Research On Estimation Method Of Carbon Source Sink Based On Big Data

Because the traditional carbon cycle process model has strong mechanism and complex process,the key parameters depend on experience setting,which brings great uncertainty to regional and global carbon cycle simulation.In order to improve the estimation accuracy of total vegetation primary productivity(GPP),this paper uses machine learning algorithm to establish a data-driven GPP estimation model.Using the remote sensing data such as EVI,NDVI,precipitation and temperature under the GEE platform,and the flux tower of the site to measure GPP data,the random forest regression model is used to establish the connection between the remote sensing data and the measured data,and the model is used to predict the test group data..The results show that the parameters such as the fitting coefficient of the random forest model are better than the Moris data based on the ecological process model,and more accurate GPP prediction results can be obtained,which provides a new method for GPP estimation.The research work of this paper mainly includes:(1)The current GPP estimation methods at home and abroad are summarized and analyzed,the related concepts are explained,and the development trends of remote sensing and machine learning are introduced.(2)Eight highly representative sites in the country were selected as research areas.Obtain the flux tower data of the site and collect the remote sensing data of the study area using the GEE platform for preprocessing.Familiarize with the use of the GEE platform and make full use of the functions of the GEE platform to enable high-quality data support for subsequent experiments.(3)A random forest model was built using data from eight sites and the site GPP was estimated using the model.Analyze the importance of input influence factors,and improve the accuracy of random forest model prediction by means of adjustment and other means.Finally,the random forest model is compared with other machine learning models to illustrate the advantages of the random forest regression model.(4)Finally,R2,RMSE and other related precision indicators are used to evaluate the model prediction effect and compare it with the globally recognized MODIS data products.Verify the reliability of the prediction data of random forest models by comparison。After a large number of experimental analysis and verification,the method based on the random forest model estimation method adopted in this paper has better accuracy in GPP estimation.