Model Construction And Mitigation Potential Assessment In Carbon Sequestration And Greenhouse Gases Emission Of Paddy Fields In China

Rice is a major food crop,and rice cultivation is an important source of greenhouse gas emissions.Meanwhile,paddy soil also has a huge carbon sequestration potential.The development of reliable quantification methods for accurate estimation of national-scale paddy field greenhouse gas emissions and topsoil organic carbon changes is of great significance to clarify the contribution of rice cultivation to climate change.Based on literature collection,this study constructed a database including the observation of greenhouse gas emissions and topsoil organic carbon changes in rice fields in China.Considering soil,management and climate variables,the models of rice yield,topsoil organic carbon changes,CH₄ and N₂O emissions were established by using multiple regression method and three machine learning algorithms（Random Forest,Support Vector Machine and Artificial Neural Network）,and the effects of various variables and their relative contribution to the models were explored.Based on model evaluation and optimal model screening,climate data under current situation（2018）and future climate change scenarios（2050）were input into the optimal models to estimate and predict the characteristics of rice yield,topsoil organic carbon changes,CH₄ and N₂O emissions in Chinese paddy field under current climate and future climate change scenarios.The spatial distribution pattern of climate change impacts was analyzed and the key driving factors were identified.Finally,the challenges of sustainable rice field management under future climate change were discussed through scenario analysis.The main results are as follows:1.Based on the collection of published literatures,247 effective literatures on greenhouse gas emissions and topsoil organic carbon changes in paddy fields of China were obtained,and a database including 536 sets of rice yield,1100 sets of CH₄ emissions,578 sets of N₂O emissions and 638 sets of topsoil organic carbon changes was established.Spatial datasets including soil properties,management scenarios,current climate（2018）and future climate change scenarios（2050）were constructed for spatial simulation and prediction of greenhouse gas emissions and topsoil organic carbon changes in paddy fields through spatial processing of data.2.Based on the rice yield dataset,a multiple regression model and three machine learning models were constructed,taking into account the effects of climate,soil and farmland management.According to the model evaluation index,it was found that the random forest model had the best performance,with the highest R² of 0.72 and model efficiency of 72%.It was further found that chemical nitrogen fertilizer application was the most key driving variable affecting rice yield,and its contribution to the model was 11.8%.This study also found that,the increase effect of livestock manure application on rice yield was greater than that of green manure and straw,and mid-term drainage,high CO₂ concentration and temperature can increase rice yield.It was finally estimated that the total production of rice from Chinese paddy field in 2018 was 245.36 Tg（95% confidence interval of 233.09²57.63 Tg）using high spatial resolution data.The Yangtze River region is the main rice producing areas,accounting for 30.2% of China’s total rice production,followed by the Southwest and Northeast agricultural region.3.Based on the CH₄ emission dataset,a set of CH₄ emission models reflecting the characteristics of agricultural regions and three machine learning models were established.Based on the model evaluation indexes,it was found that the region-specific multiple regression models had better model performance than the single national model.The performance of random forest model with a R² of 0.64 and model efficiency of 63% is better than multiple regression models.Pre-season water regime was the most critical variable affecting CH₄ emissions from paddy fields.Compared with the long-term drainage pre-season water regime,the short-term drainage and flooding pre-season water regime resulted in more CH₄ emissions.Water regime during rice growing season and organic materials input also made significant contributions to the model,which were 6.9% and 11.0%,respectively.When the average temperature of rice growing season exceeds about 25℃,the temperature rise may offset part of the CH₄ emissions stimulated by the increase of atmospheric CO₂ concentration.The total CH₄ emissions from Chinese paddy field in 2018 were estimated to be 6.12 Tg CH₄（95% confidence interval of 5.81⁶.43 Tg CH₄）using high spatial resolution data,of which the CH₄ emissions from singleseason rice planting accounted for 64.4%,while early rice,and late rice contributed 15.8% and 19.8% of the total emissions,respectively.4.Based on the N₂O emission dataset,a multiple regression model and three machine learning models were developed.According to the model evaluation index,it was found that the random forest model had the best performance,with the highest R² of 0.59 and model efficiency of 51%.It was further found that chemical nitrogen fertilizer application was the most critical variable affecting N₂O emission,and its contribution to the model was quantified as 11.1%.Soil organic carbon were also the important variable,which contributing 7.5% to the model.This study also found that increasing CO₂ concentration and temperature would promote N₂O emission from paddy fields.Using high spatial resolution data,the total N₂O emissions from Chinese paddy field in 2018 were estimated to be 36.48 Gg N₂O emissions（95% confidence interval of 32.83⁴0.86 Gg N₂O）.Single-season rice planting accounted for 76.9%,while early rice and late rice contributed 11.2% and 11.9% to the total emissions,respectively.The crop rotation type（i.e.paddy rice-upland crop rotation）and high nitrogen fertilizer application rate were the reasons for the high N₂O emissions in Southwest agricultural region.5.Based on the dataset of topsoil organic carbon changes,a multiple regression model and three machine learning algorithms were developed.Random forest model had the best performance,with the highest R² of 0.64 and model efficiency of 63% compared with other models.The variables entered into the random forest model included organic materials input,experimental duration,initial soil organic carbon content,atmospheric CO₂ concentration,annual temperature and rainfall et al.Organic material input and experimental duration were the most critical variables,which contributed 9.0% and 7.4% to the random forest model,respectively,according to variable importance analysis.The average annual increment of topsoil organic carbon brought by the application of livestock manure was greater than that of straw and green manure,and the topsoil organic carbon content would eventually tend to be stable with the continuous increase of organic material input.When the experimental duration exceeded 16 years,the carbon content tend to be stable gradually.In addition,soils with low initial carbon content had a larger increase in topsoil organic carbon content.The application of nitrogen fertilizer promoted the accumulation of soil carbon,and the optimum soil pH value was about 7 for carbon sequestration.The average annual increment of soil carbon in doublecropping rice rotation was higher than that in single-cropping rice rotation due to longer flooding time and more organic matter input.The increase of atmospheric CO₂ concentration will promote carbon sequestration,and the most suitable annual average temperature for SOC accumulation was about 17℃.The average topsoil organic carbon of paddy fields in China increased from 17.81 g kg^-1 to 21.46 g kg^-1,and the topsoil organic carbon density increased by 7.42 tCha^-1(95% confidence interval is 6.01⁸.83 tCha^-1)during 1981 to 2018.The increment of soil carbon in the Yangtze River region was higher than that in South China,mainly because of the higher annual temperature and lower soil pH in the South.6.Based on four representative concentration pathways（RCPs）and five global climate models（GCMs）,20 sets of future climate data were input into the optimal random forest models to predict the impact of future climate change on rice yield,topsoil organic carbon changes and greenhouse gas emissions in Chinese paddy fields.It was expected that the national total rice production will decrease slightly and total greenhouse gas emissions will increase by 2.1%-4.9% by 2050 if the current management measures remain unchanged.There are large spatial variations in rice yield and greenhouse gas emissions under future climate change.Taking the RCP8.5 scenario as an example,the rice yield reduction in Yangtze River region may be the largest due to the high temperature rise,while the rice yield and CH₄ emissions in Northeast region will increase.Correlation analysis showed that temperature was the most critical climate factor affecting rice yield and greenhouse gas emissions in various agricultural regions.Compared with current climate conditions,the change of topsoil organic carbon pool in paddy fields under future climate change was weak by 2050（the average annual loss of 10.78-57.89 Gg C）.Furthermore,the variability of different GCMs in predicting yield and greenhouse gas emissions was close to or even higher than that among the four RCPs,suggesting that the accuracy of future climate change data is very important.When considering the optimization of water regime and reasonable reduction of nitrogen fertilizer application,greenhouse gas reduction of 13.64¹4.66 Tg CO₂-eq can be brought.However,taking into account the negative impact of climate change,the optimal management measures can only bring 2.86⁹.16 Tg CO₂-eq emission reduction,if compared with 2018.In conclusion,based on the literature data collection,this study systematically established a series of empirical models of rice yield,topsoil organic carbon changes and greenhouse gas emissions in Chinese paddy fields.The random forest model had the best performance,the key variables affecting the models and the nonlinear marginal effects were further explored.It was found that future climate change may increase the risk of greenhouse gas emissions and there was a large spatial variability.Optimizing farmland management measures for agricultural regions can offset the negative impacts of climate change.This study can provide scientific basis and technical support for coping with and mitigating climate change in Chinese paddy field.