Research On The Generation Method Of NDVI Data Set With High Spatial And Temporal Resolusion

High spatial-temporal resolution NDVI data set can better reflect the change of vegetation phenology,which is of great significance for monitoring vegetation change information in time domain.In this thesis,a new spatiotemporal fusion model is proposed to construct a high spatiotemporal resolution NDVI data set,and filter and reconstruct the generated data set.In this thesis,the junction of Shandong,Henan and Anhui provinces is selected as the research area.Firstly,aiming at the problem of "space-time conflict" between Landsat 8 and MODIS satellite data,an improved spatiotemporal NDVI data fusion model is proposed(sptiotemporal NDVI data fusion model,STNDFM);Secondly,the current mainstream spatiotemporal fusion models STARFM,ESTARFM,FSDAF and the STNDFM method proposed in this thesis are fully compared and analyzed;Finally,the spatiotemporal fusion model proposed in this thesis is used to generate the high spatiotemporal resolution NDVI data set of 138 remote sensing data in 2014-2016 with a spatial resolution of 8 days and 30 meters,and A-G Method,D-L method and S-G filtering method are used to reconstruct the generated data set,and then the results of the three methods are compared and analyzed.The specific research content and results of this thesis mainly include the following three aspects:1.By combining the method based on weight function and the method based on mixed pixel decomposition,and applying the spatiotemporal fusion model directly to the calculation of vegetation index NDVI,an improved spatiotemporal fusion model STNDFM is proposed.This method not only considers the conversion coefficient and weight function between similar pixels by using the method based on weight function,but also uses the method of mixed pixel decomposition for low Instead of resampling data directly,resolution pixels are used for spectral unmixing to obtain NDVI values of specific surface types as the basis for further fusion.2.This thesis discusses the image prediction quality and applicability of four spatiotemporal fusion models in different wavebands and different terrain types from qualitative and quantitative perspectives.The comprehensive performance of the four spatiotemporal fusion models is as follows:STDNFM>ESTARFM>FSDAF>STARFM,which shows that the spatiotemporal fusion model proposed in this thesis has higher prediction accuracy,and it also proves that two pairs of data can be input at the reference time It can provide more spatial detail information for the prediction time,so as to improve the prediction accuracy.Then,the parameter sensitivity analysis of the spatiotemporal fusion model proposed in this thesis is carried out.The results show that: when selecting the data pairs at the reference time,we should try to select the data with smaller prediction date time interval and in the same vegetation development period,and the size of the moving window and the number of selected classification have an important impact on the accuracy of the fusion data.3.From the perspective of the reconstruction effect of NDVI data set,three kinds of time series data reconstruction methods can remove the noise in the original NDVI time series data,but the reconstructed time series curve has different degrees of improvement compared with the original time series curve,among which A-G method is the best,D-L method is the second,and S-G method is the worst.So using A-G method to reconstruct NDVI time series data is the best,and it can better describe the change characteristics and growth trend of vegetation in the original NDVI time series curve.