Estimation Method For Carbon Emissions Of Greenhouse Gases From Reservoir In Beijing

The greenhouse gas emissions from reservoirs and pools have been received great attention by researchers.In this study,the annual methane and carbon dioxide fluxes,and environmental parameters were measured from three reservoirs in Beijing.Differences of effect of environmental parameters on methane and carbon dioxide fluxes between two scales were studied,based on regional scale(Miyun,Yudushan and Baihepu,data source S1)and sampling scale(Miyun Reservoir,data source S2),in order to further explore the effect of environmental parameters on the greenhouse gas emissions from reservoirs at different scales.Moreover,statistics and machine learning method were used to establish greenhouse gas prediction models.Based on increase of model transfer learning and decrease of overfitting,method for adaptation of models were studied.CH4MODwetland were used to simulate methane fluxes in littoral zones.Furthermore,the effect of different estimation methods on annual greenhouse gas estimation were compared.The main results and conclusions are as follows:(1)The relationship between greenhouse gases and environmental parameters were different due to the greenhouse gas type,spatial distribution and research scales.Environmental parameters had similar effects on methane and carbon dioxide fluxes.Temperature(T),wind speed(WS),biomass(Bio),and soil organic carbon(SOC)could promote the emissions of methane and carbon dioxide.Moreover,air pressure(AP),pH,and dissolved oxygen(DO)can both inhibit emissions.Compared to the regional scale(S1),temperature(T),pressure(AP),wind speed(WS),dissolved oxygen(DO)could more likely had significant effects on methane and dioxide emissions in the sample scale(S2).(2)It was not necessary to select model parameters with correlation and regression analysis.If possible,it is better to use all available environmental parameters as candidates for development of prediction models.Otherwise,correlation analysis and regression analysis were used together for environmental parameters selection could reduce the risk of decrease of model accuracy.(3)The decision tree(DT)model showed the best adaptation and performance,compared to other five models.When the data source changes from S2 to S1,the adaptation of model(r2 as an indicator)did not change much in the Miyun dataset.However,r2 in Yudushan and Baihepu dataset were improved.In addition,the performance of methane model(DT)for the open water and littoral zones,were improved,except of carbon dioxide model(DT)from open water zone and littoral zones(take R2 as an example).(4)According to the observed value methods,the methane emission estimated by the IPCC default emission factor method was the highest that the Miyun Reservoir is a sink of carbon dioxide.This could overestimate annual methane emission from reservoirs in China,and misjudge the annual carbon sink/source attribute of reservoir based on carbon dioxide emissions.Compared to DT,the annual simulated value of methane emission with CH4MODwetland had less difference form observed value methods.However,it required more input parameters using CH4MODwetland.In this study,statistics model adaptation method for assessment of model transfer learning and decrease of overfitting were development.According to IPCC,methods in Tier2 and Tier3 for establishment and optimization of greenhouse gas emission prediction model of reservoirs in China were studied.The uncertainty caused by different methods for estimation of methane and carbon dioxide emissions were evaluated.It provides a reference for improvement of method for greenhouse gases estimation,and the estimation of greenhouse gas carbon emissions from reservoirs,and reservoir greenhouse gas inventories.