and 108°37′ to 111°03′ E. With an estimated area of 34,000 km², it comprises 18 cities, as illustrated in Figure 1. This island presents a tropical monsoon maritime climate, with a median air temperature fluctuating between 23.8 °C and 26.2 °C and regional annual rainfall quantities fluctuating between 1000 mm and 2500 mm. The major soil type is latosol and rice soil (pH ≈ 5.0) [36]. A variety of agricultural practices are implemented throughout Hainan Island due to these diverse hydrothermal and geographical conditions, predominantly in paddy and non-paddy fields [37,38]. The primary crop cultivated here, rice, flourishes in paddy fields. In contrast, non-paddy fields are mostly used for the cultivation of soybeans, maize, and other crops, with most of the cultivable land being located along the coastlines.

2.2. Study Data

2.2.1. GHG Data

In this study, the monthly data on agricultural soil emissions were acquired from the Emissions Database for Global Atmospheric Research (EDGAR) version 6.0 [34], with a spatial resolution of 0.1° and a temporal resolution of monthly scales. EDGAR v6.0 provides emission products from 1970 to 2018. The EDGAR emission source inventory is divided into countries and spatial grids, including power, industrial, residential, agriculture, surface transport, and shipping sectors for greenhouse gases such as CO, NOx, SO₂, etc. The data are recorded in grid points as emission intensity in kg·m⁻²·s⁻¹. As several studies have pointed out, the EDGAR database exists with uncertainty compared with the various observed datasets. This study revised the raw EDGAR data based on a correction coefficient to ensure the credibility of the result analysis.

2.2.2. Climate Data

The climate data used in this study were collected from the China Meteorological Forcing Dataset (CMFD) [39]. CMFD is China’s first regional multi-meteorological forcing dataset developed for the study of land surface processes, with long time series and high spatial, including daily products of precipitation (Prec, mm), air temperature (°C), and downward shortwave radiation (Srad, w·m⁻²). The follow-up study utilized the highest, mean, and lowest values of daily data to calculate the annual maximum air temperature (Tmax), annual mean air temperature (Tmean), and annual minimum air temperature (Tmin). The mean of the daily data was used to represent the annual data for all other climate factors. The spatial resolution of the data is 0.1°. All climate data ranged from 2009 to 2018.

2.2.3. Land Use and Land Cover Change Data

The land use and land cover change (LUCC) data were derived from the Hainan Statistical Yearbook [40], which includes the city area (CA, ha), cultivated land area (CL, ha), and paddy field area (PF, ha). All the above data are annual data from 2009 to 2018 for 18 different cities in Hainan. The non-paddy field area (NF) was calculated using CL−PF. The ratio of cultivated land area to city area (RC) was calculated using CL/CA. The ratio of paddy field area to cultivated land area (RP) was calculated using PF/CL.

2.3. Methodology

2.3.1. Coefficient Bias Correction Method

In this study, we used a coefficient bias correction method to revise the raw EDGAR data [41]. Generally, when the GHGe of EDGAR (EDGAR_China) are consistent with the GHGe of existing research (GHG-NCO2_results), the data used in this study do not need to be corrected. However, if the EDGAR_China is significantly higher or lower than GHG-NCO2_results, it is necessary to correct the GHG data of EDGAR to improve the accuracy and credibility of the results.

$${\mathsf{\beta }}_{\mathrm{j}}=\frac{\mathrm{G}\mathrm{H}\mathrm{G}-\mathrm{N}\mathrm{C}\mathrm{O}{2}_{\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{u}\mathrm{l}\mathrm{t}\mathrm{s},\mathrm{j}}}{{\mathrm{EDGAR}}_{\mathrm{C}\mathrm{h}\mathrm{i}\mathrm{n}\mathrm{a},\mathrm{j}}}$$

(1)

where GHG-NCO2_results is the GHGe of existing research in a given year j. It should be noted that there are significant differences in the coefficients β for different years. EDGAR_China is the annual GHGe in China extracted from the raw EDGAR global data.

Therefore, this study used averaged annual value β to systematically correct the EDGAR data from 2009 to 2018. If there was more than one coefficient value of β for the same year, the mean value for that year was calculated first, and then the annual average β was calculated to ensure consistent correction weights for each typical year involved in the calculation. The revised EDGAR can be bias-corrected as follows:

$$\mathrm{E}\mathrm{D}\mathrm{G}\mathrm{A}{\mathrm{R}}_{\mathrm{R}\mathrm{e}\mathrm{v}\mathrm{i}\mathrm{s}\mathrm{e}\mathrm{d}}=\overline{\mathsf{\beta }}·\mathrm{E}\mathrm{D}\mathrm{G}\mathrm{A}{\mathrm{R}}_{\mathrm{H}\mathrm{N}}$$

(2)

where EDGAR_Revised is the GHG-NCO₂ emissions of Hainan Island following bias correction from 2009 to 2018, and EDGAR_HN is Hainan Island data extracted from EDGAR raw data used in follow-up research.

2.3.2. Geographical Detector

Spatial differentiation is an essential feature of geographical phenomena. Geographic detectors are statistical tools that are used to detect spatial differentiation and reveal its drivers. This study uses two common sub-modules: factor detector and interaction detector [42].

The factor detector measures the degree of influence of geographical factor X on the dependent variable Y using the q-value. X represents 9 factors (CL, PF, RC, RP, Tmax, Tmin, Tmean, Prec, and Srad); Y represents the GHG_intensity.

$$\mathrm{q}=1-\frac{{\displaystyle \sum _{\mathrm{h}=1}^{\mathrm{L}}{\mathrm{N}}_{\mathrm{h}}{\mathsf{\sigma }}_{\mathrm{h}}^2}}{{\mathrm{N}\mathsf{\sigma }}^2}$$

(3)

where q is the explanatory power of each influencing factor X on the variable Y, taking values in the range of [0, 1], and its more considerable value indicates the more decisive influence of X on the spatial variation of Y; h = 1, …, L, where L represents the stratification of X and Y; N_h and N refer to the number of layers h and the units of the whole region, respectively; and σ²_h and σ² are the variances of Y in layer h and the whole region, respectively.

The interaction detectors can be used to identify interactions between different geographic factors X; that is, whether the combined effect of geographic factor 1 (such as Prec) and geographic factor 2 (such as Srad) enhances or weakens their explanatory power on variable Y or the effects of these influences on variable Y will be independent of each other. The first step of the evaluation method is to calculate the q-values of geographic factors, respectively: q (X₁) and q (X₂), then calculate the q-values of factor interactions: q (X₁ ∩ X₂) and finally compare q (X₁), q (X₂) and q (X₁ ∩ X₂). The five interactions of the two factors are shown in

Table 1. Interaction relationship between two geographical factors.

Interaction	Criteria
Nonlinear–weaken	q (X1 ∩ X2) < Min [q (X1), q (X2)]
Unilinear–weaken	Min [q (X1), q (X2)] < q (X1 ∩ X2) < Max [q (X1), q (X2)]
Bivariate–enhance	Max [q (X1), q (X2)] < q (X1 ∩ X2)
Independent	q (X1 ∩ X2) = q (X1) + q (X2)
Nonlinear–enhance	q (X1 ∩ X2) > q (X1) + q (X2)

.

2.3.3. Multiple Linear Regression

Multiple linear regression models (MLR) are statistical models that are used to explain the dependent variable with multiple independent variables. They can be applied to predict GHGe by analyzing the relationship between environmental and soil properties. This study considers GHGe as the dependent variable (Y) and the areas of paddy and non-paddy fields as the independent variables (X₁ and X₂, respectively). The formula for the MLR model is presented below:

$${\mathrm{Y}}_{\mathrm{i}}{=\mathsf{\alpha }}_1{\mathrm{X}}_1{+\mathsf{\alpha }}_2{\mathrm{X}}_2+\mathsf{\epsilon }(\mathrm{i}=1,2,···,\mathrm{n})$$

(4)

where i is the year; Y_i is the GHGe in each year; X₁ and X₂ are the areas of paddy and non-paddy fields in each year; α₁ and α₂ are the regression coefficients, which mean the GHG_intensity factors for paddy and non-paddy fields, respectively; and ε (i = 1, 2, ···, n) is the residual; since the GHG has no emissions in the no-till scenario, this study forces an intercept of 0 in the model construction.

2.3.4. Pearson Correlation Analysis

Pearson correlation analysis is an empirical formula that is used to obtain a correlation coefficient that reflects the linear relationship between two or more variables. The differences were considered to be statistically significant when p < 0.05 or 0.01. This study was used to calculate the correlation between climate factors and GHG_intensity. The formula is calculated as follows:

$${\mathrm{R}}_{\mathrm{xy}}=\frac{{\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}{(\mathrm{x}}_{\mathrm{i}}-\overline{\mathrm{x}}}{\left)\right(\mathrm{y}}_{\mathrm{i}}-\overline{\mathrm{y}})}{\sqrt{{\displaystyle {\sum }_{\mathrm{n}=1}^{\mathrm{n}}{{(\mathrm{x}}_{\mathrm{i}}-\overline{\mathrm{x}})}^2{\displaystyle {\sum }_{\mathrm{n}=1}^{\mathrm{n}}{(\mathrm{y}}_{\mathrm{i}}-\overline{\mathrm{y}}}{)}^2}}}$$

(5)

where R_xy is the correlation coefficient between two variables x and y; x is the climate factors; y is the GHG_intensity.

2.3.5. Bias Sensitivity Analysis

This study used a climate sensitivity factor to measure the effect of temperature change on GHG_intensity from agricultural soils. The climate sensitivity factor was defined as the emission rate ratio to the temperature change rate. The formula is calculated as follows:

$${\mathrm{S}}_{\mathrm{x}}=\underset{△\mathrm{x}/{\mathrm{x}}_{\mathrm{i}}}{\lim }(\frac{△\mathrm{E}/\mathrm{E}}{△\mathrm{x}/\mathrm{x}})=\frac{\partial \mathrm{E}}{\partial \mathrm{x}}·\frac{\mathrm{x}}{\mathrm{E}}$$

(6)

where S_x is the dimensionless sensitivity factor of GHG_intensity with regard to the temperature factor x, and E is the potential GHG_intensity rate. The advantage of this equation is that the climate factor is dimensionless through the rate of change, which can facilitate comparison between different studies [43]. The larger the absolute value of the sensitivity factor, the greater the effect of the climate variable on GHG_intensity.

3. Results

3.1. Bias Correction and Uncertainty of Agricultural Soil from EDGAR Data

We included the results of previous studies for agricultural soil GHGe from different years to the greatest extent possible, and then calculated the corresponding annual GHGe of EDGAR for the same year. In this study, we summarized results including two different periods with fourteen typical years of CH₄ ranging from 1980 to 2018 (

Table 2. CH₄ emissions datasets of EDGAR and references (unit: Tg yr⁻¹).

Year	EDGAR	References	Emissions	Bias Correction Coefficient (β)
2018	13.89	Zhang et al. [44]	6.40	0.46
2015	14.16	Huang et al. [45]	9.77	0.69
		Gong and Shi [46]	11.45	0.81
2014	13.95	CSBUR 1 [13]	7.35	0.53
		Du et al. [47]	5.78	0.41
2012	13.82	Wang et al. [48]	8.20	0.59
2011	13.83	Shang et al. [49]	9.69	0.70
2010	13.74	Shang et al. [49]	9.58	0.70
		Peng et al. [50]	7.40	0.54
		TNCCC 2 [51]	8.73	0.64
2009	13.63	Shang et al. [49]	9.61	0.71
		Chen and Pan [52]	11.16	0.82
2008	13.42	Chen et al. [53]	8.10	0.60
		Shang et al. [49]	9.53	0.71
2007	13.17	Kai et al. [54]	7.67	0.58
		Zhang et al. [55]	5.54	0.42
2005	12.87	Yue et al. [56]	6.99	0.54
		SNCCC 3 [57]	7.93	0.62
2003	11.59	Fu and Yu. [58]	5.25	0.45
2000	12.65	Xie and Wang [59]	9.26	0.73
		Streets et al. [60]	9.78	0.77
1990	16.23	Peng et al. [50]	10.00	0.62
1980	18.88	Peng et al. [50]	11.20	0.59
Mean	13.99			0.62

¹ CSBUR: The People’s Republic of China second biennial update report on climate change. ² TNCCC: The People’s Republic of China third national communication on climate change. ³ SNCCC: The People’s Republic of China second national communication on climate change.

) and five typical years of N₂O (

Table 3. N₂O emission datasets of EDGAR and references (unit: Tg yr⁻¹).

Year	EDGAR	References	Emissions	Bias Correction Coefficient (β)
2007	0.35	Zhang and Jv [61]	0.29	0.81
2005	0.37	SNCCC 1 [57]	0.67	1.82
1997	0.34	Lu et al. [62]	0.29	0.85
1995	0.36	Yan et al. [63]	0.48	1.34
		Xing [64]	0.40	1.12
1990	0.32	Wang and Li [65]	0.31	0.98
		Li et al. [66]	0.31	0.98
Mean	0.35			0.98

¹ SNCCC: The People’s Republic of China second national communication on climate change.

) ranging from 1990 to 2007. In this study, we used a coefficient bias correction method to correct the raw EDGAR data, and the bias correction coefficient of β can be calculated by Equation (1), and the revised EDGAR can be bias-corrected as in Equation (2).

The averaged bias correction coefficients of β illustrate the differences between two greenhouse gases. The coefficient of CH₄ (0.62) is much less than one, indicating that the agricultural soil CH₄ emissions from EDGAR are seriously overestimated. Although the bias correction coefficient of N₂O (0.98) is close to one, the fluctuation range of the coefficient is larger than that of CH₄, and the uncertainty of the data on agricultural soil N₂O emissions is higher (in Figure 2).

A comparison of total GHGe pre and post correction suggests that EDGAR tends to overestimate the CH₄ emission intensity from agricultural soils, as depicted in Figure 3a. Taking 2018 as an example, the emissions dropped from 14.05 million tons per year before correction to 8.71 million tons per year post correction. CH₄ emissions steadily increased from 2009 through 2015 before tapering off slightly in 2016. Despite a brief resurgence from 2016 to 2017, there was a decline from 2017 to 2018. As the bias correction coefficient for N₂O emissions is 0.98 (indicating β is very close to 1), the total N₂O emissions before correction closely mirrored the EDGAR_Revised data from 1990 to 2018. Notable trends include a significant decrease from 2009 to 2014 and an increase from 2014 to 2016, with a clear inflection point and the lowest value over the past decade recorded in 2017, as shown in Figure 3b.

3.2. Quantification of the Influencing Factors of GHG_intensity and Their Interactions

Using a geographical detector (in Figure 4a,b), we have quantified the main factors that affect GHG_intensity in agricultural soil. The ratio of paddy field area to the total cultivated land area (RP) was identified as the most vital explanatory factor (q = 0.46) in influencing CH₄ emission intensity, succeeded by maximum air temperature (Tmax) (q = 0.46). These variables demonstrated significantly stronger explanatory power than others, such as the total paddy field area (PF) and minimum air temperature (Tmin). Major factors impacting N₂O emission intensity included average temperature (Tmean) (q = 0.80), total cultivated land area (CL) (q = 0.60), RP (q = 0.65), and Tmax (q = 0.63).

To further investigate the effect of the interaction between the nine factors mentioned above on the GHG_intensity, an interaction detector was used for the analysis. This detector can assess the strength of the interaction between two independent factors based on their ability to explain surface temperature. Our results are presented in Figure 5. In CH₄, the interaction between Prec and Tmax explains the largest extent of CH₄ emission intensity, reaching 0.909. This indicates that the interaction between Prec and Tmax is the main factor influencing the spatial distribution pattern and heterogeneity of GHG_intensity on Hainan Island. In terms of N₂O, the interaction between Srad and Tmean was the strongest in relation to the intensity of N₂O emissions, with an explanation of 0.971, followed by Srad and RC, with an explanation of 0.968. We found that the interaction between Prec/Srad of CH₄ and arbitrary natural factors showed a non-linear enhancement. This suggests that the interaction between Prec/Srad and any natural factor has a greater impact on the GHG_intensity than the sum of their individual effects, which is also reflected for N₂O.

3.3. Quantifying the Sensitivity of GHGe in Different Types of Cultivated Land

According to different irrigation conditions, cultivated land is divided into three categories: dry land, dry field, and paddy field, as recommended by the Hainan County Statistical Yearbook. As dry lands and dry fields are associated with rain-fed cultivated lands, we further integrated these two types of land use into non-paddy fields. Non-paddy fields accounted for 60% of the total area of cultivated land on Hainan Island in 2018, although they failed to provide an adequate water supply during the dry season. To investigate the difference in agricultural soil GHG_intensity between paddy fields and non-paddy fields, a multiple linear regression model was used to separate CH₄ and N₂O emission intensity in the two different cultivation methods.

The scatterplots depicting the simulated and measured results from the model demonstrate that the CH₄ model (in Figure 6a) meets the requirements of model construction in terms of both stability (with a slope of 0.81, close to the 1:1 line) and accuracy (R² = 0.727, p < 0.01). Similarly, the N₂O model (in Figure 6b) exhibits stability (with a slope of 1.1, close to the 1:1 line) and accuracy (R² = 0.878, p < 0.01) that meet the criteria for model development. Notably, the simulation effect of the N₂O model was superior to that of the CH₄ model.

$$\mathrm{G}\mathrm{H}{\mathrm{G}}_{\mathrm{s}\mathrm{i}\mathrm{m},\mathrm{i},\mathrm{j}}={\mathsf{\alpha }}_{\mathrm{j}}\mathrm{a}\mathrm{r}\mathrm{e}{\mathrm{a}}_{\mathrm{p}\mathrm{a}\mathrm{d}\mathrm{d}\mathrm{y},\mathrm{i},\mathrm{j}}+{\mathsf{\beta }}_{\mathrm{j}}\mathrm{a}\mathrm{r}\mathrm{e}{\mathrm{a}}_{\mathrm{n}\mathrm{o}\mathrm{n}-\mathrm{p}\mathrm{a}\mathrm{d}\mathrm{d}\mathrm{y},\mathrm{i},\mathrm{j}}$$

(7)

$${\mathrm{GHG}}_{\mathrm{E},\mathrm{i},\mathrm{j}}={\mathrm{Intensity}}_{\mathrm{r}\mathrm{a}\mathrm{w}\mathrm{E}}{}_{,\mathrm{i},\mathrm{j}}·\mathrm{a}\mathrm{r}\mathrm{e}{\mathrm{a}}_{\mathrm{c}\mathrm{ity},\mathrm{i}}·\mathsf{\beta }$$

(8)

where GHG_sim,i,j represents the GHGe after MLR simulation in a given city i and a given year j. The variable α_j represents α₁ in Equation (4), which denotes the GHG_intensity of paddy fields. Similarly, the variable β_j represents α₂ in Equation (4), indicating the GHG_intensity of non-paddy fields. Further, area_paddy,i,j and area_{non-paddy,i,j} represent the areas of paddy fields and non-paddy fields. GHG_E,i,j represents the revised EDGAR emissions data in a given city i and a given year j, and Intensity_rawE,i,j is the emission intensity of the raw EDGAR data in a given city i and a given year j. We use the grid data in each city to calculate the mean value, which represents the emission intensity of the city, and area_city,i is the area of city i. β represents the correction coefficient.

A multiple regression model was used to analyze the GHG_intensity from paddy and non-paddy fields. In Figure 7, the orange bars represent paddy field emission intensity, and the blue bars represent non-paddy field emission intensity. The range of CH₄ emission intensity from paddy fields was 0.39 to 0.49 t·hm⁻², while the range for non-paddy fields was 0.02 to 0.07 t·hm⁻². Similarly, the range of N₂O emission intensity from paddy fields was 0.006 to 0.0072 t·hm⁻², while the range for non-paddy fields was 0.0014 to 0.0029 t·hm⁻². The CH₄ emission intensity from non-paddy fields decreased over time, indicating an increasing difference in emission intensity over a ten-year period. Similarly, the difference in N₂O emission intensity also increased. The average CH₄ emission intensity from paddy fields was 0.42 t·hm⁻², while the average from non-paddy fields was 0.04 t·hm⁻². The ratio of CH₄ emission intensity between paddy and non-paddy fields (Ratio-CH₄) was 9.3. The mean N₂O emission intensity from paddy fields was 0.0068 t·hm⁻², while the average from non-paddy fields was 0.002 t·hm⁻². The ratio of N₂O emission intensity between paddy and non-paddy fields (Ratio-N₂O) was 3.4. Notably, Ratio-CH₄ was 2.75 times larger than Ratio-N₂O, indicating that GHGe are dependent on soil moisture.

3.4. The Main Climate-Driven Forces for GHGe from Paddy Fields

Our study found that GHGe mainly derive from paddy fields. Thus, we present a comparative analysis of paddy field emission intensity below. The geodetector results in Section 3.1 indicate that paddy fields are also important LUCC indicators. Hence, we further investigate which climate factor determines paddy field emission intensity. This study correlated CH₄ and N₂O emission intensity trends from 2009 to 2018 with trends in key climate factors, including Tmax, Tmin, Tmean, Prec, and Srad (

Table 4. Correlation between climate factors and GHG_intensity from 2009 to 2018 over Hainan Island.

Climate Factor	CH4			N2O
	R2	p	Sx	R2	p	Sx
Tmax	0.851	0.002 *	14.20%	0.799	0.006 *	11.14%
Tmin	−0.025	0.946		−0.163	0.652
Tmean	0.419	0.228		0.203	0.575
Prec	−0.504	0.137		−0.342	0.333
Srad	0.360	0.307		0.150	0.680

* Correlation is significant at the 0.01 level.

). We found that all trends in CH₄ emission intensity from paddy fields showed a highly significant positive correlation with trends in Tmax (r = 0.92, p < 0.01). However, there was no significant correlation between CH₄ emission intensity and other climate factors. Similarly, all trends in N₂O emission intensity from paddy fields showed a highly significant positive correlation with Tmax (r = 0.89, p < 0.01), with no significant correlation with other climate factors.

From 2009 to 2018, there was a strong consistency between the fluctuation of Tmax and CH₄ emission intensity (R² = 0.851, p < 0.01, Figure 8a). Similarly, N₂O emission intensity was also highly correlated over the same ten-year period (R² = 0.799, p < 0.01, Figure 8b), further validating that Tmax is the most sensitive climate factor affecting GHGe on Hainan Island. Next, we quantified the sensitivity of GHG_intensity to global warming using Equation (6), as mentioned in Section 2.3.4. The sensitivity coefficient S_x for CH₄ was 0.0557, and, for N₂O, it was 0.0007. These results suggest that each per Kelvin increase in temperature would result in an increase of 14.2% and 11.14% in CH₄ and N₂O emission intensity, respectively.

4. Discussion

4.1. The Sensitivity of Climate Change to GHG from Agricultural Soils

GHGe from agricultural soils are strongly influenced by climate conditions and soil status. Soil microorganisms are the drivers of many biochemical processes, and the microbial activity of CH₄ and N₂O in the soil is also regulated by the hydrothermal conditions of the environment, including denitrifying bacteria, nitrifying, methanogenic and methane-oxidizing, etc. [19,20]. Deposition of unstable C and N in root secretions and soil provides the main substance source for GHG-NCO₂ production. In previous studies, it was found that the increase in soil surface temperature can increase the content of organic carbon in soil and then promote the increase in soil N₂O and CH₄ emissions [67]. When the number of methanogens did not change, the soil CH₄ emission rate was the fastest in the season with the highest temperature [68]. It is also believed that the emission of CH₄ in paddy fields is affected by the increase in temperature. The increase in temperature below 2 °C can promote CH₄ emission, while, above 2 °C, it will inhibit CH₄ emission [69]. In addition, experimental field studies have shown a positive correlation between CH₄ emissions and the temperature of continuously flooded rice paddies with sufficient organic matter [15]. This study found that Tmax was the primary climate factor influencing CH₄ emissions, while Tmean was the primary influence factor for N₂O emissions. We also confirmed a notable correlation between Tmax and GHG_intensity, with a 14.2% increase in CH₄ emission intensity and an 11.14% increase in N₂O emission intensity in paddy fields per Kelvin Tmax increase. The results were similar to the previous study [70]. For every 10 °C increase in paddy field temperature, CH₄ emissions would increase by 25 times in this study. These observations may be due to the increased activity of microorganisms in a warm environment, resulting in changes in their respiration rate and activity, thereby changing the GHG_intensity.

Soil moisture content is also an important factor in environmental aspects, which are mainly regulated by precipitation and evapotranspiration in nature [71]. Rapid increases in soil water content due to rainfall disrupt the soil aggregate structure, creating an anaerobic environment that stimulates the activity of methanogenic and denitrifying bacteria, leading to the release of carbon and nitrogen [72]. However, excessive rainfall impedes GHG releases [73]. This study observed a negative correlation between precipitation and paddy field intensity. It may be due to frequent rainfall maintaining high humidity, lowering temperatures, and suppressing GHG_intensity. It may also be the case that the paddy fields are mostly irrigated with reliable water sources and irrigation facilities, and fluctuations in precipitation do not adversely impact GHG_intensity.

4.2. Comparison of CH₄ and N₂O Emission Intensity in Different Farming Practices

Paddy fields with irrigation are a strong source of greenhouse gases. Management practices in agricultural cultivation have a significant impact on CH₄ and N₂O emission intensity, especially water management practices [74]. CH₄ emissions from rice-plant-mediated transport account for 80% of the total emissions from paddy fields [75]. A large amount of water is required during rice cultivation, creating an anaerobic environment that inhibits N₂O production but promotes the growth of methanogens, resulting in a large amount of CH₄ emissions [76]. Meanwhile, in rain-fed non-paddy fields, the utilization of nitrogen fertilizer and improved soil drainage conditions might contribute to an increase in N₂O emissions [77]. These studies showed that the presence or absence of artificial irrigation practices plays a key role in GHG emissions from agricultural soils, and the same is also reflected in the current study. We showed that the GHGe of 18 cities is largely determined by the proportion of the total area of paddy fields and that, under conditions of adequate water supply (usually provided by irrigation), the emission intensities of CH₄ and N₂O from paddy fields were 9.3 times and 3.4 times higher, respectively, than those from non-paddy fields. This suggests that the difference in the emission intensity of N₂O between the two farming practices is smaller compared to the emission intensity of CH₄ under different farming practices. Soil permeability is a major factor affecting CH₄ emission [78], and diffusion of CH₄ in soil can limit the rate of CH₄ oxidation [79]. This may lead to the fact that, in agroecosystems, the CH₄ emission from non-paddy fields is much smaller than that from paddy fields. Under flooded conditions, 87% of N₂O emissions were transported through rice plants and 18% were transported under non-flooded conditions [80]. Owing to their different rates of microbial decomposition and material cycling in different farming practices and physicochemical properties such as soil water content, pH, carbon, and nitrogen content, there are some differences between the emission mechanisms of CH₄ and N₂O. The differences between Ratio-CH₄ and Ratio-N₂O in this study also confirmed this point. Wang et al. [81] also concluded that dryland fields emit more N₂O and less CH₄ than paddy fields and that the overall global warming potential has decreased.

Therefore, the conversion of paddy fields to non-paddy fields may be beneficial in reducing GHGe from agricultural soils. This conclusion can be used as a basis for adjusting land use strategies. The tropics are an important source of CH₄ emissions, with the region accounting for about 60% of total global CH₄ emissions [82]. So, we can mitigate the increase in global warming by adjusting the cultivation structure of the tropics to reduce GHGe. The use of rice varieties affects yield and GHGe. Rice variety replacement in 2000 significantly increased rice yields and reduced GHG emissions compared to 1960 varieties [83]. The advocacy for modern rice varieties emerges as a novel approach to mitigating environmental impacts. For example, the water-saving and drought-resistant rice with dry cultivation (D-WDR) cropping model can achieve GHGe mitigation and yield stabilization [74]. For rice producers, it is also a win–win model that balances rice yield and GHGe reduction.

4.3. Limitations

The results of this study are more reliable, but there are some limitations. First, the geographic detectors exploring GHG_intensity selected whether to use irrigation measures as an influencing factor of farmland management measures; however, different farming practices and fertilization measures may affect the results of the study [84]. Then, the research objects selected in this study to explore GHGe are mainly paddy fields and non-paddy fields; however, other crop types and different soil types have also contributed to GHGe [21]. Finally, this study investigated the correlation between air temperature and GHGe, and the relationship between soil temperature and GHGe can be considered because soil temperature influences the decomposition cycle process of microorganisms in the soil, which may lead to different research results. Therefore, in future studies, more environmental indicators and whether there are correlations and interactions between different anthropogenic cropping choices and farm management practices on GHGe emissions could be considered.

5. Conclusions

The EDGAR agricultural soil data were bias-corrected to produce annual correction coefficients, with average bias correction coefficients of 0.62 for CH₄ and 0.98 for N₂O. The maximum air temperature constituted the primary climatic factor influencing CH₄ emissions, whereas the mean air temperature prominently affected N₂O emissions. Regarding land use, the ratio of the paddy field area to the total region was a significant component influencing both CH₄ and N₂O emissions. The mean CH₄/N₂O emission intensity of paddy under sufficient water supply conditions is 0.42 t·hm⁻²/0.0068 t·hm⁻², while that of non-paddy is 0.04 t·hm⁻²/0.002 t·hm⁻², with Ratio-CH₄ and Ratio-N₂O being 9.3 and 3.4, respectively. The Tmax showed the greatest consistency with GHG_intensity from paddy (CH₄: R² = 0.851, p < 0.01; N₂O: R² = 0.799, p < 0.01). The per Kelvin increase in temperature heightened the CH₄ and N₂O emission intensity in paddy fields by 14.2% and 11.14%. This emission intensity is anticipated to manifest non-stable positive feedback within the context of global warming.