Analysis And System Development Of Vegetation Phenology Parameters Extraction Based On Near-Surface Multispectral Images

As an important parameter reflecting the growth status of vegetation and the relationship between vegetation and climate change,vegetation phenology has been a hotspot in ecosystem studies.Based on near-ground multispectral digital camera photo time series data,automatic and visual analysis of vegetation phenology on landscape scale and smaller scale is an effective means to make up for the lack of large-scale remote sensing and manual observation.In this paper,a vegetation phenology parameter extraction and analysis system based on near-surface multispectral images is constructed based on R-Shiny,including site distribution visualization,ROI(Region of interest)selection and mapping,vegetation index calculation,data filtering,growth curve trajectory fitting and phenological parameters extraction,and pixel-scale phenology extraction analysis modules.The Web system relies on the Shiny software package in the R language,implements the interface design and parameter input at the UI(User Interface)end,and receives the input parameters for the algorithm operation at the Server end,and finally realizes the output and storage of the result.The system has friendly interface and good real-time interaction verified by phenology camera observation data from Phenocam.And it effectively improves the research efficiency of vegetation phenology based on near-ground multispectral cameras.In addition,based on the application results of the system,the effect and applicability of methods of vegetation index sequence data filtering,growth curve fitting and key phenological parameters extraction are explored in the ROI scale,and then the uncertainty estimation is carried out.And the pixel by pixel phenology parameters are extracted in the ROI.The main conclusions are as follows:(1)Relative greenness index(GI)is significantly higher than other relative vegetation indices and single-band brightness values,and its time series can be used to characterize vegetation growth trajectory after filtering the noise effects caused by instruments,lighting and weather using three-day sliding window algorithm(Max method);The NDVI(Normalized Difference Vegetation Index)time series data obtained by NIR(Near Infrared Radiation)and R bands is decomposed by the EMD(Empirical Mode Decomposition)method,and its time distribution characteristics reflected by the trend item(residue)are basically the same as those reflected by GI,which can be a reference for the phenological results from GI index.(2)Combination of different fitting methods and phenological parameter extraction methods apply to different vegetation types.The combination of KLOSTERMAN fitting method and TRS,Klosterman,GU extraction method are suitable for woodland and forest;The best combination methods for one-season crops(wheat,corn)are GU-fitting method and TRS,Derivatives,Klosterman,and GU extraction methods;And it is best to use the SPLINE fitting method to fit the growth trajectory for the multi-growing peak vegetation,and then use the change point method to extract the key phenological parameters.Uncertainty estimation was made on the combination of the above-mentioned preferred growth curve fitting and parameter extraction methods,and it was found that extraction results based on the above method combination have no significant outliers and have strong robustness.In addition,the phenology parameters result of MODIS EVI extraction is used to verify the phenological parameters of camera data,and the results are consistent.(3)Extracting key phenological parameters from each pixel by within ROI can effectively identify phenological differences between species and individuals under a uniform community scale.In the future,it can be used as an automated method for analyzing community-scale biodiversity.