Water-soluble humic-like substances(HULISWS)are a group of organic compounds which is harmful to human health.Oxygen potential(OP)could evaluate the oxygen ability of aerosols in the lungs and quantifying its source could help the precise emission reduction works.Here,the daily aerosol samples were collected in Northeast China Plain during autumn and winter.In this work,the sources of HULISWS concentration were firstly quantified using the positive-definite matrix factorization model(PMF).Although lots of work using PMF quantified the sources of the OP of HULISWS,the non-linear relation between concentration and OP could cause lots of uncertainties.Random forest(RF)algorithm,which is an easy tool to fit the complex non-linear relationship,was used to quantify the potential sources of OP.However,the generalization error will be much higher when the sample size is small.Here,we conducted a few-shot learning(FSL)method which improved the learning ability of the RF model by strengthening the recognition of characteristic variables[FSL-RF].Combining FSL-RF with PMF,the contribution of sources to the oxygen potential(OP)of HULISWS was quantified.The results indicated that biomass burning emission contributed 72％of mass concentration and 63％to OP of HULISWS.Besides,cooking emissions,which contributed 4％of the mass concentration of HULISWS,contributed 19％to OP of HULISWS.Our results showed that although biomass burning emissions domaint the OP of HULISWS,reducing the cooking emission might be the crucial way to reduce the OP of HULISWS in the Northeast China Plain.
维普
万方数据