Abstract: The results of fitting and reconstruction of long-term series vegetation index data can provide more accurate information and more reliable data source for vegetation dynamic monitoring, biomass information extraction, crop yield prediction and area estimation, vegetation phenological information extraction, ecological quality assessment and ecosystem carbon cycle research, which can effectively reflect the quality of ecological environment. After extensive research and verification, it is found that different fitting reconstruction methods have different suitability for different geographical environments. A large number of studies on comparative analysis methods mainly focus on qualitative analysis based on sample curve analysis and visual comparison, and quantitative analysis based on root mean square error, correlation coefficient, Akaike information criterion and Bayesian information standard. However, the evaluation indexes that quantitative analyses adopt are mostly the mean value, the maximum value, and the minimum value, which ignore the influence of abnormal values and spatial pattern differences on reconstruction result. In this study, the unchanging areas of cultivated land, forests, grasslands and shrubs in the Beijing-Tianjin-Hebei region were extracted through the spatial analysis tool in ArcGIS. Then, weights to all pixels were assigned in combination with the quality reliability of VI pixel. Next, the fitting reconstruction of the time series data of MDOSI EVI 16 d in Beijing-Tianjin-Hebei region from 2001 to 2015 were finished by asymmetric Gaussian function fitting method (AG), double logistic function fitting method (DL), and SG filtering method (SG). Before analyzing the fitting results, the fitted and original vegetation growth curves of Beijing station in 2006 were firstly extracted, then the start and end time of growing season were extracted by the dynamic threshold method. Verification was made combined with China National Specimen Information Infrastructure data and typical plant phenological observation dataset of Chinese phenological observation network. The results illustrated that the fitted vegetation growth curve by the three methods was consistent with the field observation data. The fitting result of the sampling point curve in the past 15 years was analyzed based on the analysis result of the relationship between noise ratio and fitting method. Combined with correlation coefficient, root mean square error, Akaike information criterion, Bayesian information standard, the spatial pattern of fitting result was analyzed. Finally, the method of mathematical statistics was used to quantitatively analyze the fitting result. The results showed that there was no significant difference between AG fitting and DL fitting in the denoising result. AG fitting showed better fitting reconstruction result at some pixel points, while SG filtering can preserve the original vegetation features more effectively. The reconstruction results of the three methods showed the difference related to the spatial distribution of land types. For the long-term time series data of Beijing-Tianjin-Hebei region, AG fitting showed better reconstruction result in grassland, forest and shrub areas with less human disturbance, and the result of SG filtering was better in the reconstruction of cultivated areas with stronger human activities. This study can provide reference for the fitting of time series data of vegetation in Beijing-Tianjin-Hebei region, and provide more objective and clear method support for the evaluation of the result of fitting reconstruction of time series data.