Comparative Study Of GPP Estimation Based On Multi-dimensional Data Datasets(MDD) Fusion Data Set

The study of Gross Primary Productivity(GPP)is an important topic in the field of global change research such as land surface ecological balance and globally carbon cycle.At present,a lot of research has been done on the prediction and evaluation of GPP based on remote sensing methods.However,there are many issues to estimate GPP over a long time series can be concluded as follows:First of all,in order to ensure a certain signal-to-noise ratio of remote sensing images,there is a mutual restriction of sensor's spatial resolution and temporal resolution;Secondly,when dealing with massive long time series remote sensing data,the present process is complicated,low efficiency,and high error rate;Finally,accurate estimation of GPP still has a problem.This paper in the site scale to estimate the GPP of two different kinds of radiant flux tower sites(Mixed forest site US-PFa and Deciduous broadleaf forests site US-WCr)for a period of one year with an interval of 8 days.The main research contents and conclusions of this work are as follows:(1)In this paper,we first put a study on the storage management and application of massive long time series remote sensing data.We constructed 6 types of vegetation index(CIgreen,EVI,NDVI,SAVI,UNVI,WDRVI)long time series cubes based on Multi-dimensional Data Datasets(MDD)which include the time dimension.The former work provides available data and research basis for GPP estimation in the following part and also provides a new methods for long time series analysis research.(2)Four common spatial-spectral fusion methods for GPP estimation were performed on real data and simulation data,and the characteristics of various fusion methods were compared.The conclusion as follows:All of them could meet most of the fusion tasks;the result of CNMF method has the best visual effect and remains a complete spatial detail information;GS method performs best in the infrared band;Wavelet transform fusion has the highest spectral fidelity and the best spectral information;CRISP fusion method has a good comprehensive performance.This work not only provides a theoretical basis and guarantee for the subsequent generation of reliable spatio-temporal data by the spatial-spectral fusion method in this paper,but also provides a reference for the selection of appropriate fusion method when the fusion tasks are required.(3)This paper proposes a fusion idea and framework,which based on MDD data format to generate spatio-temporal fusion data by using the spatial-spectral fusion method.We utilized this method to generate 6 kinds of long time series VIs data and used these data to estimate GPP based on GPP-VI estimation model,and compared with the result based on traditional spatio-temporal fusion method ESTARFM.The new method has greatly improves the estimation accuracy that the coefficient of determination(R~2)is all greater than 0.8 and the time cost of this method is also very low.